Merge pull request #641 from meilisearch/bump-version

Bump version to v0.10.1

.dockerignore

@@ -0,0 +1,5 @@
+target
+Dockerfile
+.dockerignore
+.git
+.gitignore

.github/workflows/README.md

@@ -0,0 +1,18 @@
+# GitHub actions workflow for MeiliDB
+
+> **Note:**
+
+> - We do not use [cache](https://github.com/actions/cache) yet but we could use it to speed up CI
+
+## Workflow
+
+- On each pull request, we are triggering `cargo test`.
+- On each tag, we are building:
+    - the tagged docker image
+    - the binaries for MacOS, Ubuntu, and Windows
+    - the debian package
+- On each stable release, we are build the latest docker image.
+
+## Problems
+
+- We do not test on Windows because we are unable to make it work, there is a disk space problem.

.github/workflows/check-updated-changelog.yml

@@ -0,0 +1,16 @@
+name: Check if the CHANGELOG.md has been updated
+
+on: [pull_request]
+
+jobs:
+  check:
+    name: Test on ${{ matrix.os }}
+    if: ${{ !contains(github.event.pull_request.labels.*.name, 'ignore-changelog') }}
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - name: Checking the CHANGELOG.md has been updated in this PR
+      run: |
+        set -e
+        git fetch origin ${{ github.base_ref }}
+        git diff --name-only origin/${{ github.base_ref }} | grep -q CHANGELOG.md

.github/workflows/publish-binaries.yml

@@ -0,0 +1,87 @@
+name: Publish binaries to GitHub release
+
+on:
+  push:
+    tags:
+      - '*'
+
+jobs:
+  publish:
+    name: Publish for ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [ubuntu-latest, macos-latest, windows-latest]
+        include:
+          - os: ubuntu-latest
+            artifact_name: meilisearch
+            asset_name: meilisearch-linux-amd64
+          - os: macos-latest
+            artifact_name: meilisearch
+            asset_name: meilisearch-macos-amd64
+          - os: windows-latest
+            artifact_name: meilisearch
+            asset_name: meilisearch-windows-amd64
+
+    steps:
+    - uses: hecrj/setup-rust-action@master
+      with:
+        rust-version: stable
+    - uses: actions/checkout@v1
+    - name: Build
+      run: cargo build --release --locked
+    - name: Upload binaries to release
+      uses: svenstaro/upload-release-action@v1-release
+      with:
+        repo_token: ${{ secrets.PUBLISH_TOKEN }}
+        file: target/release/${{ matrix.artifact_name }}
+        asset_name: ${{ matrix.asset_name }}
+        tag: ${{ github.ref }}
+
+  publish-armv7:
+    name: Publish for ARMv7
+    runs-on: ubuntu-18.04
+    steps:
+      - uses: actions/checkout@v1.0.0
+      - uses: uraimo/run-on-arch-action@v1.0.7
+        id: runcmd
+        with:
+          architecture: armv7
+          distribution: ubuntu18.04
+          run: |
+            apt update
+            apt install -y curl gcc make
+            curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --default-toolchain stable
+            source $HOME/.cargo/env
+            cargo build --release --locked
+      - name: Upload the binary to release
+        uses: svenstaro/upload-release-action@v1-release
+        with:
+          repo_token: ${{ secrets.PUBLISH_TOKEN }}
+          file: target/release/meilisearch
+          asset_name: meilisearch-linux-armv7
+          tag: ${{ github.ref }}
+
+  publish-armv8:
+    name: Publish for ARMv8
+    runs-on: ubuntu-18.04
+    steps:
+      - uses: actions/checkout@v1.0.0
+      - uses: uraimo/run-on-arch-action@v1.0.7
+        id: runcmd
+        with:
+          architecture: aarch64 # aka ARMv8
+          distribution: ubuntu18.04
+          run: |
+            apt update
+            apt install -y curl gcc make
+            curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --default-toolchain stable
+            source $HOME/.cargo/env
+            cargo build --release --locked
+      - name: Upload the binary to release
+        uses: svenstaro/upload-release-action@v1-release
+        with:
+          repo_token: ${{ secrets.PUBLISH_TOKEN }}
+          file: target/release/meilisearch
+          asset_name: meilisearch-linux-armv8
+          tag: ${{ github.ref }}

.github/workflows/publish-deb-brew-pkg.yml

@@ -0,0 +1,39 @@
+name: Publish deb pkg to GitHub release & apt repository & Homebrew
+
+on:
+  push:
+    tags:
+      - 'v[0-9]+.[0-9]+.[0-9]+'
+
+jobs:
+  debian:
+    name: Publish debian packagge
+    runs-on: ubuntu-latest
+    steps:
+    - uses: hecrj/setup-rust-action@master
+      with:
+        rust-version: stable
+    - name: Install cargo-deb
+      run: cargo install cargo-deb
+    - uses: actions/checkout@v1
+    - name: Build deb package
+      run: cargo deb -p meilisearch-http -o target/debian/meilisearch.deb
+    - name: Upload debian pkg to release
+      uses: svenstaro/upload-release-action@v1-release
+      with:
+        repo_token: ${{ secrets.GITHUB_TOKEN }}
+        file: target/debian/meilisearch.deb
+        asset_name: meilisearch.deb
+        tag: ${{ github.ref }}
+    - name: Upload debian pkg to apt repository
+      run: curl -F package=@target/debian/meilisearch.deb https://${{ secrets.GEMFURY_PUSH_TOKEN }}@push.fury.io/meilisearch/
+
+  homebrew:
+    name: Bump Homebrew formula
+    runs-on: ubuntu-latest
+    steps:
+      - uses: mislav/bump-homebrew-formula-action@v1
+        with:
+          formula-name: meilisearch
+        env:
+          COMMITTER_TOKEN: ${{ secrets.HOMEBREW_COMMITTER_TOKEN }}

.github/workflows/publish-docker-latest.yml

@@ -0,0 +1,19 @@
+---
+on:
+  push:
+    tags:
+      - 'v[0-9]+.[0-9]+.[0-9]+'
+
+name: Publish latest image to Docker Hub
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v1
+      - name: Publish to Registry
+        uses: elgohr/Publish-Docker-Github-Action@master
+        with:
+          name: getmeili/meilisearch
+          username: ${{ secrets.DOCKER_USERNAME }}
+          password: ${{ secrets.DOCKER_PASSWORD }}

.github/workflows/publish-docker-tag.yml

@@ -0,0 +1,20 @@
+---
+on:
+  push:
+    tags:
+      - '*'
+
+name: Publish tagged image to Docker Hub
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v1
+      - name: Publish to Registry
+        uses: elgohr/Publish-Docker-Github-Action@master
+        with:
+          name: getmeili/meilisearch
+          username: ${{ secrets.DOCKER_USERNAME }}
+          password: ${{ secrets.DOCKER_PASSWORD }}
+          tag_names: true

.github/workflows/test.yml

@@ -0,0 +1,25 @@
+---
+on: [pull_request]
+
+name: Test binaries with cargo test
+
+jobs:
+  check:
+    name: Test on ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+
+    steps:
+    - uses: actions/checkout@v1
+    - uses: actions-rs/toolchain@v1
+      with:
+        profile: minimal
+        toolchain: stable
+        override: true
+    - name: Run cargo test
+      uses: actions-rs/cargo@v1
+      with:
+        command: test
+        args: --locked --release

.gitignore

@@ -1,7 +1,8 @@
-/rocksdb
 /target
-/Cargo.lock
-**/*.rs.bk
+meilisearch-core/target
 **/*.csv
 **/*.json_lines
-**/*.rdb
+**/*.rs.bk
+/*.mdb
+/query-history.txt
+/data.ms

.travis.yml

@@ -1,22 +0,0 @@
-language: rust
-
-cache: cargo
-
-branches:
-  only:
-    - master
-
-matrix:
-  fast_finish: true
-  include:
-
-  # Test crates on their minimum Rust versions.
-  - rust: 1.31.0
-    name: "meilidb on 1.31.0"
-    script: ./ci/meilidb.sh
-
-  # Test crates on nightly Rust.
-  - rust: nightly
-    name: "meilidb on nightly"
-    script: ./ci/meilidb.sh
-

CHANGELOG.md

@@ -0,0 +1,19 @@
+## v0.10.1
+
+  - Add support for floating points in filters (#640)
+  - Add '@' character as tokenizer separator (#607)
+  - Add support for filtering on arrays of strings (#611)
+
+## v0.10
+
+  - Refined filtering (#592)
+  - Add the number of hits in search result (#541)
+  - Add support for aligned crop in search result (#543)
+  - Sanitize the content displayed in the web interface (#539)
+  - Add support of nested null, boolean and seq values (#571 and #568, #574)
+  - Fixed the core benchmark (#576)
+  - Publish an ARMv7 and ARMv8 binaries on releases (#540 and #581)
+  - Fixed a bug where the result of the update status after the first update was empty (#542)
+  - Fixed a bug where stop words were not handled correctly (#594)
+  - Fix CORS issues (#602)
+  - Support wildcard on attributes to retrieve, highlight, and crop (#549, #565, and #598)

Cargo.toml

@@ -1,55 +1,11 @@
-[package]
-edition = "2018"
-name = "meilidb"
-version = "0.2.0"
-authors = ["Kerollmops <renault.cle@gmail.com>"]
-
-[dependencies]
-bincode = "1.0"
-byteorder = "1.2"
-crossbeam = "0.6"
-fst = "0.3"
-hashbrown = { version = "0.1", features = ["serde"] }
-lazy_static = "1.1"
-levenshtein_automata = { version = "0.1", features = ["fst_automaton"] }
-linked-hash-map = { version = "0.5", features = ["serde_impl"] }
-log = "0.4"
-sdset = "0.3"
-serde = "1.0"
-serde_derive = "1.0"
-unidecode = "0.3"
-
-[dependencies.toml]
-git = "https://github.com/Kerollmops/toml-rs.git"
-features = ["preserve_order"]
-rev = "0372ba6"
-
-[dependencies.rocksdb]
-git = "https://github.com/pingcap/rust-rocksdb.git"
-rev = "306e201"
-
-[dependencies.group-by]
-git = "https://github.com/Kerollmops/group-by.git"
-rev = "5a113fe"
-
-[features]
-default = ["simd"]
-i128 = ["bincode/i128", "byteorder/i128"]
-portable = ["rocksdb/portable"]
-simd = ["rocksdb/sse"]
-nightly = ["hashbrown/nightly", "group-by/nightly"]
-
-[dev-dependencies]
-csv = "1.0"
-elapsed = "0.1"
-env_logger = "0.6"
-jemallocator = "0.1"
-quickcheck = "0.8"
-rand = "0.6"
-rand_xorshift = "0.1"
-structopt = "0.2"
-tempfile = "3.0"
-termcolor = "1.0"
+[workspace]
+members = [
+    "meilisearch-core",
+    "meilisearch-http",
+    "meilisearch-schema",
+    "meilisearch-tokenizer",
+    "meilisearch-types",
+]
 
 [profile.release]
 debug = true

Dockerfile

@@ -0,0 +1,27 @@
+# Compile
+FROM    alpine:3.10 AS compiler
+
+RUN     apk update --quiet
+RUN     apk add curl
+RUN     apk add build-base
+
+RUN     curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
+
+WORKDIR /meilisearch
+
+COPY    . .
+
+ENV     RUSTFLAGS="-C target-feature=-crt-static"
+
+RUN     $HOME/.cargo/bin/cargo build --release
+
+# Run
+FROM    alpine:3.10
+
+RUN     apk update --quiet
+RUN     apk add libgcc
+
+COPY    --from=compiler /meilisearch/target/release/meilisearch .
+
+ENV     MEILI_HTTP_ADDR 0.0.0.0:7700
+CMD     ./meilisearch

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2018 Clément Renault
+Copyright (c) 2019-2020 Meili SAS
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -1,60 +1,188 @@
-# MeiliDB
-
-[![Build Status](https://travis-ci.org/Kerollmops/MeiliDB.svg?branch=master)](https://travis-ci.org/Kerollmops/MeiliDB)
-[![dependency status](https://deps.rs/repo/github/Kerollmops/MeiliDB/status.svg)](https://deps.rs/repo/github/Kerollmops/MeiliDB)
-[![License](https://img.shields.io/github/license/Kerollmops/MeiliDB.svg)](https://github.com/Kerollmops/MeiliDB)
-[![Rust 1.31+](https://img.shields.io/badge/rust-1.31+-lightgray.svg)](
-https://www.rust-lang.org)
-
-A _full-text search database_ using a key-value store internally.
-
-It uses [RocksDB](https://github.com/facebook/rocksdb) as the internal key-value store. The key-value store allows us to handle updates and queries with small memory and CPU overheads.
-
-You can [read the deep dive](deep-dive.md) if you want more information on the engine, it describes the whole process of generating updates and handling queries.
-
-We will be proud if you submit issues and pull requests. You can help to grow this project and start contributing by checking [issues tagged "good-first-issue"](https://github.com/Kerollmops/MeiliDB/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22). It is a good start!
-
-The project is only a library yet. It means that there is no binary provided yet. To get started, you can check the examples wich are made to work with the data located in the `misc/` folder.
-
-MeiliDB will be a binary in a near future so you will be able to use it as a database out-of-the-box. We should be able to query it using a [to-be-defined](https://github.com/Kerollmops/MeiliDB/issues/38) protocol. This is our current goal, [see the milestones](https://github.com/Kerollmops/MeiliDB/milestones). In the end, the binary will be a bunch of network protocols and wrappers around the library - which will also be published on [crates.io](https://crates.io). Both the binary and the library will follow the same update cycle.
+<p align="center">
+  <img src="assets/logo.svg" alt="MeiliSearch" width="200" height="200" />
+</p>
 
 
+<h1 align="center">MeiliSearch</h1>
 
-## Performances
+<h4 align="center">
+  <a href="https://www.meilisearch.com">Website</a> | 
+  <a href="https://blog.meilisearch.com">Blog</a> | 
+  <a href="https://fr.linkedin.com/company/meilisearch">LinkedIn</a> | 
+  <a href="https://twitter.com/meilisearch">Twitter</a> | 
+  <a href="https://docs.meilisearch.com">Documentation</a> | 
+  <a href="https://docs.meilisearch.com/resources/faq.html">FAQ</a>
+</h4>
 
-With a database composed of _100 353_ documents with _352_ attributes each and _90_ of them indexed.
-So nearly _9 million_ fields indexed for _35 million_ stored we can handle more than _1.2k req/sec_ on an Intel i7-7700 (8) @ 4.2GHz.
+<p align="center">
+  <a href="https://github.com/meilisearch/MeiliSearch/actions"><img src="https://github.com/meilisearch/MeiliSearch/workflows/Cargo%20test/badge.svg" alt="Build Status"></a>
+  <a href="https://deps.rs/repo/github/meilisearch/MeiliSearch"><img src="https://deps.rs/repo/github/meilisearch/MeiliSearch/status.svg" alt="Dependency status"></a>
+  <a href="https://github.com/meilisearch/MeiliSearch/blob/master/LICENSE"><img src="https://img.shields.io/badge/license-MIT-informational" alt="License"></a>
+  <a href="https://slack.meilisearch.com"><img src="https://img.shields.io/badge/slack-MeiliSearch-blue.svg?logo=slack" alt="Slack"></a>
+  <a href="https://github.com/meilisearch/MeiliSearch/discussions" alt="Discussions"><img src="https://img.shields.io/badge/github-discussions-red" /></a>
+</p>
 
-Requests are made using [wrk](https://github.com/wg/wrk) and scripted to generate real users queries.
+<p align="center">⚡ Lightning Fast, Ultra Relevant, and Typo-Tolerant Search Engine 🔍</p>
 
-```
-Running 10s test @ http://localhost:2230
-  2 threads and 12 connections
-  Thread Stats   Avg      Stdev     Max   +/- Stdev
-    Latency    18.86ms   49.39ms 614.89ms   95.23%
-    Req/Sec   620.41     59.53   790.00     65.00%
-  12359 requests in 10.00s, 3.26MB read
-Requests/sec:   1235.54
-Transfer/sec:    334.22KB
-```
+**MeiliSearch** is a powerful, fast, open-source, easy to use and deploy search engine. Both searching and indexing are highly customizable. Features such as typo-tolerance, filters, and synonyms are provided out-of-the-box.
+For more information about features go to [our documentation](https://docs.meilisearch.com/).
 
-### Notes
+<p align="center">
+  <a href="https://crates.meilisearch.com"><img src="assets/crates-io-demo.gif" alt="crates.io demo gif" /></a>
+</p>
 
-The default Rust allocator has recently been [changed to use the system allocator](https://github.com/rust-lang/rust/pull/51241/).
-We have seen much better performances when [using jemalloc as the global allocator](https://github.com/alexcrichton/jemallocator#documentation).
+> MeiliSearch helps the Rust community find crates on [crates.meilisearch.com](https://crates.meilisearch.com)
 
-## Usage and examples
+## Features
+* Search as-you-type experience (answers < 50 milliseconds)
+* Full-text search
+* Typo tolerant (understands typos and miss-spelling)
+* Supports Kanji characters
+* Supports Synonym
+* Easy to install, deploy, and maintain
+* Whole documents are returned
+* Highly customizable
+* RESTful API
 
-MeiliDB runs with an index like most search engines.
-So to test the library you can create one by indexing a simple csv file.
+## Get started
+
+### Deploy the Server
+
+#### Run it using Docker
 
 ```bash
-cargo run --release --example create-database -- test.mdb misc/kaggle.csv --schema schema-example.toml --stop-words misc/fr.stopwords.txt
+docker run -p 7700:7700 -v $(pwd)/data.ms:/data.ms getmeili/meilisearch
 ```
 
-Once the command is executed, the index should be in the `test.mdb` folder. You are now able to run the `query-database` example and play with MeiliDB.
+#### Installing with Homebrew
 
 ```bash
-cargo run --release --example query-database -- test.mdb -n 10 id title
+brew update && brew install meilisearch
+meilisearch
 ```
 
+#### Installing with APT
+
+```bash
+echo "deb [trusted=yes] https://apt.fury.io/meilisearch/ /" > /etc/apt/sources.list.d/fury.list
+apt update && apt install meilisearch-http
+meilisearch
+```
+
+#### Download the binary
+
+```bash
+curl -L https://install.meilisearch.com | sh
+./meilisearch
+```
+
+#### Compile and run it from sources
+
+If you have the Rust toolchain already installed on your local system, clone the repository and change it to your working directory.
+
+```bash
+git clone https://github.com/meilisearch/MeiliSearch.git
+cd MeiliSearch
+```
+
+In the cloned repository, compile MeiliSearch.
+
+```bash
+cargo run --release
+```
+
+### Create an Index and Upload Some Documents
+
+Let's create an index! If you need a sample dataset, use [this movie database](https://www.notion.so/meilisearch/A-movies-dataset-to-test-Meili-1cbf7c9cfa4247249c40edfa22d7ca87#b5ae399b81834705ba5420ac70358a65). You can also find it in the `datasets/` directory.
+
+```bash
+curl -L 'https://bit.ly/2PAcw9l' -o movies.json
+```
+
+MeiliSearch can serve multiple indexes, with different kinds of documents.
+It is required to create an index before sending documents to it.
+
+```bash
+curl -i -X POST 'http://127.0.0.1:7700/indexes' --data '{ "name": "Movies", "uid": "movies" }'
+```
+
+Now that the server knows about your brand new index, you're ready to send it some data.
+
+```bash
+curl -i -X POST 'http://127.0.0.1:7700/indexes/movies/documents' \
+  --header 'content-type: application/json' \
+  --data-binary @movies.json
+```
+
+### Search for Documents
+
+#### In command line
+
+The search engine is now aware of your documents and can serve those via an HTTP server. 
+
+The [`jq` command-line tool](https://stedolan.github.io/jq/) can greatly help you read the server responses.
+
+```bash
+curl 'http://127.0.0.1:7700/indexes/movies/search?q=botman+robin&limit=2' | jq
+```
+
+```json
+{
+  "hits": [
+    {
+      "id": "415",
+      "title": "Batman & Robin",
+      "poster": "https://image.tmdb.org/t/p/w1280/79AYCcxw3kSKbhGpx1LiqaCAbwo.jpg",
+      "overview": "Along with crime-fighting partner Robin and new recruit Batgirl...",
+      "release_date": "1997-06-20",
+    },
+    {
+      "id": "411736",
+      "title": "Batman: Return of the Caped Crusaders",
+      "poster": "https://image.tmdb.org/t/p/w1280/GW3IyMW5Xgl0cgCN8wu96IlNpD.jpg",
+      "overview": "Adam West and Burt Ward returns to their iconic roles of Batman and Robin...",
+      "release_date": "2016-10-08",
+    }
+  ],
+  "offset": 0,
+  "limit": 2,
+  "processingTimeMs": 1,
+  "query": "botman robin"
+}
+```
+
+#### Use the Web Interface
+
+We also deliver an **out-of-the-box web interface** in which you can test MeiliSearch interactively.
+
+You can access the web interface in your web browser at the root of the server. The default URL is [http://127.0.0.1:7700](http://127.0.0.1:7700). All you need to do is open your web browser and enter MeiliSearch’s address to visit it. This will lead you to a web page with a search bar that will allow you to search in the selected index.
+
+<p align="center">
+  <img src="assets/movies-web-demo.gif" alt="Web interface gif" />
+</p>
+
+### Documentation
+
+Now that your MeiliSearch server is up and running, you can learn more about how to tune your search engine in [the documentation](https://docs.meilisearch.com). 
+
+
+## Contributing
+
+Hey! We're glad you're thinking about contributing to MeiliSearch! If you think something is missing or could be improved, please open issues and pull requests. If you'd like to help this project grow, we'd love to have you! To start contributing, checking [issues tagged as "good-first-issue"](https://github.com/meilisearch/MeiliSearch/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) is a good start!
+
+### Analytic Events
+
+Once a day, events are being sent to our Amplitude instance so we can know how many people are using MeiliSearch.<br/>
+Only information about the platform on which the server runs is stored. No other information is being sent.<br/>
+If this doesn't suit you, you can disable these analytics by using the `MEILI_NO_ANALYTICS` env variable.
+
+## Contact
+
+Feel free to contact us about any questions you may have:
+* At [bonjour@meilisearch.com](mailto:bonjour@meilisearch.com): English or French is welcome! 🇬🇧 🇫🇷
+* Via the chat box available on every page of [our documentation](https://docs.meilisearch.com/) and on [our landing page](https://www.meilisearch.com/).
+* 🆕 Join our [GitHub Discussions forum](https://github.com/meilisearch/MeiliSearch/discussions) (BETA hype!)
+* Join our [Slack community](https://slack.meilisearch.com/).
+* By opening an issue.
+
+Any suggestion or feedback is highly appreciated. Thank you for your support!

assets/logo.svg

@@ -0,0 +1,17 @@
+<svg width="360" height="360" viewBox="0 0 360 360" fill="none" xmlns="http://www.w3.org/2000/svg">
+<g id="logo_main">
+<rect id="Rectangle" x="107.333" y="0.150146" width="274.315" height="274.315" rx="98.8334" transform="rotate(23 107.333 0.150146)" fill="url(#paint0_linear)"/>
+<path id="Rectangle_2" fill-rule="evenodd" clip-rule="evenodd" d="M61.3296 230.199C46.2224 194.608 38.6688 176.813 38.208 160.329C37.5286 136.025 47.0175 112.539 64.3891 95.5282C76.1718 83.9904 93.9669 76.4368 129.557 61.3296C165.147 46.2224 182.943 38.6688 199.427 38.208C223.731 37.5286 247.217 47.0175 264.228 64.3891C275.766 76.1718 283.319 93.9669 298.426 129.557C313.534 165.147 321.087 182.943 321.548 199.427C322.227 223.731 312.738 247.217 295.367 264.228C283.584 275.766 265.789 283.319 230.199 298.426C194.608 313.534 176.813 321.087 160.329 321.548C136.025 322.227 112.539 312.738 95.5282 295.367C83.9903 283.584 76.4368 265.789 61.3296 230.199Z" fill="url(#paint1_linear)"/>
+<path id="m" fill-rule="evenodd" clip-rule="evenodd" d="M219.568 130.748C242.363 130.748 259.263 147.451 259.263 174.569V229.001H227.232V179.678C227.232 166.119 220.747 159.634 210.136 159.634C205.223 159.634 200.311 161.796 195.595 167.494C195.791 169.852 195.988 172.21 195.988 174.569V229.001H164.154V179.678C164.154 166.119 157.472 159.634 147.057 159.634C142.145 159.634 137.429 161.992 132.712 168.084V229.001H100.878V133.695H132.712V139.394C139.197 133.892 145.878 130.748 156.49 130.748C168.477 130.748 178.695 135.267 185.769 143.52C195.791 134.678 205.42 130.748 219.568 130.748Z" fill="white"/>
+</g>
+<defs>
+<linearGradient id="paint0_linear" x1="-13.6248" y1="129.208" x2="244.49" y2="403.522" gradientUnits="userSpaceOnUse">
+<stop stop-color="#E41359"/>
+<stop offset="1" stop-color="#F23C79"/>
+</linearGradient>
+<linearGradient id="paint1_linear" x1="11.0088" y1="111.65" x2="111.65" y2="348.747" gradientUnits="userSpaceOnUse">
+<stop stop-color="#24222F"/>
+<stop offset="1" stop-color="#2B2937"/>
+</linearGradient>
+</defs>
+</svg>

ci/meilidb.sh

@@ -1,15 +0,0 @@
-#!/bin/bash
-
-cd "$(dirname "$0")"/..
-set -ex
-
-export RUSTFLAGS="-D warnings"
-
-cargo check --no-default-features
-cargo check --bins --examples --tests
-cargo test
-
-if [[ "$TRAVIS_RUST_VERSION" == "nightly" ]]; then
-    cargo check --no-default-features --features nightly
-    cargo test --features nightly
-fi

datasets/movies/README.md

@@ -0,0 +1 @@
+_datas in movies.csv are from https://www.themoviedb.org/_

datasets/movies/settings.json

@@ -0,0 +1,10 @@
+{
+    "searchableAttributes": ["title", "overview"],
+    "displayedAttributes": [
+        "id",
+        "title",
+        "overview",
+        "release_date",
+        "poster"
+    ]
+}

deep-dive.md

@@ -1,140 +0,0 @@
-# A deep dive in MeiliDB
-
-On the 9 of december 2018.
-
-MeiliDB is a full text search engine based on a final state transducer named [fst](https://github.com/BurntSushi/fst) and a key-value store named [RocksDB](https://github.com/facebook/rocksdb). The goal of a search engine is to store data and to respond to queries as accurate and fast as possible. To achieve this it must save the data as an [inverted index](https://en.wikipedia.org/wiki/Inverted_index).
-
-
-
-<!-- MarkdownTOC autolink="true" -->
-
-- [Where is the data stored?](#where-is-the-data-stored)
-- [What does the key-value store contains?](#what-does-the-key-value-store-contains)
-    - [The blob type](#the-blob-type)
-        - [A final state transducer](#a-final-state-transducer)
-        - [Document indexes](#document-indexes)
-        - [Document ids](#document-ids)
-    - [The schema](#the-schema)
-    - [Document attributes](#document-attributes)
-- [How is an update handled?](#how-is-an-update-handled)
-    - [The merge operation is CPU consuming](#the-merge-operation-is-cpu-consuming)
-- [How is a request processed?](#how-is-a-request-processed)
-    - [Query lexemes](#query-lexemes)
-    - [Automatons and query index](#automatons-and-query-index)
-    - [Sort by criteria](#sort-by-criteria)
-    - [Retrieve original documents](#retrieve-original-documents)
-
-<!-- /MarkdownTOC -->
-
-## Where is the data stored?
-
-MeiliDB is entirely backed by a key-value store like any good database (i.e. Postgres, MySQL). This brings a great flexibility in the way documents can be stored and updates handled along time.
-
-[RocksDB brings some](https://rocksdb.org/blog/2015/02/27/write-batch-with-index.html) of the [A.C.I.D. properties](https://en.wikipedia.org/wiki/ACID_(computer_science)) to help us be sure the saved data is consistent, for example we use SST files and the key-value store ability to load them in one time to manage updates.
-
-Note that the SST file have the same restriction as the fst, it needs its keys to be added in order at creation.
-
-
-
-## What does the key-value store contains?
-
-It contain the blob, the schema and the documents stored attributes.
-
-### The blob type
-
-[The Blob type](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/blob/mod.rs#L16-L19) is a data structure that indicate if an update is a positive or a negative one. In the case where the update is considered positive, the blob will contain [an fst map and the document indexes](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/blob/positive/blob.rs#L15-L18) associated. In the other case it will only contain [all the document ids](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/blob/negative/blob.rs#L12-L14) that must be considered removed.
-
-The Blob type [is stored under the "*data-index*" entry](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/update/positive/update.rs#L497-L499) and marked as [a merge operation](https://github.com/facebook/rocksdb/wiki/Merge-Operator-Implementation) in the key-value store.
-
-#### A final state transducer
-
-_...also abbreviated fst_
-
-This is the first entry point of the engine, you can read more about how it work with the beautiful blog post of @BurntSushi, [Index 1,600,000,000 Keys with Automata and Rust](https://blog.burntsushi.net/transducers/).
-
-To make it short it is a powerful way to store all the words that are present in the indexed documents. You construct it by giving it all the words you want to index associated with a value that, for the moment, can only be an `u64`. When you want to search in it you can provide any automaton you want, in MeiliDB [a custom levenshtein automaton](https://github.com/tantivy-search/levenshtein-automata/) is used.
-
-Note that the number under each word is auto-incremental, each new word have a new number that is greater than the prevous one.
-
-Another powerful feature of `fst` is that it can nearly avoid using RAM and be streamed to disk for example, the problem is that the keys must be always added in lexicographic order, so you must sort them before, for the moment MeiliDB uses a [BTreeMap](https://github.com/Kerollmops/raptor-rs/blob/8abdb0a228e2808fe1814a6a0641a4b72d158579/src/metadata/doc_indexes.rs#L107-L112).
-
-#### Document indexes
-
-As it has been specified, the `fst` can only store a number corresponding to a word, an `u64`, but the goal of the search engine is to retrieve a match in a document when a query is made. You want it to return some sort of position in an attribute in a document, an information about where the given word match.
-
-To make it possible, a custom data structure has been developed, the document indexes is composed of two arrays, the ranges array and all the docindexes corresponding to a given range, each range identify the word number. The [DocIndexes](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/data/doc_indexes.rs#L23) type is designed to be streamed when constructed, consumming a minimum amount of ram like the fst. Another advantage is that the slices are accessible in `O(1)` when you know the word associated number.
-
-#### Document ids
-
-This is a simple ordered list of all documents ids which must be considered deleted. It is used with [the sdset library](https://docs.rs/sdset/0.3.0/sdset/duo/struct.DifferenceByKey.html), the docindexes and the `DifferenceByKey` operation builder when merging blobs.
-
-When a blob represent a negative update it only contains this simple slice of deleted documents ids.
-
-### The schema
-
-The schema is a data struture that represents which documents attributes should be stored and which should be indexed. It is stored under the "_data-schema_" entry and given to MeiliDB only at the creation.
-
-Each document attribute is associated to a unique 32 bit number named `SchemaAttr`.
-
-In the future this schema type could be given along with updates and probably be different from the original, the database could be able to handled this document structure and reindex it.
-
-### Document attributes
-
-When the engine handle a query the result that the requester want is a document, not only the [match](https://github.com/Kerollmops/MeiliDB/blob/fc2cdf92596fc002ce278e3aa8718640ac44724d/src/lib.rs#L51-L79) associated to it, fields of the original document must be returned too.
-
-So MeiliDB again uses the power of the underlying key-value store and save the documents attributes marked as _STORE_. The key is prefixed by "_doc_" followed by the 64 bit document id in bytes and the schema attribute number in bytes corresponding to the document attribute stored.
-
-When a document field is saved in the key-value store its value is binary encoded using the [bincode](https://docs.rs/bincode/) library, so a document must be serializable using serde.
-
-
-
-## How is an update handled?
-
-First of all an update in MeiliDB is nothing more than [a RocksDB SST file](https://github.com/facebook/rocksdb/wiki/Creating-and-Ingesting-SST-files). It contains the blob and all the documents attributes binary encoded like described above. Note that the blob is stored under the "_data-index_" key marked as [a merge operation](https://github.com/facebook/rocksdb/wiki/Merge-Operator-Implementation).
-
-### The merge operation is CPU consuming
-
-When [the database ingest an update](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/mod.rs#L108-L145) it gives the SST file to the underlying RocksDB, once it has ingested it there is a "_data-index_" entry available, we can request it but the key-value store will call a function before, a merge operation is performed.
-
-This merge operation is done on multiple blobs as you have understood and will compute a [PositiveBlob](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/blob/positive/blob.rs#L15), this type contains the fst and document indexes structures allowing us to search for documents. This two data structures can be considered as the inverted index.
-
-The computation time of this merge is important, RocksDB doesn't keep the previous merged result, it will call our merge operation each time until it decided to do a compaction. So [we must force this compaction earlier](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/mod.rs#L129-L131) when we receive an update to reduce this cost.
-
-This way when we request the "_data-index_" value it will gives us the previously merged positive blob without any other merge overhead.
-
-
-
-## How is a request processed?
-
-Now that we have our "_data-index_" we are able to return results based on a query. In the MeiliDB universe a query is a string.
-
-### Query lexemes
-
-The first step to be able to call the underlying structures is to split the query in words, for that we use a [custom tokenizer](https://github.com/Kerollmops/MeiliDB/blob/fc2cdf92596fc002ce278e3aa8718640ac44724d/src/tokenizer/mod.rs) that is not finished for the moment, [there is an open issue](https://github.com/Kerollmops/MeiliDB/issues/3). Note that a tokenizer is specialized for a human language, this is the hard part.
-
-### Automatons and query index
-
-So to query the fst we need an automaton, in MeiliDB we use a [levenshtein automaton](https://en.wikipedia.org/wiki/Levenshtein_automaton), this automaton is constructed using a string and a maximum distance. According to the [Algolia's blog post](https://blog.algolia.com/inside-the-algolia-engine-part-3-query-processing/#algolia%e2%80%99s-way-of-searching-for-alternatives) we [created the DFAs](https://github.com/Kerollmops/MeiliDB/blob/fc2cdf92596fc002ce278e3aa8718640ac44724d/src/automaton.rs#L62-L75) with different settings.
-
-Thanks to the power of the fst library [it is possible to union multiple automatons](https://docs.rs/fst/0.3.2/fst/map/struct.OpBuilder.html#method.union) on the same fst map, it will allow us to know which [automaton returns a word according to its index](https://github.com/Kerollmops/MeiliDB/blob/fc2cdf92596fc002ce278e3aa8718640ac44724d/src/metadata/ops.rs#L111). The `Stream` is able to return all the numbers associated to the words. We use these numbers to find the whole list of `DocIndexes` associated and do the union set operation.
-
-With all these informations it is possible [to reconstruct a list of all the DocIndexes associated](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/rank/query_builder.rs#L62-L99) with the words queried.
-
-### Sort by criteria
-
-Now that we are able to get a big list of [DocIndexes](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/lib.rs#L21-L36) it is not enough to sort them by criteria, we need more informations like the levenshtein distance or the fact that a query word match exactly the word stored in the fst. So [we stuff it a little bit](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/rank/query_builder.rs#L86-L93), and aggregate all these [Matches](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/lib.rs#L47-L74) for each document. This way it will be easy to sort a simple vector of document using a bunch of functions.
-
-With this big list of documents and associated matches [we are able to sort only the part of the slice that we want](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/rank/query_builder.rs#L108-L119) using bucket sorting. [Each criterion](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/rank/criterion/mod.rs#L75-L87) is evaluated on each subslice without copy, thanks to [GroupByMut](https://github.com/Kerollmops/group-by/blob/cab857bae01463dbd0edb99b0e0d7f3624e6c6f5/src/lib.rs#L180-L185) which, I hope [will soon be merged](https://github.com/rust-lang/rfcs/pull/2477).
-
-Note that it is possible to customize the criteria used by using the `QueryBuilder::with_criteria` constructor, this way you can implement some custom ranking based on the document attributes using the appropriate structure and the `retrieve_document` method.
-
-### Retrieve original documents
-
-The [DatabaseView](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/database_view.rs#L18-L24) structure that you must have created to be able to query the database have [two functions](https://github.com/Kerollmops/MeiliDB/blob/550dc1e99224e386516877450320f694947332d4/src/database/database_view.rs#L60-L76) that allows you to retrieve a full (or not) document according to the schema you specified at creation time (i.e. the _STORED_ attributes).
-
-As you can see, these functions force the created type `T` to implement [the serde Deserialize trait](https://docs.rs/serde/1.0.81/serde/trait.Deserialize.html), MeiliDB will use the `bincode::deserialise` function for each attribute to construct your type and return it to you.
-
-
-
-At this point, MeiliDB work is over 🎉
-

download-latest.sh

@@ -0,0 +1,129 @@
+#!/bin/sh
+
+# COLORS
+RED='\033[31m'
+GREEN='\033[32m'
+DEFAULT='\033[0m'
+
+# GLOBALS
+GREP_SEMVER_REGEXP='\"v\([0-9]*\)[.]\([0-9]*\)[.]\([0-9]*\)\"' # i.e. "v[number].[number].[number]"
+BINARY_NAME='meilisearch'
+
+# semverParseInto and semverLT from https://github.com/cloudflare/semver_bash/blob/master/semver.sh
+
+# usage: semverParseInto version major minor patch special
+# version: the string version
+# major, minor, patch, special: will be assigned by the function
+semverParseInto() {
+    local RE='[^0-9]*\([0-9]*\)[.]\([0-9]*\)[.]\([0-9]*\)\([0-9A-Za-z-]*\)'
+    #MAJOR
+    eval $2=`echo $1 | sed -e "s#$RE#\1#"`
+    #MINOR
+    eval $3=`echo $1 | sed -e "s#$RE#\2#"`
+    #MINOR
+    eval $4=`echo $1 | sed -e "s#$RE#\3#"`
+    #SPECIAL
+    eval $5=`echo $1 | sed -e "s#$RE#\4#"`
+}
+
+# usage: semverLT version1 version2
+semverLT() {
+    local MAJOR_A=0
+    local MINOR_A=0
+    local PATCH_A=0
+    local SPECIAL_A=0
+
+    local MAJOR_B=0
+    local MINOR_B=0
+    local PATCH_B=0
+    local SPECIAL_B=0
+
+    semverParseInto $1 MAJOR_A MINOR_A PATCH_A SPECIAL_A
+    semverParseInto $2 MAJOR_B MINOR_B PATCH_B SPECIAL_B
+
+    if [ $MAJOR_A -lt $MAJOR_B ]; then
+        return 0
+    fi
+    if [ $MAJOR_A -le $MAJOR_B ] && [ $MINOR_A -lt $MINOR_B ]; then
+        return 0
+    fi
+    if [ $MAJOR_A -le $MAJOR_B ] && [ $MINOR_A -le $MINOR_B ] && [ $PATCH_A -lt $PATCH_B ]; then
+        return 0
+    fi
+    if [ "_$SPECIAL_A"  == "_" ] && [ "_$SPECIAL_B"  == "_" ] ; then
+        return 1
+    fi
+    if [ "_$SPECIAL_A"  == "_" ] && [ "_$SPECIAL_B"  != "_" ] ; then
+        return 1
+    fi
+    if [ "_$SPECIAL_A"  != "_" ] && [ "_$SPECIAL_B"  == "_" ] ; then
+        return 0
+    fi
+    if [ "_$SPECIAL_A" < "_$SPECIAL_B" ]; then
+        return 0
+    fi
+
+    return 1
+}
+
+success_usage() {
+    printf "$GREEN%s\n$DEFAULT" "MeiliSearch binary successfully downloaded as '$BINARY_NAME' file."
+    echo ''
+    echo 'Run it:'
+    echo '    $ ./meilisearch'
+    echo 'Usage:'
+    echo '    $ ./meilisearch --help'
+}
+
+failure_usage() {
+    printf "$RED%s\n$DEFAULT" 'ERROR: MeiliSearch binary is not available for your OS distribution yet.'
+    echo ''
+    echo 'However, you can easily compile the binary from the source files.'
+    echo 'Follow the steps on the docs: https://docs.meilisearch.com/advanced_guides/binary.html#how-to-compile-meilisearch'
+}
+
+# OS DETECTION
+echo 'Detecting OS distribution...'
+os_name=$(uname -s)
+if [ "$os_name" != "Darwin" ]; then
+    os_name=$(cat /etc/os-release | grep '^ID=' | tr -d '"' | cut -d '=' -f 2)
+fi
+echo "OS distribution detected: $os_name"
+case "$os_name" in
+'Darwin')
+    os='macos'
+    ;;
+'ubuntu' | 'debian')
+    os='linux'
+    ;;
+*)
+    failure_usage
+    exit 1
+esac
+
+# GET LATEST VERSION
+tags=$(curl -s 'https://api.github.com/repos/meilisearch/MeiliSearch/tags' \
+    | grep "$GREP_SEMVER_REGEXP" \
+    | grep 'name' \
+    | tr -d '"' | tr -d ',' | cut -d 'v' -f 2)
+
+latest=""
+for tag in $tags; do
+    if [ "$latest" = "" ]; then
+        latest="$tag"
+    else
+        semverLT $tag $latest
+        if [ $? -eq 1 ]; then
+            latest="$tag"
+        fi
+    fi
+done
+
+# DOWNLOAD THE LATEST
+echo "Downloading MeiliSearch binary v$latest for $os..."
+release_file="meilisearch-$os-amd64"
+link="https://github.com/meilisearch/MeiliSearch/releases/download/v$latest/$release_file"
+curl -OL "$link"
+mv "$release_file" "$BINARY_NAME"
+chmod 744 "$BINARY_NAME"
+success_usage

examples/create-database.rs

@@ -1,138 +0,0 @@
-#[global_allocator]
-static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc;
-
-use std::io::{self, BufRead, BufReader};
-use std::path::{Path, PathBuf};
-use std::error::Error;
-use std::borrow::Cow;
-use std::fs::File;
-
-use hashbrown::{HashMap, HashSet};
-use serde_derive::{Serialize, Deserialize};
-use structopt::StructOpt;
-
-use meilidb::database::{Database, Schema, UpdateBuilder};
-use meilidb::tokenizer::DefaultBuilder;
-
-#[derive(Debug, StructOpt)]
-pub struct Opt {
-    /// The destination where the database must be created.
-    #[structopt(parse(from_os_str))]
-    pub database_path: PathBuf,
-
-    /// The csv file to index.
-    #[structopt(parse(from_os_str))]
-    pub csv_data_path: PathBuf,
-
-    /// The path to the schema.
-    #[structopt(long = "schema", parse(from_os_str))]
-    pub schema_path: PathBuf,
-
-    /// The path to the list of stop words (one by line).
-    #[structopt(long = "stop-words", parse(from_os_str))]
-    pub stop_words_path: Option<PathBuf>,
-
-    #[structopt(long = "update-group-size")]
-    pub update_group_size: Option<usize>,
-}
-
-#[derive(Serialize, Deserialize)]
-struct Document<'a> (
-    #[serde(borrow)]
-    HashMap<Cow<'a, str>, Cow<'a, str>>
-);
-
-fn index(
-    schema: Schema,
-    database_path: &Path,
-    csv_data_path: &Path,
-    update_group_size: Option<usize>,
-    stop_words: &HashSet<String>,
-) -> Result<Database, Box<Error>>
-{
-    let database = Database::create(database_path, &schema)?;
-
-    let mut rdr = csv::Reader::from_path(csv_data_path)?;
-    let mut raw_record = csv::StringRecord::new();
-    let headers = rdr.headers()?.clone();
-
-    let mut i = 0;
-    let mut end_of_file = false;
-
-    while !end_of_file {
-        let tokenizer_builder = DefaultBuilder::new();
-        let update_path = tempfile::NamedTempFile::new()?;
-        let mut update = UpdateBuilder::new(update_path.path().to_path_buf(), schema.clone());
-
-        loop {
-            end_of_file = !rdr.read_record(&mut raw_record)?;
-            if end_of_file { break }
-
-            let document: Document = match raw_record.deserialize(Some(&headers)) {
-                Ok(document) => document,
-                Err(e) => {
-                    eprintln!("{:?}", e);
-                    continue;
-                }
-            };
-
-            update.update_document(&document, &tokenizer_builder, &stop_words)?;
-
-            print!("\rindexing document {}", i);
-            i += 1;
-
-            if let Some(group_size) = update_group_size {
-                if i % group_size == 0 { break }
-            }
-        }
-
-        println!();
-
-        println!("building update...");
-        let update = update.build()?;
-        println!("ingesting update...");
-        database.ingest_update_file(update)?;
-    }
-
-    Ok(database)
-}
-
-fn retrieve_stop_words(path: &Path) -> io::Result<HashSet<String>> {
-    let f = File::open(path)?;
-    let reader = BufReader::new(f);
-    let mut words = HashSet::new();
-
-    for line in reader.lines() {
-        let line = line?;
-        let word = line.trim().to_string();
-        words.insert(word);
-    }
-
-    Ok(words)
-}
-
-fn main() -> Result<(), Box<Error>> {
-    let _ = env_logger::init();
-    let opt = Opt::from_args();
-
-    let schema = {
-        let file = File::open(&opt.schema_path)?;
-        Schema::from_toml(file)?
-    };
-
-    let stop_words = match opt.stop_words_path {
-        Some(ref path) => retrieve_stop_words(path)?,
-        None           => HashSet::new(),
-    };
-
-    let (elapsed, result) = elapsed::measure_time(|| {
-        index(schema, &opt.database_path, &opt.csv_data_path, opt.update_group_size, &stop_words)
-    });
-
-    if let Err(e) = result {
-        return Err(e.into())
-    }
-
-    println!("database created in {} at: {:?}", elapsed, opt.database_path);
-    Ok(())
-}

examples/query-database.rs

@@ -1,170 +0,0 @@
-#[global_allocator]
-static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc;
-
-use std::collections::btree_map::{BTreeMap, Entry};
-use std::iter::FromIterator;
-use std::io::{self, Write};
-use std::path::PathBuf;
-use std::error::Error;
-
-use hashbrown::{HashMap, HashSet};
-use termcolor::{Color, ColorChoice, ColorSpec, StandardStream, WriteColor};
-use structopt::StructOpt;
-
-use meilidb::database::schema::SchemaAttr;
-use meilidb::database::Database;
-use meilidb::Match;
-
-#[derive(Debug, StructOpt)]
-pub struct Opt {
-    /// The destination where the database must be created
-    #[structopt(parse(from_os_str))]
-    pub database_path: PathBuf,
-
-    /// Fields that must be displayed.
-    pub displayed_fields: Vec<String>,
-
-    /// The number of returned results
-    #[structopt(short = "n", long = "number-results", default_value = "10")]
-    pub number_results: usize,
-}
-
-type Document = HashMap<String, String>;
-
-fn display_highlights(text: &str, ranges: &[usize]) -> io::Result<()> {
-    let mut stdout = StandardStream::stdout(ColorChoice::Always);
-    let mut highlighted = false;
-
-    for range in ranges.windows(2) {
-        let [start, end] = match range { [start, end] => [*start, *end], _ => unreachable!() };
-        if highlighted {
-            stdout.set_color(ColorSpec::new().set_fg(Some(Color::Yellow)))?;
-        }
-        write!(&mut stdout, "{}", &text[start..end])?;
-        stdout.reset()?;
-        highlighted = !highlighted;
-    }
-
-    Ok(())
-}
-
-fn char_to_byte_range(index: usize, length: usize, text: &str) -> (usize, usize) {
-    let mut byte_index = 0;
-    let mut byte_length = 0;
-
-    for (n, (i, c)) in text.char_indices().enumerate() {
-        if n == index {
-            byte_index = i;
-        }
-
-        if n + 1 == index + length {
-            byte_length = i - byte_index + c.len_utf8();
-            break;
-        }
-    }
-
-    (byte_index, byte_length)
-}
-
-fn create_highlight_areas(text: &str, matches: &[Match], attribute: SchemaAttr) -> Vec<usize> {
-    let mut byte_indexes = BTreeMap::new();
-
-    for match_ in matches {
-        let match_attribute = match_.attribute.attribute();
-        if SchemaAttr::new(match_attribute) == attribute {
-            let word_area = match_.word_area;
-
-            let char_index = word_area.char_index() as usize;
-            let char_length = word_area.length() as usize;
-            let (byte_index, byte_length) = char_to_byte_range(char_index, char_length, text);
-
-            match byte_indexes.entry(byte_index) {
-                Entry::Vacant(entry) => { entry.insert(byte_length); },
-                Entry::Occupied(mut entry) => {
-                    if *entry.get() < byte_length {
-                        entry.insert(byte_length);
-                    }
-                },
-            }
-        }
-    }
-
-    let mut title_areas = Vec::new();
-    title_areas.push(0);
-    for (byte_index, length) in byte_indexes {
-        title_areas.push(byte_index);
-        title_areas.push(byte_index + length);
-    }
-    title_areas.push(text.len());
-    title_areas.sort_unstable();
-    title_areas
-}
-
-fn main() -> Result<(), Box<Error>> {
-    let _ = env_logger::init();
-    let opt = Opt::from_args();
-
-    let (elapsed, result) = elapsed::measure_time(|| Database::open(&opt.database_path));
-    let database = result?;
-    println!("database prepared for you in {}", elapsed);
-
-    let mut buffer = String::new();
-    let input = io::stdin();
-
-    loop {
-        print!("Searching for: ");
-        io::stdout().flush()?;
-
-        if input.read_line(&mut buffer)? == 0 { break }
-        let query = buffer.trim_end_matches('\n');
-
-        let view = database.view();
-        let schema = view.schema();
-
-        let (elapsed, documents) = elapsed::measure_time(|| {
-            let builder = view.query_builder().unwrap();
-            builder.query(query, 0..opt.number_results)
-        });
-
-        let number_of_documents = documents.len();
-        for doc in documents {
-            match view.document_by_id::<Document>(doc.id) {
-                Ok(document) => {
-                    for name in &opt.displayed_fields {
-                        let attr = match schema.attribute(name) {
-                            Some(attr) => attr,
-                            None => continue,
-                        };
-                        let text = match document.get(name) {
-                            Some(text) => text,
-                            None => continue,
-                        };
-
-                        print!("{}: ", name);
-                        let areas = create_highlight_areas(&text, &doc.matches, attr);
-                        display_highlights(&text, &areas)?;
-                        println!();
-                    }
-                },
-                Err(e) => eprintln!("{}", e),
-            }
-
-            let mut matching_attributes = HashSet::new();
-            for _match in doc.matches {
-                let attr = SchemaAttr::new(_match.attribute.attribute());
-                let name = schema.attribute_name(attr);
-                matching_attributes.insert(name);
-            }
-
-            let matching_attributes = Vec::from_iter(matching_attributes);
-            println!("matching in: {:?}", matching_attributes);
-
-            println!();
-        }
-
-        eprintln!("===== Found {} results in {} =====", number_of_documents, elapsed);
-        buffer.clear();
-    }
-
-    Ok(())
-}

examples/schema-example.toml

@@ -1,19 +0,0 @@
-# This schema has been generated ...
-# The order in which the attributes are declared is important,
-# it specify the attribute xxx...
-
-identifier = "id"
-
-[attributes.id]
-stored = true
-
-[attributes.title]
-stored = true
-indexed = true
-
-[attributes.description]
-stored = true
-indexed = true
-
-[attributes.image]
-stored = true

meilisearch-core/Cargo.toml

@@ -0,0 +1,53 @@
+[package]
+name = "meilisearch-core"
+version = "0.10.1"
+license = "MIT"
+authors = ["Kerollmops <clement@meilisearch.com>"]
+edition = "2018"
+
+[dependencies]
+arc-swap = "0.4.5"
+bincode = "1.2.1"
+byteorder = "1.3.4"
+chrono = { version = "0.4.11", features = ["serde"] }
+compact_arena = "0.4.0"
+crossbeam-channel = "0.4.2"
+deunicode = "1.1.0"
+env_logger = "0.7.1"
+fst = { version = "0.3.5", default-features = false }
+hashbrown = { version = "0.7.1", features = ["serde"] }
+heed = "0.7.0"
+indexmap = { version = "1.3.2", features = ["serde-1"] }
+intervaltree = "0.2.5"
+itertools = "0.9.0"
+levenshtein_automata = { version = "0.1.1", features = ["fst_automaton"] }
+log = "0.4.8"
+meilisearch-schema = { path = "../meilisearch-schema", version = "0.10.1" }
+meilisearch-tokenizer = { path = "../meilisearch-tokenizer", version = "0.10.1" }
+meilisearch-types = { path = "../meilisearch-types", version = "0.10.1" }
+once_cell = "1.3.1"
+ordered-float = { version = "1.0.2", features = ["serde"] }
+pest = { git = "https://github.com/MarinPostma/pest.git", tag = "meilisearch-patch1" }
+pest_derive = "2.0"
+regex = "1.3.6"
+sdset = "0.4.0"
+serde = { version = "1.0.105", features = ["derive"] }
+serde_json = "1.0.50"
+siphasher = "0.3.2"
+slice-group-by = "0.2.6"
+unicase = "2.6.0"
+zerocopy = "0.3.0"
+
+[dev-dependencies]
+assert_matches = "1.3.0"
+criterion = "0.3.1"
+csv = "1.1.3"
+jemallocator = "0.3.2"
+rustyline = { version = "6.0.0", default-features = false }
+structopt = "0.3.12"
+tempfile = "3.1.0"
+termcolor = "1.1.0"
+
+[[bench]]
+name = "search_benchmark"
+harness = false

meilisearch-core/benches/search_benchmark.rs

@@ -0,0 +1,104 @@
+#[cfg(test)]
+#[macro_use]
+extern crate assert_matches;
+
+use std::sync::mpsc;
+use std::path::Path;
+use std::fs::File;
+use std::io::BufReader;
+use std::iter;
+
+use meilisearch_core::Database;
+use meilisearch_core::{ProcessedUpdateResult, UpdateStatus};
+use meilisearch_core::settings::{Settings, SettingsUpdate};
+use meilisearch_schema::Schema;
+use serde_json::Value;
+
+use criterion::{criterion_group, criterion_main, Criterion, BenchmarkId};
+
+fn prepare_database(path: &Path) -> Database {
+    let database = Database::open_or_create(path).unwrap();
+    let db = &database;
+
+    let (sender, receiver) = mpsc::sync_channel(100);
+    let update_fn = move |_name: &str, update: ProcessedUpdateResult| {
+        sender.send(update.update_id).unwrap()
+    };
+    let index = database.create_index("bench").unwrap();
+
+    database.set_update_callback(Box::new(update_fn));
+    
+    let mut writer = db.main_write_txn().unwrap();
+    index.main.put_schema(&mut writer, &Schema::with_primary_key("id")).unwrap();
+    writer.commit().unwrap();
+
+    let settings_update: SettingsUpdate = {
+        let path = concat!(env!("CARGO_MANIFEST_DIR"), "/../datasets/movies/settings.json");
+        let file = File::open(path).unwrap();
+        let reader = BufReader::new(file);
+        let settings: Settings = serde_json::from_reader(reader).unwrap();
+        settings.into_update().unwrap()
+    };
+
+    let mut update_writer = db.update_write_txn().unwrap();
+    let _update_id = index.settings_update(&mut update_writer, settings_update).unwrap();
+    update_writer.commit().unwrap();
+
+    let mut additions = index.documents_addition();
+
+    let json: Value = {
+        let path = concat!(env!("CARGO_MANIFEST_DIR"), "/../datasets/movies/movies.json");
+        let movies_file = File::open(path).expect("find movies");
+        serde_json::from_reader(movies_file).unwrap()
+    };
+
+    let documents = json.as_array().unwrap();
+
+    for document in documents {
+        additions.update_document(document);
+    }
+
+    let mut update_writer = db.update_write_txn().unwrap();
+    let update_id = additions.finalize(&mut update_writer).unwrap();
+    update_writer.commit().unwrap();
+
+    // block until the transaction is processed
+    let _ = receiver.into_iter().find(|id| *id == update_id);
+
+    let update_reader = db.update_read_txn().unwrap();
+    let result = index.update_status(&update_reader, update_id).unwrap();
+    assert_matches!(result, Some(UpdateStatus::Processed { content }) if content.error.is_none());
+
+    database
+}
+
+pub fn criterion_benchmark(c: &mut Criterion) {
+    let dir = tempfile::tempdir().unwrap();
+    let database = prepare_database(dir.path());
+
+    let reader = database.main_read_txn().unwrap();
+    let index = database.open_index("bench").unwrap();
+
+    let mut count = 0;
+    let query = "I love paris ";
+
+    let iter = iter::from_fn(|| {
+        count += 1;
+        query.get(0..count)
+    });
+
+    let mut group = c.benchmark_group("searching in movies (19654 docs)");
+    group.sample_size(10);
+
+    for query in iter {
+        let bench_name = BenchmarkId::from_parameter(format!("{:?}", query));
+        group.bench_with_input(bench_name, &query, |b, query| b.iter(|| {
+            let builder = index.query_builder();
+            builder.query(&reader, query, 0..20).unwrap();
+        }));
+    }
+    group.finish();
+}
+
+criterion_group!(benches, criterion_benchmark);
+criterion_main!(benches);

meilisearch-core/examples/from_file.rs

@@ -0,0 +1,473 @@
+use std::collections::HashSet;
+use std::collections::btree_map::{BTreeMap, Entry};
+use std::error::Error;
+use std::io::{Read, Write};
+use std::iter::FromIterator;
+use std::path::{Path, PathBuf};
+use std::time::{Duration, Instant};
+use std::{fs, io, sync::mpsc};
+
+use rustyline::{Config, Editor};
+use serde::{Deserialize, Serialize};
+use structopt::StructOpt;
+use termcolor::{Color, ColorChoice, ColorSpec, StandardStream, WriteColor};
+
+use meilisearch_core::{Database, Highlight, ProcessedUpdateResult};
+use meilisearch_core::settings::Settings;
+use meilisearch_schema::FieldId;
+
+// #[cfg(target_os = "linux")]
+#[global_allocator]
+static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc;
+
+#[derive(Debug, StructOpt)]
+struct IndexCommand {
+    /// The destination where the database must be created.
+    #[structopt(parse(from_os_str))]
+    database_path: PathBuf,
+
+    #[structopt(long, default_value = "default")]
+    index_uid: String,
+
+    /// The csv file path to index, you can also use `-` to specify the standard input.
+    #[structopt(parse(from_os_str))]
+    csv_data_path: PathBuf,
+
+    /// The path to the settings.
+    #[structopt(long, parse(from_os_str))]
+    settings: PathBuf,
+
+    #[structopt(long)]
+    update_group_size: Option<usize>,
+
+    #[structopt(long, parse(from_os_str))]
+    compact_to_path: Option<PathBuf>,
+}
+
+#[derive(Debug, StructOpt)]
+struct SearchCommand {
+    /// The path of the database to work with.
+    #[structopt(parse(from_os_str))]
+    database_path: PathBuf,
+
+    #[structopt(long, default_value = "default")]
+    index_uid: String,
+
+    /// Timeout after which the search will return results.
+    #[structopt(long)]
+    fetch_timeout_ms: Option<u64>,
+
+    /// The number of returned results
+    #[structopt(short, long, default_value = "10")]
+    number_results: usize,
+
+    /// The number of characters before and after the first match
+    #[structopt(short = "C", long, default_value = "35")]
+    char_context: usize,
+
+    /// A filter string that can be `!adult` or `adult` to
+    /// filter documents on this specfied field
+    #[structopt(short, long)]
+    filter: Option<String>,
+
+    /// Fields that must be displayed.
+    displayed_fields: Vec<String>,
+}
+
+#[derive(Debug, StructOpt)]
+struct ShowUpdatesCommand {
+    /// The path of the database to work with.
+    #[structopt(parse(from_os_str))]
+    database_path: PathBuf,
+
+    #[structopt(long, default_value = "default")]
+    index_uid: String,
+}
+
+#[derive(Debug, StructOpt)]
+enum Command {
+    Index(IndexCommand),
+    Search(SearchCommand),
+    ShowUpdates(ShowUpdatesCommand),
+}
+
+impl Command {
+    fn path(&self) -> &Path {
+        match self {
+            Command::Index(command) => &command.database_path,
+            Command::Search(command) => &command.database_path,
+            Command::ShowUpdates(command) => &command.database_path,
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize)]
+#[serde(transparent)]
+struct Document(indexmap::IndexMap<String, String>);
+
+fn index_command(command: IndexCommand, database: Database) -> Result<(), Box<dyn Error>> {
+    let start = Instant::now();
+
+    let (sender, receiver) = mpsc::sync_channel(100);
+    let update_fn =
+        move |_name: &str, update: ProcessedUpdateResult| sender.send(update.update_id).unwrap();
+    let index = match database.open_index(&command.index_uid) {
+        Some(index) => index,
+        None => database.create_index(&command.index_uid).unwrap(),
+    };
+
+    database.set_update_callback(Box::new(update_fn));
+
+    let db = &database;
+
+    let settings = {
+        let string = fs::read_to_string(&command.settings)?;
+        let settings: Settings = serde_json::from_str(&string).unwrap();
+        settings.into_update().unwrap()
+    };
+
+    let mut update_writer = db.update_write_txn().unwrap();
+    index.settings_update(&mut update_writer, settings)?;
+    update_writer.commit().unwrap();
+
+    let mut rdr = if command.csv_data_path.as_os_str() == "-" {
+        csv::Reader::from_reader(Box::new(io::stdin()) as Box<dyn Read>)
+    } else {
+        let file = std::fs::File::open(command.csv_data_path)?;
+        csv::Reader::from_reader(Box::new(file) as Box<dyn Read>)
+    };
+
+    let mut raw_record = csv::StringRecord::new();
+    let headers = rdr.headers()?.clone();
+
+    let mut max_update_id = 0;
+    let mut i = 0;
+    let mut end_of_file = false;
+
+    while !end_of_file {
+        let mut additions = index.documents_addition();
+
+        loop {
+            end_of_file = !rdr.read_record(&mut raw_record)?;
+            if end_of_file {
+                break;
+            }
+
+            let document: Document = match raw_record.deserialize(Some(&headers)) {
+                Ok(document) => document,
+                Err(e) => {
+                    eprintln!("{:?}", e);
+                    continue;
+                }
+            };
+
+            additions.update_document(document);
+
+            print!("\rindexing document {}", i);
+            i += 1;
+
+            if let Some(group_size) = command.update_group_size {
+                if i % group_size == 0 {
+                    break;
+                }
+            }
+        }
+
+        println!();
+
+        let mut update_writer = db.update_write_txn().unwrap();
+        println!("committing update...");
+        let update_id = additions.finalize(&mut update_writer)?;
+        update_writer.commit().unwrap();
+        max_update_id = max_update_id.max(update_id);
+        println!("committed update {}", update_id);
+    }
+
+    println!("Waiting for update {}", max_update_id);
+    for id in receiver {
+        if id == max_update_id {
+            break;
+        }
+    }
+
+    println!(
+        "database created in {:.2?} at: {:?}",
+        start.elapsed(),
+        command.database_path
+    );
+
+    if let Some(path) = command.compact_to_path {
+        fs::create_dir_all(&path)?;
+        let start = Instant::now();
+        let _file = database.copy_and_compact_to_path(path.join("data.mdb"))?;
+        println!(
+            "database compacted in {:.2?} at: {:?}",
+            start.elapsed(),
+            path
+        );
+    }
+
+    Ok(())
+}
+
+fn display_highlights(text: &str, ranges: &[usize]) -> io::Result<()> {
+    let mut stdout = StandardStream::stdout(ColorChoice::Always);
+    let mut highlighted = false;
+
+    for range in ranges.windows(2) {
+        let [start, end] = match range {
+            [start, end] => [*start, *end],
+            _ => unreachable!(),
+        };
+        if highlighted {
+            stdout.set_color(
+                ColorSpec::new()
+                    .set_fg(Some(Color::Yellow))
+                    .set_underline(true),
+            )?;
+        }
+        write!(&mut stdout, "{}", &text[start..end])?;
+        stdout.reset()?;
+        highlighted = !highlighted;
+    }
+
+    Ok(())
+}
+
+fn char_to_byte_range(index: usize, length: usize, text: &str) -> (usize, usize) {
+    let mut byte_index = 0;
+    let mut byte_length = 0;
+
+    for (n, (i, c)) in text.char_indices().enumerate() {
+        if n == index {
+            byte_index = i;
+        }
+
+        if n + 1 == index + length {
+            byte_length = i - byte_index + c.len_utf8();
+            break;
+        }
+    }
+
+    (byte_index, byte_length)
+}
+
+fn create_highlight_areas(text: &str, highlights: &[Highlight]) -> Vec<usize> {
+    let mut byte_indexes = BTreeMap::new();
+
+    for highlight in highlights {
+        let char_index = highlight.char_index as usize;
+        let char_length = highlight.char_length as usize;
+        let (byte_index, byte_length) = char_to_byte_range(char_index, char_length, text);
+
+        match byte_indexes.entry(byte_index) {
+            Entry::Vacant(entry) => {
+                entry.insert(byte_length);
+            }
+            Entry::Occupied(mut entry) => {
+                if *entry.get() < byte_length {
+                    entry.insert(byte_length);
+                }
+            }
+        }
+    }
+
+    let mut title_areas = Vec::new();
+    title_areas.push(0);
+    for (byte_index, length) in byte_indexes {
+        title_areas.push(byte_index);
+        title_areas.push(byte_index + length);
+    }
+    title_areas.push(text.len());
+    title_areas.sort_unstable();
+    title_areas
+}
+
+/// note: matches must have been sorted by `char_index` and `char_length` before being passed.
+///
+/// ```no_run
+/// matches.sort_unstable_by_key(|m| (m.char_index, m.char_length));
+///
+/// let matches = matches.matches.iter().filter(|m| SchemaAttr::new(m.attribute) == attr).cloned();
+///
+/// let (text, matches) = crop_text(&text, matches, 35);
+/// ```
+fn crop_text(
+    text: &str,
+    highlights: impl IntoIterator<Item = Highlight>,
+    context: usize,
+) -> (String, Vec<Highlight>) {
+    let mut highlights = highlights.into_iter().peekable();
+
+    let char_index = highlights
+        .peek()
+        .map(|m| m.char_index as usize)
+        .unwrap_or(0);
+    let start = char_index.saturating_sub(context);
+    let text = text.chars().skip(start).take(context * 2).collect();
+
+    let highlights = highlights
+        .take_while(|m| (m.char_index as usize) + (m.char_length as usize) <= start + (context * 2))
+        .map(|highlight| Highlight {
+            char_index: highlight.char_index - start as u16,
+            ..highlight
+        })
+        .collect();
+
+    (text, highlights)
+}
+
+fn search_command(command: SearchCommand, database: Database) -> Result<(), Box<dyn Error>> {
+    let db = &database;
+    let index = database
+        .open_index(&command.index_uid)
+        .expect("Could not find index");
+
+    let reader = db.main_read_txn().unwrap();
+    let schema = index.main.schema(&reader)?;
+    reader.abort();
+
+    let schema = schema.ok_or(meilisearch_core::Error::SchemaMissing)?;
+
+    let fields = command.displayed_fields.iter().map(String::as_str);
+    let fields = HashSet::from_iter(fields);
+
+    let config = Config::builder().auto_add_history(true).build();
+    let mut readline = Editor::<()>::with_config(config);
+    let _ = readline.load_history("query-history.txt");
+
+    for result in readline.iter("Searching for: ") {
+        match result {
+            Ok(query) => {
+                let start_total = Instant::now();
+
+                let reader = db.main_read_txn().unwrap();
+                let ref_index = &index;
+                let ref_reader = &reader;
+
+                let mut builder = index.query_builder();
+                if let Some(timeout) = command.fetch_timeout_ms {
+                    builder.with_fetch_timeout(Duration::from_millis(timeout));
+                }
+
+                if let Some(ref filter) = command.filter {
+                    let filter = filter.as_str();
+                    let (positive, filter) = if filter.chars().next() == Some('!') {
+                        (false, &filter[1..])
+                    } else {
+                        (true, filter)
+                    };
+
+                    let attr = schema
+                        .id(filter)
+                        .expect("Could not find filtered attribute");
+
+                    builder.with_filter(move |document_id| {
+                        let string: String = ref_index
+                            .document_attribute(ref_reader, document_id, attr)
+                            .unwrap()
+                            .unwrap();
+                        (string == "true") == positive
+                    });
+                }
+
+                let (documents, _nb_hits) = builder.query(ref_reader, &query, 0..command.number_results)?;
+
+                let mut retrieve_duration = Duration::default();
+
+                let number_of_documents = documents.len();
+                for mut doc in documents {
+                    doc.highlights
+                        .sort_unstable_by_key(|m| (m.char_index, m.char_length));
+
+                    let start_retrieve = Instant::now();
+                    let result = index.document::<Document>(&reader, Some(&fields), doc.id);
+                    retrieve_duration += start_retrieve.elapsed();
+
+                    match result {
+                        Ok(Some(document)) => {
+                            println!("raw-id: {:?}", doc.id);
+                            for (name, text) in document.0 {
+                                print!("{}: ", name);
+
+                                let attr = schema.id(&name).unwrap();
+                                let highlights = doc
+                                    .highlights
+                                    .iter()
+                                    .filter(|m| FieldId::new(m.attribute) == attr)
+                                    .cloned();
+                                let (text, highlights) =
+                                    crop_text(&text, highlights, command.char_context);
+                                let areas = create_highlight_areas(&text, &highlights);
+                                display_highlights(&text, &areas)?;
+                                println!();
+                            }
+                        }
+                        Ok(None) => eprintln!("missing document"),
+                        Err(e) => eprintln!("{}", e),
+                    }
+
+                    let mut matching_attributes = HashSet::new();
+                    for highlight in doc.highlights {
+                        let attr = FieldId::new(highlight.attribute);
+                        let name = schema.name(attr);
+                        matching_attributes.insert(name);
+                    }
+
+                    let matching_attributes = Vec::from_iter(matching_attributes);
+                    println!("matching in: {:?}", matching_attributes);
+
+                    println!();
+                }
+
+                eprintln!(
+                    "whole documents fields retrieve took {:.2?}",
+                    retrieve_duration
+                );
+                eprintln!(
+                    "===== Found {} results in {:.2?} =====",
+                    number_of_documents,
+                    start_total.elapsed()
+                );
+            }
+            Err(err) => {
+                println!("Error: {:?}", err);
+                break;
+            }
+        }
+    }
+
+    readline.save_history("query-history.txt").unwrap();
+
+    Ok(())
+}
+
+fn show_updates_command(
+    command: ShowUpdatesCommand,
+    database: Database,
+) -> Result<(), Box<dyn Error>> {
+    let db = &database;
+    let index = database
+        .open_index(&command.index_uid)
+        .expect("Could not find index");
+
+    let reader = db.update_read_txn().unwrap();
+    let updates = index.all_updates_status(&reader)?;
+    println!("{:#?}", updates);
+    reader.abort();
+
+    Ok(())
+}
+
+fn main() -> Result<(), Box<dyn Error>> {
+    env_logger::init();
+
+    let opt = Command::from_args();
+    let database = Database::open_or_create(opt.path())?;
+
+    match opt {
+        Command::Index(command) => index_command(command, database),
+        Command::Search(command) => search_command(command, database),
+        Command::ShowUpdates(command) => show_updates_command(command, database),
+    }
+}

meilisearch-core/src/automaton/dfa.rs

@@ -0,0 +1,53 @@
+use levenshtein_automata::{LevenshteinAutomatonBuilder as LevBuilder, DFA};
+use once_cell::sync::OnceCell;
+
+static LEVDIST0: OnceCell<LevBuilder> = OnceCell::new();
+static LEVDIST1: OnceCell<LevBuilder> = OnceCell::new();
+static LEVDIST2: OnceCell<LevBuilder> = OnceCell::new();
+
+#[derive(Copy, Clone)]
+enum PrefixSetting {
+    Prefix,
+    NoPrefix,
+}
+
+fn build_dfa_with_setting(query: &str, setting: PrefixSetting) -> DFA {
+    use PrefixSetting::{NoPrefix, Prefix};
+
+    match query.len() {
+        0..=4 => {
+            let builder = LEVDIST0.get_or_init(|| LevBuilder::new(0, true));
+            match setting {
+                Prefix => builder.build_prefix_dfa(query),
+                NoPrefix => builder.build_dfa(query),
+            }
+        }
+        5..=8 => {
+            let builder = LEVDIST1.get_or_init(|| LevBuilder::new(1, true));
+            match setting {
+                Prefix => builder.build_prefix_dfa(query),
+                NoPrefix => builder.build_dfa(query),
+            }
+        }
+        _ => {
+            let builder = LEVDIST2.get_or_init(|| LevBuilder::new(2, true));
+            match setting {
+                Prefix => builder.build_prefix_dfa(query),
+                NoPrefix => builder.build_dfa(query),
+            }
+        }
+    }
+}
+
+pub fn build_prefix_dfa(query: &str) -> DFA {
+    build_dfa_with_setting(query, PrefixSetting::Prefix)
+}
+
+pub fn build_dfa(query: &str) -> DFA {
+    build_dfa_with_setting(query, PrefixSetting::NoPrefix)
+}
+
+pub fn build_exact_dfa(query: &str) -> DFA {
+    let builder = LEVDIST0.get_or_init(|| LevBuilder::new(0, true));
+    builder.build_dfa(query)
+}

meilisearch-core/src/automaton/mod.rs

@@ -0,0 +1,15 @@
+mod dfa;
+
+use meilisearch_tokenizer::is_cjk;
+
+pub use self::dfa::{build_dfa, build_prefix_dfa, build_exact_dfa};
+
+pub fn normalize_str(string: &str) -> String {
+    let mut string = string.to_lowercase();
+
+    if !string.contains(is_cjk) {
+        string = deunicode::deunicode_with_tofu(&string, "");
+    }
+
+    string
+}

meilisearch-core/src/bucket_sort.rs

@@ -0,0 +1,560 @@
+use std::borrow::Cow;
+use std::collections::HashMap;
+use std::mem;
+use std::ops::Deref;
+use std::ops::Range;
+use std::rc::Rc;
+use std::sync::atomic::{AtomicUsize, Ordering};
+use std::time::Instant;
+use std::fmt;
+
+use compact_arena::{SmallArena, Idx32, mk_arena};
+use log::debug;
+use meilisearch_types::DocIndex;
+use sdset::{Set, SetBuf, exponential_search};
+use slice_group_by::{GroupBy, GroupByMut};
+
+use crate::error::Error;
+use crate::criterion::{Criteria, Context, ContextMut};
+use crate::distinct_map::{BufferedDistinctMap, DistinctMap};
+use crate::raw_document::RawDocument;
+use crate::{database::MainT, reordered_attrs::ReorderedAttrs};
+use crate::{store, Document, DocumentId, MResult};
+use crate::query_tree::{create_query_tree, traverse_query_tree};
+use crate::query_tree::{Operation, QueryResult, QueryKind, QueryId, PostingsKey};
+use crate::query_tree::Context as QTContext;
+
+pub fn bucket_sort<'c, FI>(
+    reader: &heed::RoTxn<MainT>,
+    query: &str,
+    range: Range<usize>,
+    filter: Option<FI>,
+    criteria: Criteria<'c>,
+    searchable_attrs: Option<ReorderedAttrs>,
+    main_store: store::Main,
+    postings_lists_store: store::PostingsLists,
+    documents_fields_counts_store: store::DocumentsFieldsCounts,
+    synonyms_store: store::Synonyms,
+    prefix_documents_cache_store: store::PrefixDocumentsCache,
+    prefix_postings_lists_cache_store: store::PrefixPostingsListsCache,
+) -> MResult<(Vec<Document>, usize)>
+where
+    FI: Fn(DocumentId) -> bool,
+{
+    // We delegate the filter work to the distinct query builder,
+    // specifying a distinct rule that has no effect.
+    if filter.is_some() {
+        let distinct = |_| None;
+        let distinct_size = 1;
+        return bucket_sort_with_distinct(
+            reader,
+            query,
+            range,
+            filter,
+            distinct,
+            distinct_size,
+            criteria,
+            searchable_attrs,
+            main_store,
+            postings_lists_store,
+            documents_fields_counts_store,
+            synonyms_store,
+            prefix_documents_cache_store,
+            prefix_postings_lists_cache_store,
+        );
+    }
+
+    let words_set = match unsafe { main_store.static_words_fst(reader)? } {
+        Some(words) => words,
+        None => return Ok((Vec::new(), 0)),
+    };
+
+    let stop_words = main_store.stop_words_fst(reader)?.unwrap_or_default();
+
+    let context = QTContext {
+        words_set,
+        stop_words,
+        synonyms: synonyms_store,
+        postings_lists: postings_lists_store,
+        prefix_postings_lists: prefix_postings_lists_cache_store,
+    };
+
+    let (operation, mapping) = create_query_tree(reader, &context, query)?;
+    debug!("operation:\n{:?}", operation);
+    debug!("mapping:\n{:?}", mapping);
+
+    fn recurs_operation<'o>(map: &mut HashMap<QueryId, &'o QueryKind>, operation: &'o Operation) {
+        match operation {
+            Operation::And(ops) => ops.iter().for_each(|op| recurs_operation(map, op)),
+            Operation::Or(ops) => ops.iter().for_each(|op| recurs_operation(map, op)),
+            Operation::Query(query) => { map.insert(query.id, &query.kind); },
+        }
+    }
+
+    let mut queries_kinds = HashMap::new();
+    recurs_operation(&mut queries_kinds, &operation);
+
+    let QueryResult { docids, queries } = traverse_query_tree(reader, &context, &operation)?;
+    debug!("found {} documents", docids.len());
+    debug!("number of postings {:?}", queries.len());
+
+    let before = Instant::now();
+    mk_arena!(arena);
+    let mut bare_matches = cleanup_bare_matches(&mut arena, &docids, queries);
+    debug!("matches cleaned in {:.02?}", before.elapsed());
+
+    let before_bucket_sort = Instant::now();
+
+    let before_raw_documents_building = Instant::now();
+    let mut raw_documents = Vec::new();
+    for bare_matches in bare_matches.linear_group_by_key_mut(|sm| sm.document_id) {
+        let raw_document = RawDocument::new(bare_matches, &mut arena, searchable_attrs.as_ref());
+        raw_documents.push(raw_document);
+    }
+    debug!("creating {} candidates documents took {:.02?}",
+        raw_documents.len(),
+        before_raw_documents_building.elapsed(),
+    );
+
+    let before_criterion_loop = Instant::now();
+    let proximity_count = AtomicUsize::new(0);
+
+    let mut groups = vec![raw_documents.as_mut_slice()];
+
+    'criteria: for criterion in criteria.as_ref() {
+        let tmp_groups = mem::replace(&mut groups, Vec::new());
+        let mut documents_seen = 0;
+
+        for mut group in tmp_groups {
+            let before_criterion_preparation = Instant::now();
+
+            let ctx = ContextMut {
+                reader,
+                postings_lists: &mut arena,
+                query_mapping: &mapping,
+                documents_fields_counts_store,
+            };
+
+            criterion.prepare(ctx, &mut group)?;
+            debug!("{:?} preparation took {:.02?}", criterion.name(), before_criterion_preparation.elapsed());
+
+            let ctx = Context {
+                postings_lists: &arena,
+                query_mapping: &mapping,
+            };
+
+            let before_criterion_sort = Instant::now();
+            group.sort_unstable_by(|a, b| criterion.evaluate(&ctx, a, b));
+            debug!("{:?} evaluation took {:.02?}", criterion.name(), before_criterion_sort.elapsed());
+
+            for group in group.binary_group_by_mut(|a, b| criterion.eq(&ctx, a, b)) {
+                debug!("{:?} produced a group of size {}", criterion.name(), group.len());
+
+                documents_seen += group.len();
+                groups.push(group);
+
+                // we have sort enough documents if the last document sorted is after
+                // the end of the requested range, we can continue to the next criterion
+                if documents_seen >= range.end {
+                    continue 'criteria;
+                }
+            }
+        }
+    }
+
+    debug!("criterion loop took {:.02?}", before_criterion_loop.elapsed());
+    debug!("proximity evaluation called {} times", proximity_count.load(Ordering::Relaxed));
+
+    let schema = main_store.schema(reader)?.ok_or(Error::SchemaMissing)?;
+    let iter = raw_documents.into_iter().skip(range.start).take(range.len());
+    let iter = iter.map(|rd| Document::from_raw(rd, &queries_kinds, &arena, searchable_attrs.as_ref(), &schema));
+    let documents = iter.collect();
+
+    debug!("bucket sort took {:.02?}", before_bucket_sort.elapsed());
+
+    Ok((documents, docids.len()))
+}
+
+pub fn bucket_sort_with_distinct<'c, FI, FD>(
+    reader: &heed::RoTxn<MainT>,
+    query: &str,
+    range: Range<usize>,
+    filter: Option<FI>,
+    distinct: FD,
+    distinct_size: usize,
+    criteria: Criteria<'c>,
+    searchable_attrs: Option<ReorderedAttrs>,
+    main_store: store::Main,
+    postings_lists_store: store::PostingsLists,
+    documents_fields_counts_store: store::DocumentsFieldsCounts,
+    synonyms_store: store::Synonyms,
+    _prefix_documents_cache_store: store::PrefixDocumentsCache,
+    prefix_postings_lists_cache_store: store::PrefixPostingsListsCache,
+) -> MResult<(Vec<Document>, usize)>
+where
+    FI: Fn(DocumentId) -> bool,
+    FD: Fn(DocumentId) -> Option<u64>,
+{
+    let words_set = match unsafe { main_store.static_words_fst(reader)? } {
+        Some(words) => words,
+        None => return Ok((Vec::new(), 0)),
+    };
+
+    let stop_words = main_store.stop_words_fst(reader)?.unwrap_or_default();
+
+    let context = QTContext {
+        words_set,
+        stop_words,
+        synonyms: synonyms_store,
+        postings_lists: postings_lists_store,
+        prefix_postings_lists: prefix_postings_lists_cache_store,
+    };
+
+    let (operation, mapping) = create_query_tree(reader, &context, query)?;
+    debug!("operation:\n{:?}", operation);
+    debug!("mapping:\n{:?}", mapping);
+
+    fn recurs_operation<'o>(map: &mut HashMap<QueryId, &'o QueryKind>, operation: &'o Operation) {
+        match operation {
+            Operation::And(ops) => ops.iter().for_each(|op| recurs_operation(map, op)),
+            Operation::Or(ops) => ops.iter().for_each(|op| recurs_operation(map, op)),
+            Operation::Query(query) => { map.insert(query.id, &query.kind); },
+        }
+    }
+
+    let mut queries_kinds = HashMap::new();
+    recurs_operation(&mut queries_kinds, &operation);
+
+    let QueryResult { docids, queries } = traverse_query_tree(reader, &context, &operation)?;
+    debug!("found {} documents", docids.len());
+    debug!("number of postings {:?}", queries.len());
+
+    let before = Instant::now();
+    mk_arena!(arena);
+    let mut bare_matches = cleanup_bare_matches(&mut arena, &docids, queries);
+    debug!("matches cleaned in {:.02?}", before.elapsed());
+
+    let before_raw_documents_building = Instant::now();
+    let mut raw_documents = Vec::new();
+    for bare_matches in bare_matches.linear_group_by_key_mut(|sm| sm.document_id) {
+        let raw_document = RawDocument::new(bare_matches, &mut arena, searchable_attrs.as_ref());
+        raw_documents.push(raw_document);
+    }
+    debug!("creating {} candidates documents took {:.02?}",
+        raw_documents.len(),
+        before_raw_documents_building.elapsed(),
+    );
+
+    let mut groups = vec![raw_documents.as_mut_slice()];
+    let mut key_cache = HashMap::new();
+
+    let mut filter_map = HashMap::new();
+    // these two variables informs on the current distinct map and
+    // on the raw offset of the start of the group where the
+    // range.start bound is located according to the distinct function
+    let mut distinct_map = DistinctMap::new(distinct_size);
+    let mut distinct_raw_offset = 0;
+
+    'criteria: for criterion in criteria.as_ref() {
+        let tmp_groups = mem::replace(&mut groups, Vec::new());
+        let mut buf_distinct = BufferedDistinctMap::new(&mut distinct_map);
+        let mut documents_seen = 0;
+
+        for mut group in tmp_groups {
+            // if this group does not overlap with the requested range,
+            // push it without sorting and splitting it
+            if documents_seen + group.len() < distinct_raw_offset {
+                documents_seen += group.len();
+                groups.push(group);
+                continue;
+            }
+
+            let ctx = ContextMut {
+                reader,
+                postings_lists: &mut arena,
+                query_mapping: &mapping,
+                documents_fields_counts_store,
+            };
+
+            let before_criterion_preparation = Instant::now();
+            criterion.prepare(ctx, &mut group)?;
+            debug!("{:?} preparation took {:.02?}", criterion.name(), before_criterion_preparation.elapsed());
+
+            let ctx = Context {
+                postings_lists: &arena,
+                query_mapping: &mapping,
+            };
+
+            let before_criterion_sort = Instant::now();
+            group.sort_unstable_by(|a, b| criterion.evaluate(&ctx, a, b));
+            debug!("{:?} evaluation took {:.02?}", criterion.name(), before_criterion_sort.elapsed());
+
+            for group in group.binary_group_by_mut(|a, b| criterion.eq(&ctx, a, b)) {
+                // we must compute the real distinguished len of this sub-group
+                for document in group.iter() {
+                    let filter_accepted = match &filter {
+                        Some(filter) => {
+                            let entry = filter_map.entry(document.id);
+                            *entry.or_insert_with(|| (filter)(document.id))
+                        }
+                        None => true,
+                    };
+
+                    if filter_accepted {
+                        let entry = key_cache.entry(document.id);
+                        let key = entry.or_insert_with(|| (distinct)(document.id).map(Rc::new));
+
+                        match key.clone() {
+                            Some(key) => buf_distinct.register(key),
+                            None => buf_distinct.register_without_key(),
+                        };
+                    }
+
+                    // the requested range end is reached: stop computing distinct
+                    if buf_distinct.len() >= range.end {
+                        break;
+                    }
+                }
+
+                documents_seen += group.len();
+                groups.push(group);
+
+                // if this sub-group does not overlap with the requested range
+                // we must update the distinct map and its start index
+                if buf_distinct.len() < range.start {
+                    buf_distinct.transfert_to_internal();
+                    distinct_raw_offset = documents_seen;
+                }
+
+                // we have sort enough documents if the last document sorted is after
+                // the end of the requested range, we can continue to the next criterion
+                if buf_distinct.len() >= range.end {
+                    continue 'criteria;
+                }
+            }
+        }
+    }
+
+    // once we classified the documents related to the current
+    // automatons we save that as the next valid result
+    let mut seen = BufferedDistinctMap::new(&mut distinct_map);
+    let schema = main_store.schema(reader)?.ok_or(Error::SchemaMissing)?;
+
+    let mut documents = Vec::with_capacity(range.len());
+    for raw_document in raw_documents.into_iter().skip(distinct_raw_offset) {
+        let filter_accepted = match &filter {
+            Some(_) => filter_map.remove(&raw_document.id).unwrap(),
+            None => true,
+        };
+
+        if filter_accepted {
+            let key = key_cache.remove(&raw_document.id).unwrap();
+            let distinct_accepted = match key {
+                Some(key) => seen.register(key),
+                None => seen.register_without_key(),
+            };
+
+            if distinct_accepted && seen.len() > range.start {
+                documents.push(Document::from_raw(raw_document, &queries_kinds, &arena, searchable_attrs.as_ref(), &schema));
+                if documents.len() == range.len() {
+                    break;
+                }
+            }
+        }
+    }
+
+    Ok((documents, docids.len()))
+}
+
+fn cleanup_bare_matches<'tag, 'txn>(
+    arena: &mut SmallArena<'tag, PostingsListView<'txn>>,
+    docids: &Set<DocumentId>,
+    queries: HashMap<PostingsKey, Cow<'txn, Set<DocIndex>>>,
+) -> Vec<BareMatch<'tag>>
+{
+    let docidslen = docids.len() as f32;
+    let mut bare_matches = Vec::new();
+
+    for (PostingsKey { query, input, distance, is_exact }, matches) in queries {
+        let postings_list_view = PostingsListView::original(Rc::from(input), Rc::new(matches));
+        let pllen = postings_list_view.len() as f32;
+
+        if docidslen / pllen >= 0.8 {
+            let mut offset = 0;
+            for matches in postings_list_view.linear_group_by_key(|m| m.document_id) {
+                let document_id = matches[0].document_id;
+                if docids.contains(&document_id) {
+                    let range = postings_list_view.range(offset, matches.len());
+                    let posting_list_index = arena.add(range);
+
+                    let bare_match = BareMatch {
+                        document_id,
+                        query_index: query.id,
+                        distance,
+                        is_exact,
+                        postings_list: posting_list_index,
+                    };
+
+                    bare_matches.push(bare_match);
+                }
+
+                offset += matches.len();
+            }
+
+        } else {
+            let mut offset = 0;
+            for id in docids.as_slice() {
+                let di = DocIndex { document_id: *id, ..DocIndex::default() };
+                let pos = exponential_search(&postings_list_view[offset..], &di).unwrap_or_else(|x| x);
+
+                offset += pos;
+
+                let group = postings_list_view[offset..]
+                    .linear_group_by_key(|m| m.document_id)
+                    .next()
+                    .filter(|matches| matches[0].document_id == *id);
+
+                if let Some(matches) = group {
+                    let range = postings_list_view.range(offset, matches.len());
+                    let posting_list_index = arena.add(range);
+
+                    let bare_match = BareMatch {
+                        document_id: *id,
+                        query_index: query.id,
+                        distance,
+                        is_exact,
+                        postings_list: posting_list_index,
+                    };
+
+                    bare_matches.push(bare_match);
+                }
+            }
+        }
+    }
+
+    let before_raw_documents_presort = Instant::now();
+    bare_matches.sort_unstable_by_key(|sm| sm.document_id);
+    debug!("sort by documents ids took {:.02?}", before_raw_documents_presort.elapsed());
+
+    bare_matches
+}
+
+pub struct BareMatch<'tag> {
+    pub document_id: DocumentId,
+    pub query_index: usize,
+    pub distance: u8,
+    pub is_exact: bool,
+    pub postings_list: Idx32<'tag>,
+}
+
+impl fmt::Debug for BareMatch<'_> {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.debug_struct("BareMatch")
+            .field("document_id", &self.document_id)
+            .field("query_index", &self.query_index)
+            .field("distance", &self.distance)
+            .field("is_exact", &self.is_exact)
+            .finish()
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
+pub struct SimpleMatch {
+    pub query_index: usize,
+    pub distance: u8,
+    pub attribute: u16,
+    pub word_index: u16,
+    pub is_exact: bool,
+}
+
+#[derive(Clone)]
+pub enum PostingsListView<'txn> {
+    Original {
+        input: Rc<[u8]>,
+        postings_list: Rc<Cow<'txn, Set<DocIndex>>>,
+        offset: usize,
+        len: usize,
+    },
+    Rewritten {
+        input: Rc<[u8]>,
+        postings_list: SetBuf<DocIndex>,
+    },
+}
+
+impl fmt::Debug for PostingsListView<'_> {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.debug_struct("PostingsListView")
+            .field("input", &std::str::from_utf8(&self.input()).unwrap())
+            .field("postings_list", &self.as_ref())
+            .finish()
+    }
+}
+
+impl<'txn> PostingsListView<'txn> {
+    pub fn original(input: Rc<[u8]>, postings_list: Rc<Cow<'txn, Set<DocIndex>>>) -> PostingsListView<'txn> {
+        let len = postings_list.len();
+        PostingsListView::Original { input, postings_list, offset: 0, len }
+    }
+
+    pub fn rewritten(input: Rc<[u8]>, postings_list: SetBuf<DocIndex>) -> PostingsListView<'txn> {
+        PostingsListView::Rewritten { input, postings_list }
+    }
+
+    pub fn rewrite_with(&mut self, postings_list: SetBuf<DocIndex>) {
+        let input = match self {
+            PostingsListView::Original { input, .. } => input.clone(),
+            PostingsListView::Rewritten { input, .. } => input.clone(),
+        };
+        *self = PostingsListView::rewritten(input, postings_list);
+    }
+
+    pub fn len(&self) -> usize {
+        match self {
+            PostingsListView::Original { len, .. } => *len,
+            PostingsListView::Rewritten { postings_list, .. } => postings_list.len(),
+        }
+    }
+
+    pub fn input(&self) -> &[u8] {
+        match self {
+            PostingsListView::Original { ref input, .. } => input,
+            PostingsListView::Rewritten { ref input, .. } => input,
+        }
+    }
+
+    pub fn range(&self, range_offset: usize, range_len: usize) -> PostingsListView<'txn> {
+        match self {
+            PostingsListView::Original { input, postings_list, offset, len } => {
+                assert!(range_offset + range_len <= *len);
+                PostingsListView::Original {
+                    input: input.clone(),
+                    postings_list: postings_list.clone(),
+                    offset: offset + range_offset,
+                    len: range_len,
+                }
+            },
+            PostingsListView::Rewritten { .. } => {
+                panic!("Cannot create a range on a rewritten postings list view");
+            }
+        }
+    }
+}
+
+impl AsRef<Set<DocIndex>> for PostingsListView<'_> {
+    fn as_ref(&self) -> &Set<DocIndex> {
+        self
+    }
+}
+
+impl Deref for PostingsListView<'_> {
+    type Target = Set<DocIndex>;
+
+    fn deref(&self) -> &Set<DocIndex> {
+        match *self {
+            PostingsListView::Original { ref postings_list, offset, len, .. } => {
+                Set::new_unchecked(&postings_list[offset..offset + len])
+            },
+            PostingsListView::Rewritten { ref postings_list, .. } => postings_list,
+        }
+    }
+}

meilisearch-core/src/criterion/attribute.rs

@@ -0,0 +1,37 @@
+use std::cmp::Ordering;
+use slice_group_by::GroupBy;
+use crate::{RawDocument, MResult};
+use crate::bucket_sort::SimpleMatch;
+use super::{Criterion, Context, ContextMut, prepare_bare_matches};
+
+pub struct Attribute;
+
+impl Criterion for Attribute {
+    fn name(&self) -> &str { "attribute" }
+
+    fn prepare<'h, 'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: ContextMut<'h, 'p, 'tag, 'txn, 'q>,
+        documents: &mut [RawDocument<'r, 'tag>],
+    ) -> MResult<()>
+    {
+        prepare_bare_matches(documents, ctx.postings_lists, ctx.query_mapping);
+        Ok(())
+    }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        #[inline]
+        fn sum_of_attribute(matches: &[SimpleMatch]) -> usize {
+            let mut sum_of_attribute = 0;
+            for group in matches.linear_group_by_key(|bm| bm.query_index) {
+                sum_of_attribute += group[0].attribute as usize;
+            }
+            sum_of_attribute
+        }
+
+        let lhs = sum_of_attribute(&lhs.processed_matches);
+        let rhs = sum_of_attribute(&rhs.processed_matches);
+
+        lhs.cmp(&rhs)
+    }
+}

meilisearch-core/src/criterion/document_id.rs

@@ -0,0 +1,16 @@
+use std::cmp::Ordering;
+use crate::RawDocument;
+use super::{Criterion, Context};
+
+pub struct DocumentId;
+
+impl Criterion for DocumentId {
+    fn name(&self) -> &str { "stable document id" }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        let lhs = &lhs.id;
+        let rhs = &rhs.id;
+
+        lhs.cmp(rhs)
+    }
+}

meilisearch-core/src/criterion/exactness.rs

@@ -0,0 +1,78 @@
+use std::cmp::{Ordering, Reverse};
+use std::collections::hash_map::{HashMap, Entry};
+use meilisearch_schema::IndexedPos;
+use slice_group_by::GroupBy;
+use crate::{RawDocument, MResult};
+use crate::bucket_sort::BareMatch;
+use super::{Criterion, Context, ContextMut};
+
+pub struct Exactness;
+
+impl Criterion for Exactness {
+    fn name(&self) -> &str { "exactness" }
+
+    fn prepare<'h, 'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: ContextMut<'h, 'p, 'tag, 'txn, 'q>,
+        documents: &mut [RawDocument<'r, 'tag>],
+    ) -> MResult<()>
+    {
+        let store = ctx.documents_fields_counts_store;
+        let reader = ctx.reader;
+
+        'documents: for doc in documents {
+            doc.bare_matches.sort_unstable_by_key(|bm| (bm.query_index, Reverse(bm.is_exact)));
+
+            // mark the document if we find a "one word field" that matches
+            let mut fields_counts = HashMap::new();
+            for group in doc.bare_matches.linear_group_by_key(|bm| bm.query_index) {
+                for group in group.linear_group_by_key(|bm| bm.is_exact) {
+                    if !group[0].is_exact { break }
+
+                    for bm in group {
+                        for di in ctx.postings_lists[bm.postings_list].as_ref() {
+
+                            let attr = IndexedPos(di.attribute);
+                            let count = match fields_counts.entry(attr) {
+                                Entry::Occupied(entry) => *entry.get(),
+                                Entry::Vacant(entry) => {
+                                    let count = store.document_field_count(reader, doc.id, attr)?;
+                                    *entry.insert(count)
+                                },
+                            };
+
+                            if count == Some(1) {
+                                doc.contains_one_word_field = true;
+                                continue 'documents
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        #[inline]
+        fn sum_exact_query_words(matches: &[BareMatch]) -> usize {
+            let mut sum_exact_query_words = 0;
+
+            for group in matches.linear_group_by_key(|bm| bm.query_index) {
+                sum_exact_query_words += group[0].is_exact as usize;
+            }
+
+            sum_exact_query_words
+        }
+
+        // does it contains a "one word field"
+        lhs.contains_one_word_field.cmp(&rhs.contains_one_word_field).reverse()
+        // if not, with document contains the more exact words
+        .then_with(|| {
+            let lhs = sum_exact_query_words(&lhs.bare_matches);
+            let rhs = sum_exact_query_words(&rhs.bare_matches);
+            lhs.cmp(&rhs).reverse()
+        })
+    }
+}

meilisearch-core/src/criterion/mod.rs

@@ -0,0 +1,291 @@
+use std::cmp::{self, Ordering};
+use std::collections::HashMap;
+use std::ops::Range;
+
+use compact_arena::SmallArena;
+use sdset::SetBuf;
+use slice_group_by::GroupBy;
+
+use crate::bucket_sort::{SimpleMatch, PostingsListView};
+use crate::database::MainT;
+use crate::query_tree::QueryId;
+use crate::{store, RawDocument, MResult};
+
+mod typo;
+mod words;
+mod proximity;
+mod attribute;
+mod words_position;
+mod exactness;
+mod document_id;
+mod sort_by_attr;
+
+pub use self::typo::Typo;
+pub use self::words::Words;
+pub use self::proximity::Proximity;
+pub use self::attribute::Attribute;
+pub use self::words_position::WordsPosition;
+pub use self::exactness::Exactness;
+pub use self::document_id::DocumentId;
+pub use self::sort_by_attr::SortByAttr;
+
+pub trait Criterion {
+    fn name(&self) -> &str;
+
+    fn prepare<'h, 'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        _ctx: ContextMut<'h, 'p, 'tag, 'txn, 'q>,
+        _documents: &mut [RawDocument<'r, 'tag>],
+    ) -> MResult<()>
+    {
+        Ok(())
+    }
+
+    fn evaluate<'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: &Context<'p, 'tag, 'txn, 'q>,
+        lhs: &RawDocument<'r, 'tag>,
+        rhs: &RawDocument<'r, 'tag>,
+    ) -> Ordering;
+
+    #[inline]
+    fn eq<'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: &Context<'p, 'tag, 'txn, 'q>,
+        lhs: &RawDocument<'r, 'tag>,
+        rhs: &RawDocument<'r, 'tag>,
+    ) -> bool
+    {
+        self.evaluate(ctx, lhs, rhs) == Ordering::Equal
+    }
+}
+
+pub struct ContextMut<'h, 'p, 'tag, 'txn, 'q> {
+    pub reader: &'h heed::RoTxn<MainT>,
+    pub postings_lists: &'p mut SmallArena<'tag, PostingsListView<'txn>>,
+    pub query_mapping: &'q HashMap<QueryId, Range<usize>>,
+    pub documents_fields_counts_store: store::DocumentsFieldsCounts,
+}
+
+pub struct Context<'p, 'tag, 'txn, 'q> {
+    pub postings_lists: &'p SmallArena<'tag, PostingsListView<'txn>>,
+    pub query_mapping: &'q HashMap<QueryId, Range<usize>>,
+}
+
+#[derive(Default)]
+pub struct CriteriaBuilder<'a> {
+    inner: Vec<Box<dyn Criterion + 'a>>,
+}
+
+impl<'a> CriteriaBuilder<'a> {
+    pub fn new() -> CriteriaBuilder<'a> {
+        CriteriaBuilder { inner: Vec::new() }
+    }
+
+    pub fn with_capacity(capacity: usize) -> CriteriaBuilder<'a> {
+        CriteriaBuilder {
+            inner: Vec::with_capacity(capacity),
+        }
+    }
+
+    pub fn reserve(&mut self, additional: usize) {
+        self.inner.reserve(additional)
+    }
+
+    pub fn add<C: 'a>(mut self, criterion: C) -> CriteriaBuilder<'a>
+    where
+        C: Criterion,
+    {
+        self.push(criterion);
+        self
+    }
+
+    pub fn push<C: 'a>(&mut self, criterion: C)
+    where
+        C: Criterion,
+    {
+        self.inner.push(Box::new(criterion));
+    }
+
+    pub fn build(self) -> Criteria<'a> {
+        Criteria { inner: self.inner }
+    }
+}
+
+pub struct Criteria<'a> {
+    inner: Vec<Box<dyn Criterion + 'a>>,
+}
+
+impl<'a> Default for Criteria<'a> {
+    fn default() -> Self {
+        CriteriaBuilder::with_capacity(7)
+            .add(Typo)
+            .add(Words)
+            .add(Proximity)
+            .add(Attribute)
+            .add(WordsPosition)
+            .add(Exactness)
+            .add(DocumentId)
+            .build()
+    }
+}
+
+impl<'a> AsRef<[Box<dyn Criterion + 'a>]> for Criteria<'a> {
+    fn as_ref(&self) -> &[Box<dyn Criterion + 'a>] {
+        &self.inner
+    }
+}
+
+fn prepare_query_distances<'a, 'tag, 'txn>(
+    documents: &mut [RawDocument<'a, 'tag>],
+    query_mapping: &HashMap<QueryId, Range<usize>>,
+    postings_lists: &SmallArena<'tag, PostingsListView<'txn>>,
+) {
+    for document in documents {
+        if !document.processed_distances.is_empty() { continue }
+
+        let mut processed = Vec::new();
+        for m in document.bare_matches.iter() {
+            if postings_lists[m.postings_list].is_empty() { continue }
+
+            let range = query_mapping[&(m.query_index as usize)].clone();
+            let new_len = cmp::max(range.end as usize, processed.len());
+            processed.resize(new_len, None);
+
+            for index in range {
+                let index = index as usize;
+                processed[index] = match processed[index] {
+                    Some(distance) if distance > m.distance => Some(m.distance),
+                    Some(distance) => Some(distance),
+                    None => Some(m.distance),
+                };
+            }
+        }
+
+        document.processed_distances = processed;
+    }
+}
+
+fn prepare_bare_matches<'a, 'tag, 'txn>(
+    documents: &mut [RawDocument<'a, 'tag>],
+    postings_lists: &mut SmallArena<'tag, PostingsListView<'txn>>,
+    query_mapping: &HashMap<QueryId, Range<usize>>,
+) {
+    for document in documents {
+        if !document.processed_matches.is_empty() { continue }
+
+        let mut processed = Vec::new();
+        for m in document.bare_matches.iter() {
+            let postings_list = &postings_lists[m.postings_list];
+            processed.reserve(postings_list.len());
+            for di in postings_list.as_ref() {
+                let simple_match = SimpleMatch {
+                    query_index: m.query_index,
+                    distance: m.distance,
+                    attribute: di.attribute,
+                    word_index: di.word_index,
+                    is_exact: m.is_exact,
+                };
+                processed.push(simple_match);
+            }
+        }
+
+        let processed = multiword_rewrite_matches(&mut processed, query_mapping);
+        document.processed_matches = processed.into_vec();
+    }
+}
+
+fn multiword_rewrite_matches(
+    matches: &mut [SimpleMatch],
+    query_mapping: &HashMap<QueryId, Range<usize>>,
+) -> SetBuf<SimpleMatch>
+{
+    matches.sort_unstable_by_key(|m| (m.attribute, m.word_index));
+
+    let mut padded_matches = Vec::with_capacity(matches.len());
+
+    // let before_padding = Instant::now();
+    // for each attribute of each document
+    for same_document_attribute in matches.linear_group_by_key(|m| m.attribute) {
+        // padding will only be applied
+        // to word indices in the same attribute
+        let mut padding = 0;
+        let mut iter = same_document_attribute.linear_group_by_key(|m| m.word_index);
+
+        // for each match at the same position
+        // in this document attribute
+        while let Some(same_word_index) = iter.next() {
+            // find the biggest padding
+            let mut biggest = 0;
+            for match_ in same_word_index {
+                let mut replacement = query_mapping[&(match_.query_index as usize)].clone();
+                let replacement_len = replacement.len();
+                let nexts = iter.remainder().linear_group_by_key(|m| m.word_index);
+
+                if let Some(query_index) = replacement.next() {
+                    let word_index = match_.word_index + padding as u16;
+                    let match_ = SimpleMatch { query_index, word_index, ..*match_ };
+                    padded_matches.push(match_);
+                }
+
+                let mut found = false;
+
+                // look ahead and if there already is a match
+                // corresponding to this padding word, abort the padding
+                'padding: for (x, next_group) in nexts.enumerate() {
+                    for (i, query_index) in replacement.clone().enumerate().skip(x) {
+                        let word_index = match_.word_index + padding as u16 + (i + 1) as u16;
+                        let padmatch = SimpleMatch { query_index, word_index, ..*match_ };
+
+                        for nmatch_ in next_group {
+                            let mut rep = query_mapping[&(nmatch_.query_index as usize)].clone();
+                            let query_index = rep.next().unwrap();
+                            if query_index == padmatch.query_index {
+                                if !found {
+                                    // if we find a corresponding padding for the
+                                    // first time we must push preceding paddings
+                                    for (i, query_index) in replacement.clone().enumerate().take(i) {
+                                        let word_index = match_.word_index + padding as u16 + (i + 1) as u16;
+                                        let match_ = SimpleMatch { query_index, word_index, ..*match_ };
+                                        padded_matches.push(match_);
+                                        biggest = biggest.max(i + 1);
+                                    }
+                                }
+
+                                padded_matches.push(padmatch);
+                                found = true;
+                                continue 'padding;
+                            }
+                        }
+                    }
+
+                    // if we do not find a corresponding padding in the
+                    // next groups so stop here and pad what was found
+                    break;
+                }
+
+                if !found {
+                    // if no padding was found in the following matches
+                    // we must insert the entire padding
+                    for (i, query_index) in replacement.enumerate() {
+                        let word_index = match_.word_index + padding as u16 + (i + 1) as u16;
+                        let match_ = SimpleMatch { query_index, word_index, ..*match_ };
+                        padded_matches.push(match_);
+                    }
+
+                    biggest = biggest.max(replacement_len - 1);
+                }
+            }
+
+            padding += biggest;
+        }
+    }
+
+    // debug!("padding matches took {:.02?}", before_padding.elapsed());
+
+    // With this check we can see that the loop above takes something
+    // like 43% of the search time even when no rewrite is needed.
+    // assert_eq!(before_matches, padded_matches);
+
+    SetBuf::from_dirty(padded_matches)
+}

meilisearch-core/src/criterion/proximity.rs

@@ -0,0 +1,68 @@
+use std::cmp::{self, Ordering};
+use slice_group_by::GroupBy;
+use crate::bucket_sort::{SimpleMatch};
+use crate::{RawDocument, MResult};
+use super::{Criterion, Context, ContextMut, prepare_bare_matches};
+
+const MAX_DISTANCE: u16 = 8;
+
+pub struct Proximity;
+
+impl Criterion for Proximity {
+    fn name(&self) -> &str { "proximity" }
+
+    fn prepare<'h, 'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: ContextMut<'h, 'p, 'tag, 'txn, 'q>,
+        documents: &mut [RawDocument<'r, 'tag>],
+    ) -> MResult<()>
+    {
+        prepare_bare_matches(documents, ctx.postings_lists, ctx.query_mapping);
+        Ok(())
+    }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        fn index_proximity(lhs: u16, rhs: u16) -> u16 {
+            if lhs < rhs {
+                cmp::min(rhs - lhs, MAX_DISTANCE)
+            } else {
+                cmp::min(lhs - rhs, MAX_DISTANCE) + 1
+            }
+        }
+
+        fn attribute_proximity(lhs: SimpleMatch, rhs: SimpleMatch) -> u16 {
+            if lhs.attribute != rhs.attribute { MAX_DISTANCE }
+            else { index_proximity(lhs.word_index, rhs.word_index) }
+        }
+
+        fn min_proximity(lhs: &[SimpleMatch], rhs: &[SimpleMatch]) -> u16 {
+            let mut min_prox = u16::max_value();
+            for a in lhs {
+                for b in rhs {
+                    let prox = attribute_proximity(*a, *b);
+                    min_prox = cmp::min(min_prox, prox);
+                }
+            }
+            min_prox
+        }
+
+        fn matches_proximity(matches: &[SimpleMatch],) -> u16 {
+            let mut proximity = 0;
+            let mut iter = matches.linear_group_by_key(|m| m.query_index);
+
+            // iterate over groups by windows of size 2
+            let mut last = iter.next();
+            while let (Some(lhs), Some(rhs)) = (last, iter.next()) {
+                proximity += min_proximity(lhs, rhs);
+                last = Some(rhs);
+            }
+
+            proximity
+        }
+
+        let lhs = matches_proximity(&lhs.processed_matches);
+        let rhs = matches_proximity(&rhs.processed_matches);
+
+        lhs.cmp(&rhs)
+    }
+}

meilisearch-core/src/criterion/sort_by_attr.rs

@@ -0,0 +1,129 @@
+use std::cmp::Ordering;
+use std::error::Error;
+use std::fmt;
+use meilisearch_schema::{Schema, FieldId};
+use crate::{RankedMap, RawDocument};
+use super::{Criterion, Context};
+
+/// An helper struct that permit to sort documents by
+/// some of their stored attributes.
+///
+/// # Note
+///
+/// If a document cannot be deserialized it will be considered [`None`][].
+///
+/// Deserialized documents are compared like `Some(doc0).cmp(&Some(doc1))`,
+/// so you must check the [`Ord`] of `Option` implementation.
+///
+/// [`None`]: https://doc.rust-lang.org/std/option/enum.Option.html#variant.None
+/// [`Ord`]: https://doc.rust-lang.org/std/option/enum.Option.html#impl-Ord
+///
+/// # Example
+///
+/// ```ignore
+/// use serde_derive::Deserialize;
+/// use meilisearch::rank::criterion::*;
+///
+/// let custom_ranking = SortByAttr::lower_is_better(&ranked_map, &schema, "published_at")?;
+///
+/// let builder = CriteriaBuilder::with_capacity(8)
+///        .add(Typo)
+///        .add(Words)
+///        .add(Proximity)
+///        .add(Attribute)
+///        .add(WordsPosition)
+///        .add(Exactness)
+///        .add(custom_ranking)
+///        .add(DocumentId);
+///
+/// let criterion = builder.build();
+///
+/// ```
+pub struct SortByAttr<'a> {
+    ranked_map: &'a RankedMap,
+    field_id: FieldId,
+    reversed: bool,
+}
+
+impl<'a> SortByAttr<'a> {
+    pub fn lower_is_better(
+        ranked_map: &'a RankedMap,
+        schema: &Schema,
+        attr_name: &str,
+    ) -> Result<SortByAttr<'a>, SortByAttrError> {
+        SortByAttr::new(ranked_map, schema, attr_name, false)
+    }
+
+    pub fn higher_is_better(
+        ranked_map: &'a RankedMap,
+        schema: &Schema,
+        attr_name: &str,
+    ) -> Result<SortByAttr<'a>, SortByAttrError> {
+        SortByAttr::new(ranked_map, schema, attr_name, true)
+    }
+
+    fn new(
+        ranked_map: &'a RankedMap,
+        schema: &Schema,
+        attr_name: &str,
+        reversed: bool,
+    ) -> Result<SortByAttr<'a>, SortByAttrError> {
+        let field_id = match schema.id(attr_name) {
+            Some(field_id) => field_id,
+            None => return Err(SortByAttrError::AttributeNotFound),
+        };
+
+        if !schema.is_ranked(field_id) {
+            return Err(SortByAttrError::AttributeNotRegisteredForRanking);
+        }
+
+        Ok(SortByAttr {
+            ranked_map,
+            field_id,
+            reversed,
+        })
+    }
+}
+
+impl Criterion for SortByAttr<'_> {
+    fn name(&self) -> &str {
+        "sort by attribute"
+    }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        let lhs = self.ranked_map.get(lhs.id, self.field_id);
+        let rhs = self.ranked_map.get(rhs.id, self.field_id);
+
+        match (lhs, rhs) {
+            (Some(lhs), Some(rhs)) => {
+                let order = lhs.cmp(&rhs);
+                if self.reversed {
+                    order.reverse()
+                } else {
+                    order
+                }
+            }
+            (None, Some(_)) => Ordering::Greater,
+            (Some(_), None) => Ordering::Less,
+            (None, None) => Ordering::Equal,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub enum SortByAttrError {
+    AttributeNotFound,
+    AttributeNotRegisteredForRanking,
+}
+
+impl fmt::Display for SortByAttrError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        use SortByAttrError::*;
+        match self {
+            AttributeNotFound => f.write_str("attribute not found in the schema"),
+            AttributeNotRegisteredForRanking => f.write_str("attribute not registered for ranking"),
+        }
+    }
+}
+
+impl Error for SortByAttrError {}

meilisearch-core/src/criterion/typo.rs

@@ -0,0 +1,55 @@
+use std::cmp::Ordering;
+use crate::{RawDocument, MResult};
+use super::{Criterion, Context, ContextMut, prepare_query_distances};
+
+pub struct Typo;
+
+impl Criterion for Typo {
+    fn name(&self) -> &str { "typo" }
+
+    fn prepare<'h, 'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: ContextMut<'h, 'p, 'tag, 'txn, 'q>,
+        documents: &mut [RawDocument<'r, 'tag>],
+    ) -> MResult<()>
+    {
+        prepare_query_distances(documents, ctx.query_mapping, ctx.postings_lists);
+        Ok(())
+    }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        // This function is a wrong logarithmic 10 function.
+        // It is safe to panic on input number higher than 3,
+        // the number of typos is never bigger than that.
+        #[inline]
+        fn custom_log10(n: u8) -> f32 {
+            match n {
+                0 => 0.0,     // log(1)
+                1 => 0.30102, // log(2)
+                2 => 0.47712, // log(3)
+                3 => 0.60205, // log(4)
+                _ => panic!("invalid number"),
+            }
+        }
+
+        #[inline]
+        fn compute_typos(distances: &[Option<u8>]) -> usize {
+            let mut number_words: usize = 0;
+            let mut sum_typos = 0.0;
+
+            for distance in distances {
+                if let Some(distance) = distance {
+                    sum_typos += custom_log10(*distance);
+                    number_words += 1;
+                }
+            }
+
+            (number_words as f32 / (sum_typos + 1.0) * 1000.0) as usize
+        }
+
+        let lhs = compute_typos(&lhs.processed_distances);
+        let rhs = compute_typos(&rhs.processed_distances);
+
+        lhs.cmp(&rhs).reverse()
+    }
+}

meilisearch-core/src/criterion/words.rs

@@ -0,0 +1,31 @@
+use std::cmp::Ordering;
+use crate::{RawDocument, MResult};
+use super::{Criterion, Context, ContextMut, prepare_query_distances};
+
+pub struct Words;
+
+impl Criterion for Words {
+    fn name(&self) -> &str { "words" }
+
+    fn prepare<'h, 'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: ContextMut<'h, 'p, 'tag, 'txn, 'q>,
+        documents: &mut [RawDocument<'r, 'tag>],
+    ) -> MResult<()>
+    {
+        prepare_query_distances(documents, ctx.query_mapping, ctx.postings_lists);
+        Ok(())
+    }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        #[inline]
+        fn number_of_query_words(distances: &[Option<u8>]) -> usize {
+            distances.iter().cloned().filter(Option::is_some).count()
+        }
+
+        let lhs = number_of_query_words(&lhs.processed_distances);
+        let rhs = number_of_query_words(&rhs.processed_distances);
+
+        lhs.cmp(&rhs).reverse()
+    }
+}

meilisearch-core/src/criterion/words_position.rs

@@ -0,0 +1,37 @@
+use std::cmp::Ordering;
+use slice_group_by::GroupBy;
+use crate::bucket_sort::SimpleMatch;
+use crate::{RawDocument, MResult};
+use super::{Criterion, Context, ContextMut, prepare_bare_matches};
+
+pub struct WordsPosition;
+
+impl Criterion for WordsPosition {
+    fn name(&self) -> &str { "words position" }
+
+    fn prepare<'h, 'p, 'tag, 'txn, 'q, 'r>(
+        &self,
+        ctx: ContextMut<'h, 'p, 'tag, 'txn, 'q>,
+        documents: &mut [RawDocument<'r, 'tag>],
+    ) -> MResult<()>
+    {
+        prepare_bare_matches(documents, ctx.postings_lists, ctx.query_mapping);
+        Ok(())
+    }
+
+    fn evaluate(&self, _ctx: &Context, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
+        #[inline]
+        fn sum_words_position(matches: &[SimpleMatch]) -> usize {
+            let mut sum_words_position = 0;
+            for group in matches.linear_group_by_key(|bm| bm.query_index) {
+                sum_words_position += group[0].word_index as usize;
+            }
+            sum_words_position
+        }
+
+        let lhs = sum_words_position(&lhs.processed_matches);
+        let rhs = sum_words_position(&rhs.processed_matches);
+
+        lhs.cmp(&rhs)
+    }
+}

meilisearch-core/src/distinct_map.rs

@@ -1,6 +1,5 @@
-use std::hash::Hash;
-
 use hashbrown::HashMap;
+use std::hash::Hash;
 
 pub struct DistinctMap<K> {
     inner: HashMap<K, usize>,
@@ -12,7 +11,7 @@ impl<K: Hash + Eq> DistinctMap<K> {
     pub fn new(limit: usize) -> Self {
         DistinctMap {
             inner: HashMap::new(),
-            limit: limit,
+            limit,
             len: 0,
         }
     }
@@ -31,7 +30,7 @@ pub struct BufferedDistinctMap<'a, K> {
 impl<'a, K: Hash + Eq> BufferedDistinctMap<'a, K> {
     pub fn new(internal: &'a mut DistinctMap<K>) -> BufferedDistinctMap<'a, K> {
         BufferedDistinctMap {
-            internal: internal,
+            internal,
             inner: HashMap::new(),
             len: 0,
         }

meilisearch-core/src/error.rs

@@ -0,0 +1,158 @@
+use crate::serde::{DeserializerError, SerializerError};
+use serde_json::Error as SerdeJsonError;
+use pest::error::Error as PestError;
+use crate::filters::Rule;
+use std::{error, fmt, io};
+
+pub use bincode::Error as BincodeError;
+pub use fst::Error as FstError;
+pub use heed::Error as HeedError;
+pub use pest::error as pest_error;
+
+pub type MResult<T> = Result<T, Error>;
+
+#[derive(Debug)]
+pub enum Error {
+    Io(io::Error),
+    IndexAlreadyExists,
+    MissingPrimaryKey,
+    SchemaMissing,
+    WordIndexMissing,
+    MissingDocumentId,
+    MaxFieldsLimitExceeded,
+    Schema(meilisearch_schema::Error),
+    Zlmdb(heed::Error),
+    Fst(fst::Error),
+    SerdeJson(SerdeJsonError),
+    Bincode(bincode::Error),
+    Serializer(SerializerError),
+    Deserializer(DeserializerError),
+    UnsupportedOperation(UnsupportedOperation),
+    FilterParseError(PestError<Rule>)
+}
+
+impl From<io::Error> for Error {
+    fn from(error: io::Error) -> Error {
+        Error::Io(error)
+    }
+}
+
+impl From<PestError<Rule>> for Error {
+    fn from(error: PestError<Rule>) -> Error {
+        Error::FilterParseError(error.renamed_rules(|r| {
+            let s  = match r {
+                Rule::or => "OR",
+                Rule::and => "AND",
+                Rule::not => "NOT",
+                Rule::string => "string",
+                Rule::word => "word",
+                Rule::greater => "field > value",
+                Rule::less => "field < value",
+                Rule::eq => "field = value",
+                Rule::leq => "field <= value",
+                Rule::geq => "field >= value",
+                Rule::key => "key",
+                _ => "other",
+            };
+            s.to_string()
+        }))
+    }
+} 
+
+impl From<meilisearch_schema::Error> for Error {
+    fn from(error: meilisearch_schema::Error) -> Error {
+        Error::Schema(error)
+    }
+}
+
+impl From<HeedError> for Error {
+    fn from(error: HeedError) -> Error {
+        Error::Zlmdb(error)
+    }
+}
+
+impl From<FstError> for Error {
+    fn from(error: FstError) -> Error {
+        Error::Fst(error)
+    }
+}
+
+impl From<SerdeJsonError> for Error {
+    fn from(error: SerdeJsonError) -> Error {
+        Error::SerdeJson(error)
+    }
+}
+
+impl From<BincodeError> for Error {
+    fn from(error: BincodeError) -> Error {
+        Error::Bincode(error)
+    }
+}
+
+impl From<SerializerError> for Error {
+    fn from(error: SerializerError) -> Error {
+        Error::Serializer(error)
+    }
+}
+
+impl From<DeserializerError> for Error {
+    fn from(error: DeserializerError) -> Error {
+        Error::Deserializer(error)
+    }
+}
+
+impl From<UnsupportedOperation> for Error {
+    fn from(op: UnsupportedOperation) -> Error {
+        Error::UnsupportedOperation(op)
+    }
+}
+
+impl fmt::Display for Error {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        use self::Error::*;
+        match self {
+            Io(e) => write!(f, "{}", e),
+            IndexAlreadyExists => write!(f, "index already exists"),
+            MissingPrimaryKey => write!(f, "schema cannot be built without a primary key"),
+            SchemaMissing => write!(f, "this index does not have a schema"),
+            WordIndexMissing => write!(f, "this index does not have a word index"),
+            MissingDocumentId => write!(f, "document id is missing"),
+            MaxFieldsLimitExceeded => write!(f, "maximum number of fields in a document exceeded"),
+            Schema(e) => write!(f, "schema error; {}", e),
+            Zlmdb(e) => write!(f, "heed error; {}", e),
+            Fst(e) => write!(f, "fst error; {}", e),
+            SerdeJson(e) => write!(f, "serde json error; {}", e),
+            Bincode(e) => write!(f, "bincode error; {}", e),
+            Serializer(e) => write!(f, "serializer error; {}", e),
+            Deserializer(e) => write!(f, "deserializer error; {}", e),
+            UnsupportedOperation(op) => write!(f, "unsupported operation; {}", op),
+            FilterParseError(e) => write!(f, "error parsing filter; {}", e),
+        }
+    }
+}
+
+impl error::Error for Error {}
+
+#[derive(Debug)]
+pub enum UnsupportedOperation {
+    SchemaAlreadyExists,
+    CannotUpdateSchemaPrimaryKey,
+    CannotReorderSchemaAttribute,
+    CanOnlyIntroduceNewSchemaAttributesAtEnd,
+    CannotRemoveSchemaAttribute,
+}
+
+impl fmt::Display for UnsupportedOperation {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        use self::UnsupportedOperation::*;
+        match self {
+            SchemaAlreadyExists => write!(f, "Cannot update index which already have a schema"),
+            CannotUpdateSchemaPrimaryKey => write!(f, "Cannot update the primary key of a schema"),
+            CannotReorderSchemaAttribute => write!(f, "Cannot reorder the attributes of a schema"),
+            CanOnlyIntroduceNewSchemaAttributesAtEnd => {
+                write!(f, "Can only introduce new attributes at end of a schema")
+            }
+            CannotRemoveSchemaAttribute => write!(f, "Cannot remove attributes from a schema"),
+        }
+    }
+}

meilisearch-core/src/filters/condition.rs

@@ -0,0 +1,277 @@
+use std::str::FromStr;
+use std::cmp::Ordering;
+
+use crate::error::Error;
+use crate::{store::Index, DocumentId, MainT};
+use heed::RoTxn;
+use meilisearch_schema::{FieldId, Schema};
+use pest::error::{Error as PestError, ErrorVariant};
+use pest::iterators::Pair;
+use serde_json::{Value, Number};
+use super::parser::Rule;
+
+#[derive(Debug, PartialEq)]
+enum ConditionType {
+    Greater,
+    Less,
+    Equal,
+    LessEqual,
+    GreaterEqual,
+    NotEqual,
+}
+
+/// We need to infer type when the filter is constructed
+/// and match every possible types it can be parsed into.
+#[derive(Debug)]
+struct ConditionValue<'a> {
+    string: &'a str,
+    boolean: Option<bool>,
+    number: Option<Number>
+}
+
+impl<'a> ConditionValue<'a> {
+    pub fn new(value: &Pair<'a, Rule>) -> Self {
+        let value = match value.as_rule() {
+            Rule::string | Rule::word => {
+                let string =  value.as_str();
+                let boolean = match value.as_str() {
+                    "true" => Some(true),
+                    "false" => Some(false),
+                    _ => None,
+                };
+                let number = Number::from_str(value.as_str()).ok();
+                ConditionValue { string, boolean, number }
+            },
+            _ => unreachable!(),
+        };
+        value
+    }
+
+    pub fn as_str(&self) -> &str {
+        self.string.as_ref()
+    }
+
+    pub fn as_number(&self) -> Option<&Number> {
+        self.number.as_ref()
+    }
+
+    pub fn as_bool(&self) -> Option<bool> {
+        self.boolean
+    }
+}
+
+#[derive(Debug)]
+pub struct Condition<'a> {
+    field: FieldId,
+    condition: ConditionType,
+    value: ConditionValue<'a>
+}
+
+fn get_field_value<'a>(schema: &Schema, pair: Pair<'a, Rule>) -> Result<(FieldId, ConditionValue<'a>), Error> {
+    let mut items = pair.into_inner();
+    // lexing ensures that we at least have a key
+    let key = items.next().unwrap();
+    let field = schema
+        .id(key.as_str())
+        .ok_or::<PestError<Rule>>(PestError::new_from_span(
+                ErrorVariant::CustomError {
+                    message: format!(
+                                 "attribute `{}` not found, available attributes are: {}",
+                                 key.as_str(),
+                                 schema.names().collect::<Vec<_>>().join(", ")
+                             ),
+                },
+                key.as_span()))?;
+    let value = ConditionValue::new(&items.next().unwrap());
+    Ok((field, value))
+}
+
+// undefined behavior with big numbers
+fn compare_numbers(lhs: &Number, rhs: &Number) -> Option<Ordering> {
+    match (lhs.as_i64(), lhs.as_u64(), lhs.as_f64(),
+        rhs.as_i64(), rhs.as_u64(), rhs.as_f64()) {
+    //    i64   u64  f64  i64  u64  f64
+        (Some(lhs), _, _, Some(rhs), _, _) => lhs.partial_cmp(&rhs),
+        (_, Some(lhs), _, _, Some(rhs), _) => lhs.partial_cmp(&rhs),
+        (_, _, Some(lhs), _, _, Some(rhs)) => lhs.partial_cmp(&rhs),
+        (_, _, _, _, _, _) => None,
+    }
+}
+
+impl<'a> Condition<'a> {
+    pub fn less(
+        item: Pair<'a, Rule>,
+        schema: &'a Schema,
+    ) -> Result<Self, Error> {
+        let (field, value) = get_field_value(schema, item)?;
+        let condition = ConditionType::Less;
+        Ok(Self { field, condition, value })
+    }
+
+    pub fn greater(
+        item: Pair<'a, Rule>,
+        schema: &'a Schema,
+    ) -> Result<Self, Error> {
+        let (field, value) = get_field_value(schema, item)?;
+        let condition = ConditionType::Greater;
+        Ok(Self { field, condition, value })
+    }
+
+    pub fn neq(
+        item: Pair<'a, Rule>,
+        schema: &'a Schema,
+    ) -> Result<Self, Error> {
+        let (field, value) = get_field_value(schema, item)?;
+        let condition = ConditionType::NotEqual;
+        Ok(Self { field, condition, value })
+    }
+
+    pub fn geq(
+        item: Pair<'a, Rule>,
+        schema: &'a Schema,
+    ) -> Result<Self, Error> {
+        let (field, value) = get_field_value(schema, item)?;
+        let condition = ConditionType::GreaterEqual;
+        Ok(Self { field, condition, value })
+    }
+
+    pub fn leq(
+        item: Pair<'a, Rule>,
+        schema: &'a Schema,
+    ) -> Result<Self, Error> {
+        let (field, value) = get_field_value(schema, item)?;
+        let condition = ConditionType::LessEqual;
+        Ok(Self { field, condition, value })
+    }
+
+    pub fn eq(
+        item: Pair<'a, Rule>,
+        schema: &'a Schema,
+    ) -> Result<Self, Error> {
+        let (field, value) = get_field_value(schema, item)?;
+        let condition = ConditionType::Equal;
+        Ok(Self { field, condition, value })
+    }
+
+    pub fn test(
+        &self,
+        reader: &RoTxn<MainT>,
+        index: &Index,
+        document_id: DocumentId,
+    ) -> Result<bool, Error> {
+        match index.document_attribute::<Value>(reader, document_id, self.field)? {
+            Some(Value::Array(values)) => Ok(values.iter().any(|v| self.match_value(Some(v)))),
+            other => Ok(self.match_value(other.as_ref())),
+        }
+    }
+
+    fn match_value(&self, value: Option<&Value>) -> bool {
+        match value {
+            Some(Value::String(s)) => {
+                let value = self.value.as_str();
+                match self.condition {
+                    ConditionType::Equal => unicase::eq(value, &s),
+                    ConditionType::NotEqual => !unicase::eq(value, &s),
+                    _ => false
+                }
+            },
+            Some(Value::Number(n)) => { 
+                if let Some(value) = self.value.as_number() {
+                    if let Some(ord) = compare_numbers(&n, value) {
+                        let res =  match self.condition {
+                            ConditionType::Equal => ord == Ordering::Equal,
+                            ConditionType::NotEqual => ord != Ordering::Equal,
+                            ConditionType::GreaterEqual => ord != Ordering::Less,
+                            ConditionType::LessEqual => ord != Ordering::Greater,
+                            ConditionType::Greater => ord == Ordering::Greater,
+                            ConditionType::Less => ord == Ordering::Less,
+                        };
+                        return res
+                    } 
+                } 
+                false
+            },
+            Some(Value::Bool(b)) => {
+                if let Some(value) = self.value.as_bool() {
+                    let res = match self.condition {
+                        ConditionType::Equal => *b == value,
+                        ConditionType::NotEqual => *b != value,
+                        _ => false
+                    };
+                    return res
+                }
+                false
+            },
+            // if field is not supported (or not found), all values are different from it,
+            // so != should always return true in this case.
+            _ => self.condition == ConditionType::NotEqual,
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use serde_json::Number;
+    use std::cmp::Ordering;
+
+    #[test]
+    fn test_number_comp() {
+        // test both u64
+        let n1 = Number::from(1u64);
+        let n2 = Number::from(2u64);
+        assert_eq!(Some(Ordering::Less), compare_numbers(&n1, &n2));
+        assert_eq!(Some(Ordering::Greater), compare_numbers(&n2, &n1));
+        let n1 = Number::from(1u64);
+        let n2 = Number::from(1u64);
+        assert_eq!(Some(Ordering::Equal), compare_numbers(&n1, &n2));
+
+        // test both i64
+        let n1 = Number::from(1i64);
+        let n2 = Number::from(2i64);
+        assert_eq!(Some(Ordering::Less), compare_numbers(&n1, &n2));
+        assert_eq!(Some(Ordering::Greater), compare_numbers(&n2, &n1));
+        let n1 = Number::from(1i64);
+        let n2 = Number::from(1i64);
+        assert_eq!(Some(Ordering::Equal), compare_numbers(&n1, &n2));
+
+        // test both f64
+        let n1 = Number::from_f64(1f64).unwrap();
+        let n2 = Number::from_f64(2f64).unwrap();
+        assert_eq!(Some(Ordering::Less), compare_numbers(&n1, &n2));
+        assert_eq!(Some(Ordering::Greater), compare_numbers(&n2, &n1));
+        let n1 = Number::from_f64(1f64).unwrap();
+        let n2 = Number::from_f64(1f64).unwrap();
+        assert_eq!(Some(Ordering::Equal), compare_numbers(&n1, &n2));
+
+        // test one u64 and one f64
+        let n1 = Number::from_f64(1f64).unwrap();
+        let n2 = Number::from(2u64);
+        assert_eq!(Some(Ordering::Less), compare_numbers(&n1, &n2));
+        assert_eq!(Some(Ordering::Greater), compare_numbers(&n2, &n1));
+
+        // equality
+        let n1 = Number::from_f64(1f64).unwrap();
+        let n2 = Number::from(1u64);
+        assert_eq!(Some(Ordering::Equal), compare_numbers(&n1, &n2));
+        assert_eq!(Some(Ordering::Equal), compare_numbers(&n2, &n1));
+
+        // float is neg
+        let n1 = Number::from_f64(-1f64).unwrap();
+        let n2 = Number::from(1u64);
+        assert_eq!(Some(Ordering::Less), compare_numbers(&n1, &n2));
+        assert_eq!(Some(Ordering::Greater), compare_numbers(&n2, &n1));
+
+        // float is too big
+        let n1 = Number::from_f64(std::f64::MAX).unwrap();
+        let n2 = Number::from(1u64);
+        assert_eq!(Some(Ordering::Greater), compare_numbers(&n1, &n2));
+        assert_eq!(Some(Ordering::Less), compare_numbers(&n2, &n1));
+
+        // misc
+        let n1 = Number::from_f64(std::f64::MAX).unwrap();
+        let n2 = Number::from(std::u64::MAX);
+        assert_eq!(Some(Ordering::Greater), compare_numbers(&n1, &n2));
+        assert_eq!(Some( Ordering::Less ), compare_numbers(&n2, &n1));
+    }
+}

meilisearch-core/src/filters/mod.rs

@@ -0,0 +1,127 @@
+mod parser;
+mod condition;
+
+pub(crate) use parser::Rule;
+
+use std::ops::Not;
+
+use condition::Condition;
+use crate::error::Error;
+use crate::{DocumentId, MainT, store::Index};
+use heed::RoTxn;
+use meilisearch_schema::Schema;
+use parser::{PREC_CLIMBER, FilterParser};
+use pest::iterators::{Pair, Pairs};
+use pest::Parser;
+
+type FilterResult<'a> = Result<Filter<'a>, Error>;
+
+#[derive(Debug)]
+pub enum Filter<'a> {
+    Condition(Condition<'a>),
+    Or(Box<Self>, Box<Self>),
+    And(Box<Self>, Box<Self>),
+    Not(Box<Self>),
+}
+
+impl<'a> Filter<'a> {
+    pub fn parse(expr: &'a str, schema: &'a Schema) -> FilterResult<'a> {
+        let mut lexed = FilterParser::parse(Rule::prgm, expr.as_ref())?;
+        Self::build(lexed.next().unwrap().into_inner(), schema)
+    }
+
+    pub fn test(
+        &self,
+        reader: &RoTxn<MainT>,
+        index: &Index,
+        document_id: DocumentId,
+    ) -> Result<bool, Error> {
+        use Filter::*;
+        match self {
+            Condition(c) => c.test(reader, index, document_id),
+            Or(lhs, rhs) => Ok(
+                lhs.test(reader, index, document_id)? || rhs.test(reader, index, document_id)?
+            ),
+            And(lhs, rhs) => Ok(
+                lhs.test(reader, index, document_id)? && rhs.test(reader, index, document_id)?
+            ),
+            Not(op) => op.test(reader, index, document_id).map(bool::not),
+        }
+    }
+
+    fn build(expression: Pairs<'a, Rule>, schema: &'a Schema) -> FilterResult<'a> {
+        PREC_CLIMBER.climb(
+            expression,
+            |pair: Pair<Rule>| match pair.as_rule() {
+                Rule::eq => Ok(Filter::Condition(Condition::eq(pair, schema)?)),
+                Rule::greater => Ok(Filter::Condition(Condition::greater(pair, schema)?)),
+                Rule::less => Ok(Filter::Condition(Condition::less(pair, schema)?)),
+                Rule::neq => Ok(Filter::Condition(Condition::neq(pair, schema)?)),
+                Rule::geq => Ok(Filter::Condition(Condition::geq(pair, schema)?)),
+                Rule::leq => Ok(Filter::Condition(Condition::leq(pair, schema)?)),
+                Rule::prgm => Self::build(pair.into_inner(), schema),
+                Rule::term => Self::build(pair.into_inner(), schema),
+                Rule::not => Ok(Filter::Not(Box::new(Self::build(
+                    pair.into_inner(),
+                    schema,
+                )?))),
+                _ => unreachable!(),
+            },
+            |lhs: FilterResult, op: Pair<Rule>, rhs: FilterResult| match op.as_rule() {
+                Rule::or => Ok(Filter::Or(Box::new(lhs?), Box::new(rhs?))),
+                Rule::and => Ok(Filter::And(Box::new(lhs?), Box::new(rhs?))),
+                _ => unreachable!(),
+            },
+        )
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn invalid_syntax() {
+        assert!(FilterParser::parse(Rule::prgm, "field : id").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "field=hello hello").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "field=hello OR OR").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "OR field:hello").is_err());
+        assert!(FilterParser::parse(Rule::prgm, r#"field="hello world"#).is_err());
+        assert!(FilterParser::parse(Rule::prgm, r#"field='hello world"#).is_err());
+        assert!(FilterParser::parse(Rule::prgm, "NOT field=").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "N").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "(field=1").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "(field=1))").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "field=1ORfield=2").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "field=1 ( OR field=2)").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "hello world=1").is_err());
+        assert!(FilterParser::parse(Rule::prgm, "").is_err());
+        assert!(FilterParser::parse(Rule::prgm, r#"((((((hello=world)))))"#).is_err());
+    }
+
+    #[test]
+    fn valid_syntax() {
+        assert!(FilterParser::parse(Rule::prgm, "field = id").is_ok());
+        assert!(FilterParser::parse(Rule::prgm, "field=id").is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field >= 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field <= 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field="hello world""#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field='hello world'"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field > 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field < 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field < 10 AND NOT field=5"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field < 10 AND NOT field > 7.5"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field=true OR NOT field=5"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"NOT field=true OR NOT field=5"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field='hello world' OR ( NOT field=true OR NOT field=5 )"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field='hello \'worl\'d' OR ( NOT field=true OR NOT field=5 )"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"field="hello \"worl\"d" OR ( NOT field=true OR NOT field=5 )"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"((((((hello=world))))))"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#""foo bar" > 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#""foo bar" = 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"'foo bar' = 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"'foo bar' <= 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"'foo bar' != 10"#).is_ok());
+        assert!(FilterParser::parse(Rule::prgm, r#"bar != 10"#).is_ok());
+    }
+}

meilisearch-core/src/filters/parser/grammar.pest

@@ -0,0 +1,28 @@
+key = _{quoted | word}
+value = _{quoted | word}
+quoted = _{ (PUSH("'") | PUSH("\"")) ~ string ~ POP  }
+string = {char*}
+word = ${(LETTER | NUMBER | "_" | "-" | ".")+}
+
+char =  _{ !(PEEK | "\\") ~ ANY
+    | "\\" ~ (PEEK | "\\" | "/" | "b" | "f" | "n" | "r" | "t")
+    | "\\" ~ ("u" ~ ASCII_HEX_DIGIT{4})}
+
+condition = _{eq | greater | less | geq | leq | neq}
+geq = {key ~ ">=" ~ value}
+leq = {key ~ "<=" ~ value}
+neq = {key ~ "!=" ~ value}
+eq = {key ~ "=" ~ value}
+greater = {key ~ ">" ~ value}
+less = {key ~ "<" ~ value}
+
+prgm = {SOI ~ expr ~ EOI}
+expr = _{ ( term ~ (operation ~ term)* ) }
+term = { ("(" ~ expr ~ ")") | condition | not }
+operation = _{ and | or }
+	and = {"AND"}
+	or = {"OR"}
+
+not = {"NOT" ~ term}
+
+WHITESPACE = _{ " " }

meilisearch-core/src/filters/parser/mod.rs

@@ -0,0 +1,12 @@
+use once_cell::sync::Lazy;
+use pest::prec_climber::{Operator, Assoc, PrecClimber};
+
+pub static PREC_CLIMBER: Lazy<PrecClimber<Rule>> = Lazy::new(|| {
+    use Assoc::*;
+    use Rule::*;
+    pest::prec_climber::PrecClimber::new(vec![Operator::new(or, Left), Operator::new(and, Left)])
+});
+
+#[derive(Parser)]
+#[grammar = "filters/parser/grammar.pest"]
+pub struct FilterParser;

meilisearch-core/src/levenshtein.rs

@@ -0,0 +1,134 @@
+use std::cmp::min;
+use std::collections::BTreeMap;
+use std::ops::{Index, IndexMut};
+
+// A simple wrapper around vec so we can get contiguous but index it like it's 2D array.
+struct N2Array<T> {
+    y_size: usize,
+    buf: Vec<T>,
+}
+
+impl<T: Clone> N2Array<T> {
+    fn new(x: usize, y: usize, value: T) -> N2Array<T> {
+        N2Array {
+            y_size: y,
+            buf: vec![value; x * y],
+        }
+    }
+}
+
+impl<T> Index<(usize, usize)> for N2Array<T> {
+    type Output = T;
+
+    #[inline]
+    fn index(&self, (x, y): (usize, usize)) -> &T {
+        &self.buf[(x * self.y_size) + y]
+    }
+}
+
+impl<T> IndexMut<(usize, usize)> for N2Array<T> {
+    #[inline]
+    fn index_mut(&mut self, (x, y): (usize, usize)) -> &mut T {
+        &mut self.buf[(x * self.y_size) + y]
+    }
+}
+
+pub fn prefix_damerau_levenshtein(source: &[u8], target: &[u8]) -> (u32, usize) {
+    let (n, m) = (source.len(), target.len());
+
+    assert!(
+        n <= m,
+        "the source string must be shorter than the target one"
+    );
+
+    if n == 0 {
+        return (m as u32, 0);
+    }
+    if m == 0 {
+        return (n as u32, 0);
+    }
+
+    if n == m && source == target {
+        return (0, m);
+    }
+
+    let inf = n + m;
+    let mut matrix = N2Array::new(n + 2, m + 2, 0);
+
+    matrix[(0, 0)] = inf;
+    for i in 0..n + 1 {
+        matrix[(i + 1, 0)] = inf;
+        matrix[(i + 1, 1)] = i;
+    }
+    for j in 0..m + 1 {
+        matrix[(0, j + 1)] = inf;
+        matrix[(1, j + 1)] = j;
+    }
+
+    let mut last_row = BTreeMap::new();
+
+    for (row, char_s) in source.iter().enumerate() {
+        let mut last_match_col = 0;
+        let row = row + 1;
+
+        for (col, char_t) in target.iter().enumerate() {
+            let col = col + 1;
+            let last_match_row = *last_row.get(&char_t).unwrap_or(&0);
+            let cost = if char_s == char_t { 0 } else { 1 };
+
+            let dist_add = matrix[(row, col + 1)] + 1;
+            let dist_del = matrix[(row + 1, col)] + 1;
+            let dist_sub = matrix[(row, col)] + cost;
+            let dist_trans = matrix[(last_match_row, last_match_col)]
+                + (row - last_match_row - 1)
+                + 1
+                + (col - last_match_col - 1);
+
+            let dist = min(min(dist_add, dist_del), min(dist_sub, dist_trans));
+
+            matrix[(row + 1, col + 1)] = dist;
+
+            if cost == 0 {
+                last_match_col = col;
+            }
+        }
+
+        last_row.insert(char_s, row);
+    }
+
+    let mut minimum = (u32::max_value(), 0);
+
+    for x in n..=m {
+        let dist = matrix[(n + 1, x + 1)] as u32;
+        if dist < minimum.0 {
+            minimum = (dist, x)
+        }
+    }
+
+    minimum
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn matched_length() {
+        let query = "Levenste";
+        let text = "Levenshtein";
+
+        let (dist, length) = prefix_damerau_levenshtein(query.as_bytes(), text.as_bytes());
+        assert_eq!(dist, 1);
+        assert_eq!(&text[..length], "Levenshte");
+    }
+
+    #[test]
+    #[should_panic]
+    fn matched_length_panic() {
+        let query = "Levenshtein";
+        let text = "Levenste";
+
+        // this function will panic if source if longer than target
+        prefix_damerau_levenshtein(query.as_bytes(), text.as_bytes());
+    }
+}

meilisearch-core/src/lib.rs

@@ -0,0 +1,195 @@
+#[cfg(test)]
+#[macro_use]
+extern crate assert_matches;
+#[macro_use]
+extern crate pest_derive;
+
+mod automaton;
+mod bucket_sort;
+mod database;
+mod distinct_map;
+mod error;
+mod filters;
+mod levenshtein;
+mod number;
+mod query_builder;
+mod query_tree;
+mod query_words_mapper;
+mod ranked_map;
+mod raw_document;
+mod reordered_attrs;
+mod update;
+pub mod settings;
+pub mod criterion;
+pub mod raw_indexer;
+pub mod serde;
+pub mod store;
+
+pub use self::database::{BoxUpdateFn, Database, MainT, UpdateT};
+pub use self::error::{Error, HeedError, FstError, MResult, pest_error};
+pub use self::filters::Filter;
+pub use self::number::{Number, ParseNumberError};
+pub use self::ranked_map::RankedMap;
+pub use self::raw_document::RawDocument;
+pub use self::store::Index;
+pub use self::update::{EnqueuedUpdateResult, ProcessedUpdateResult, UpdateStatus, UpdateType};
+pub use meilisearch_types::{DocIndex, DocumentId, Highlight};
+pub use meilisearch_schema::Schema;
+pub use query_words_mapper::QueryWordsMapper;
+
+use std::convert::TryFrom;
+use std::collections::HashMap;
+use compact_arena::SmallArena;
+use log::{error, trace};
+
+use crate::bucket_sort::PostingsListView;
+use crate::levenshtein::prefix_damerau_levenshtein;
+use crate::query_tree::{QueryId, QueryKind};
+use crate::reordered_attrs::ReorderedAttrs;
+
+#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord)]
+pub struct Document {
+    pub id: DocumentId,
+    pub highlights: Vec<Highlight>,
+
+    #[cfg(test)]
+    pub matches: Vec<crate::bucket_sort::SimpleMatch>,
+}
+
+fn highlights_from_raw_document<'a, 'tag, 'txn>(
+    raw_document: &RawDocument<'a, 'tag>,
+    queries_kinds: &HashMap<QueryId, &QueryKind>,
+    arena: &SmallArena<'tag, PostingsListView<'txn>>,
+    searchable_attrs: Option<&ReorderedAttrs>,
+    schema: &Schema,
+) -> Vec<Highlight>
+{
+    let mut highlights = Vec::new();
+
+    for bm in raw_document.bare_matches.iter() {
+        let postings_list = &arena[bm.postings_list];
+        let input = postings_list.input();
+        let kind = &queries_kinds.get(&bm.query_index);
+
+        for di in postings_list.iter() {
+            let covered_area = match kind {
+                Some(QueryKind::NonTolerant(query)) | Some(QueryKind::Tolerant(query)) => {
+                    let len = if query.len() > input.len() {
+                        input.len()
+                    } else {
+                        prefix_damerau_levenshtein(query.as_bytes(), input).1
+                    };
+                    u16::try_from(len).unwrap_or(u16::max_value())
+                },
+                _ => di.char_length,
+            };
+
+            let attribute = searchable_attrs
+                .and_then(|sa| sa.reverse(di.attribute))
+                .unwrap_or(di.attribute);
+
+            let attribute = match schema.indexed_pos_to_field_id(attribute) {
+                Some(field_id) => field_id.0,
+                None => {
+                    error!("Cannot convert indexed_pos {} to field_id", attribute);
+                    trace!("Schema is compromized; {:?}", schema);
+                    continue
+                }
+            };
+
+            let highlight = Highlight {
+                attribute,
+                char_index: di.char_index,
+                char_length: covered_area,
+            };
+
+            highlights.push(highlight);
+        }
+    }
+
+    highlights
+}
+
+impl Document {
+    #[cfg(not(test))]
+    pub fn from_highlights(id: DocumentId, highlights: &[Highlight]) -> Document {
+        Document { id, highlights: highlights.to_owned() }
+    }
+
+    #[cfg(test)]
+    pub fn from_highlights(id: DocumentId, highlights: &[Highlight]) -> Document {
+        Document { id, highlights: highlights.to_owned(), matches: Vec::new() }
+    }
+
+    #[cfg(not(test))]
+    pub fn from_raw<'a, 'tag, 'txn>(
+        raw_document: RawDocument<'a, 'tag>,
+        queries_kinds: &HashMap<QueryId, &QueryKind>,
+        arena: &SmallArena<'tag, PostingsListView<'txn>>,
+        searchable_attrs: Option<&ReorderedAttrs>,
+        schema: &Schema,
+    ) -> Document
+    {
+        let highlights = highlights_from_raw_document(
+            &raw_document,
+            queries_kinds,
+            arena,
+            searchable_attrs,
+            schema,
+        );
+
+        Document { id: raw_document.id, highlights }
+    }
+
+    #[cfg(test)]
+    pub fn from_raw<'a, 'tag, 'txn>(
+        raw_document: RawDocument<'a, 'tag>,
+        queries_kinds: &HashMap<QueryId, &QueryKind>,
+        arena: &SmallArena<'tag, PostingsListView<'txn>>,
+        searchable_attrs: Option<&ReorderedAttrs>,
+        schema: &Schema,
+    ) -> Document
+    {
+        use crate::bucket_sort::SimpleMatch;
+
+        let highlights = highlights_from_raw_document(
+            &raw_document,
+            queries_kinds,
+            arena,
+            searchable_attrs,
+            schema,
+        );
+
+        let mut matches = Vec::new();
+        for sm in raw_document.processed_matches {
+            let attribute = searchable_attrs
+                .and_then(|sa| sa.reverse(sm.attribute))
+                .unwrap_or(sm.attribute);
+
+            let attribute = match schema.indexed_pos_to_field_id(attribute) {
+                Some(field_id) => field_id.0,
+                None => {
+                    error!("Cannot convert indexed_pos {} to field_id", attribute);
+                    trace!("Schema is compromized; {:?}", schema);
+                    continue
+                }
+            };
+
+            matches.push(SimpleMatch { attribute, ..sm });
+        }
+        matches.sort_unstable();
+
+        Document { id: raw_document.id, highlights, matches }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::mem;
+
+    #[test]
+    fn docindex_mem_size() {
+        assert_eq!(mem::size_of::<DocIndex>(), 16);
+    }
+}

meilisearch-core/src/number.rs

@@ -0,0 +1,120 @@
+use std::cmp::Ordering;
+use std::fmt;
+use std::num::{ParseFloatError, ParseIntError};
+use std::str::FromStr;
+
+use ordered_float::OrderedFloat;
+use serde::{Deserialize, Serialize};
+
+#[derive(Serialize, Deserialize, Debug, Copy, Clone, Hash)]
+pub enum Number {
+    Unsigned(u64),
+    Signed(i64),
+    Float(OrderedFloat<f64>),
+    Null,
+}
+
+impl Default for Number {
+    fn default() -> Self {
+        Self::Null
+    }
+}
+
+impl FromStr for Number {
+    type Err = ParseNumberError;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let uint_error = match u64::from_str(s) {
+            Ok(unsigned) => return Ok(Number::Unsigned(unsigned)),
+            Err(error) => error,
+        };
+
+        let int_error = match i64::from_str(s) {
+            Ok(signed) => return Ok(Number::Signed(signed)),
+            Err(error) => error,
+        };
+
+        let float_error = match f64::from_str(s) {
+            Ok(float) => return Ok(Number::Float(OrderedFloat(float))),
+            Err(error) => error,
+        };
+
+        Err(ParseNumberError {
+            uint_error,
+            int_error,
+            float_error,
+        })
+    }
+}
+
+impl PartialEq for Number {
+    fn eq(&self, other: &Number) -> bool {
+        self.cmp(other) == Ordering::Equal
+    }
+}
+
+impl Eq for Number {}
+
+impl PartialOrd for Number {
+    fn partial_cmp(&self, other: &Number) -> Option<Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+impl Ord for Number {
+    fn cmp(&self, other: &Self) -> Ordering {
+        use Number::{Float, Signed, Unsigned, Null};
+
+        match (*self, *other) {
+            (Unsigned(a), Unsigned(b)) => a.cmp(&b),
+            (Unsigned(a), Signed(b)) => {
+                if b < 0 {
+                    Ordering::Greater
+                } else {
+                    a.cmp(&(b as u64))
+                }
+            }
+            (Unsigned(a), Float(b)) => (OrderedFloat(a as f64)).cmp(&b),
+            (Signed(a), Unsigned(b)) => {
+                if a < 0 {
+                    Ordering::Less
+                } else {
+                    (a as u64).cmp(&b)
+                }
+            }
+            (Signed(a), Signed(b)) => a.cmp(&b),
+            (Signed(a), Float(b)) => OrderedFloat(a as f64).cmp(&b),
+            (Float(a), Unsigned(b)) => a.cmp(&OrderedFloat(b as f64)),
+            (Float(a), Signed(b)) => a.cmp(&OrderedFloat(b as f64)),
+            (Float(a), Float(b)) => a.cmp(&b),
+            (Null, Null) => Ordering::Equal,
+            (_, Null) => Ordering::Less,
+            (Null, _) => Ordering::Greater,
+        }
+    }
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct ParseNumberError {
+    uint_error: ParseIntError,
+    int_error: ParseIntError,
+    float_error: ParseFloatError,
+}
+
+impl fmt::Display for ParseNumberError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        if self.uint_error == self.int_error {
+            write!(
+                f,
+                "can not parse number: {}, {}",
+                self.uint_error, self.float_error
+            )
+        } else {
+            write!(
+                f,
+                "can not parse number: {}, {}, {}",
+                self.uint_error, self.int_error, self.float_error
+            )
+        }
+    }
+}

meilisearch-core/src/query_tree.rs

@@ -0,0 +1,560 @@
+use std::borrow::Cow;
+use std::collections::HashMap;
+use std::hash::{Hash, Hasher};
+use std::ops::Range;
+use std::time::Instant;
+use std::{cmp, fmt, iter::once};
+
+use fst::{IntoStreamer, Streamer};
+use itertools::{EitherOrBoth, merge_join_by};
+use meilisearch_tokenizer::split_query_string;
+use sdset::{Set, SetBuf, SetOperation};
+use log::debug;
+
+use crate::database::MainT;
+use crate::{store, DocumentId, DocIndex, MResult};
+use crate::automaton::{normalize_str, build_dfa, build_prefix_dfa, build_exact_dfa};
+use crate::QueryWordsMapper;
+
+#[derive(Clone, PartialEq, Eq, Hash)]
+pub enum Operation {
+    And(Vec<Operation>),
+    Or(Vec<Operation>),
+    Query(Query),
+}
+
+impl fmt::Debug for Operation {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        fn pprint_tree(f: &mut fmt::Formatter<'_>, op: &Operation, depth: usize) -> fmt::Result {
+            match op {
+                Operation::And(children) => {
+                    writeln!(f, "{:1$}AND", "", depth * 2)?;
+                    children.iter().try_for_each(|c| pprint_tree(f, c, depth + 1))
+                },
+                Operation::Or(children) => {
+                    writeln!(f, "{:1$}OR", "", depth * 2)?;
+                    children.iter().try_for_each(|c| pprint_tree(f, c, depth + 1))
+                },
+                Operation::Query(query) => writeln!(f, "{:2$}{:?}", "", query, depth * 2),
+            }
+        }
+
+        pprint_tree(f, self, 0)
+    }
+}
+
+impl Operation {
+    fn tolerant(id: QueryId, prefix: bool, s: &str) -> Operation {
+        Operation::Query(Query { id, prefix, exact: true, kind: QueryKind::Tolerant(s.to_string()) })
+    }
+
+    fn non_tolerant(id: QueryId, prefix: bool, s: &str) -> Operation {
+        Operation::Query(Query { id, prefix, exact: true, kind: QueryKind::NonTolerant(s.to_string()) })
+    }
+
+    fn phrase2(id: QueryId, prefix: bool, (left, right): (&str, &str)) -> Operation {
+        let kind = QueryKind::Phrase(vec![left.to_owned(), right.to_owned()]);
+        Operation::Query(Query { id, prefix, exact: true, kind })
+    }
+}
+
+pub type QueryId = usize;
+
+#[derive(Clone, Eq)]
+pub struct Query {
+    pub id: QueryId,
+    pub prefix: bool,
+    pub exact: bool,
+    pub kind: QueryKind,
+}
+
+impl PartialEq for Query {
+    fn eq(&self, other: &Self) -> bool {
+        self.prefix == other.prefix && self.kind == other.kind
+    }
+}
+
+impl Hash for Query {
+    fn hash<H: Hasher>(&self, state: &mut H) {
+        self.prefix.hash(state);
+        self.kind.hash(state);
+    }
+}
+
+#[derive(Clone, PartialEq, Eq, Hash)]
+pub enum QueryKind {
+    Tolerant(String),
+    NonTolerant(String),
+    Phrase(Vec<String>),
+}
+
+impl fmt::Debug for Query {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let Query { id, prefix, kind, .. } = self;
+        let prefix = if *prefix { String::from("Prefix") } else { String::default() };
+        match kind {
+            QueryKind::NonTolerant(word) => {
+                f.debug_struct(&(prefix + "NonTolerant")).field("id", &id).field("word", &word).finish()
+            },
+            QueryKind::Tolerant(word) => {
+                f.debug_struct(&(prefix + "Tolerant")).field("id", &id).field("word", &word).finish()
+            },
+            QueryKind::Phrase(words) => {
+                f.debug_struct(&(prefix + "Phrase")).field("id", &id).field("words", &words).finish()
+            },
+        }
+    }
+}
+
+#[derive(Debug, Default)]
+pub struct PostingsList {
+    docids: SetBuf<DocumentId>,
+    matches: SetBuf<DocIndex>,
+}
+
+pub struct Context {
+    pub words_set: fst::Set,
+    pub stop_words: fst::Set,
+    pub synonyms: store::Synonyms,
+    pub postings_lists: store::PostingsLists,
+    pub prefix_postings_lists: store::PrefixPostingsListsCache,
+}
+
+fn split_best_frequency<'a>(reader: &heed::RoTxn<MainT>, ctx: &Context, word: &'a str) -> MResult<Option<(&'a str, &'a str)>> {
+    let chars = word.char_indices().skip(1);
+    let mut best = None;
+
+    for (i, _) in chars {
+        let (left, right) = word.split_at(i);
+
+        let left_freq = ctx.postings_lists
+            .postings_list(reader, left.as_bytes())?
+            .map(|p| p.docids.len())
+            .unwrap_or(0);
+        let right_freq = ctx.postings_lists
+            .postings_list(reader, right.as_bytes())?
+            .map(|p| p.docids.len())
+            .unwrap_or(0);
+
+        let min_freq = cmp::min(left_freq, right_freq);
+        if min_freq != 0 && best.map_or(true, |(old, _, _)| min_freq > old) {
+            best = Some((min_freq, left, right));
+        }
+    }
+
+    Ok(best.map(|(_, l, r)| (l, r)))
+}
+
+fn fetch_synonyms(reader: &heed::RoTxn<MainT>, ctx: &Context, words: &[&str]) -> MResult<Vec<Vec<String>>> {
+    let words = normalize_str(&words.join(" "));
+    let set = ctx.synonyms.synonyms(reader, words.as_bytes())?.unwrap_or_default();
+
+    let mut strings = Vec::new();
+    let mut stream = set.stream();
+    while let Some(input) = stream.next() {
+        if let Ok(input) = std::str::from_utf8(input) {
+            let alts = input.split_ascii_whitespace().map(ToOwned::to_owned).collect();
+            strings.push(alts);
+        }
+    }
+
+    Ok(strings)
+}
+
+fn create_operation<I, F>(iter: I, f: F) -> Operation
+where I: IntoIterator<Item=Operation>,
+      F: Fn(Vec<Operation>) -> Operation,
+{
+    let mut iter = iter.into_iter();
+    match (iter.next(), iter.next()) {
+        (Some(first), None) => first,
+        (first, second) => f(first.into_iter().chain(second).chain(iter).collect()),
+    }
+}
+
+const MAX_NGRAM: usize = 3;
+
+pub fn create_query_tree(
+    reader: &heed::RoTxn<MainT>,
+    ctx: &Context,
+    query: &str,
+) -> MResult<(Operation, HashMap<QueryId, Range<usize>>)>
+{
+    let words = split_query_string(query).map(str::to_lowercase);
+    let words = words.filter(|w| !ctx.stop_words.contains(w));
+    let words: Vec<_> = words.enumerate().collect();
+
+    let mut mapper = QueryWordsMapper::new(words.iter().map(|(_, w)| w));
+
+    fn create_inner(
+        reader: &heed::RoTxn<MainT>,
+        ctx: &Context,
+        mapper: &mut QueryWordsMapper,
+        words: &[(usize, String)],
+    ) -> MResult<Vec<Operation>>
+    {
+        let mut alts = Vec::new();
+
+        for ngram in 1..=MAX_NGRAM {
+            if let Some(group) = words.get(..ngram) {
+                let mut group_ops = Vec::new();
+
+                let tail = &words[ngram..];
+                let is_last = tail.is_empty();
+
+                let mut group_alts = Vec::new();
+                match group {
+                    [(id, word)] => {
+                        let mut idgen = ((id + 1) * 100)..;
+                        let range = (*id)..id+1;
+
+                        let phrase = split_best_frequency(reader, ctx, word)?
+                            .map(|ws| {
+                                let id = idgen.next().unwrap();
+                                idgen.next().unwrap();
+                                mapper.declare(range.clone(), id, &[ws.0, ws.1]);
+                                Operation::phrase2(id, is_last, ws)
+                            });
+
+                        let synonyms = fetch_synonyms(reader, ctx, &[word])?
+                            .into_iter()
+                            .map(|alts| {
+                                let exact = alts.len() == 1;
+                                let id = idgen.next().unwrap();
+                                mapper.declare(range.clone(), id, &alts);
+
+                                let mut idgen = once(id).chain(&mut idgen);
+                                let iter = alts.into_iter().map(|w| {
+                                    let id = idgen.next().unwrap();
+                                    let kind = QueryKind::NonTolerant(w);
+                                    Operation::Query(Query { id, prefix: false, exact, kind })
+                                });
+
+                                create_operation(iter, Operation::And)
+                            });
+
+                        let original = Operation::tolerant(*id, is_last, word);
+
+                        group_alts.push(original);
+                        group_alts.extend(synonyms.chain(phrase));
+                    },
+                    words => {
+                        let id = words[0].0;
+                        let mut idgen = ((id + 1) * 100_usize.pow(ngram as u32))..;
+                        let range = id..id+ngram;
+
+                        let words: Vec<_> = words.iter().map(|(_, s)| s.as_str()).collect();
+
+                        for synonym in fetch_synonyms(reader, ctx, &words)? {
+                            let exact = synonym.len() == 1;
+                            let id = idgen.next().unwrap();
+                            mapper.declare(range.clone(), id, &synonym);
+
+                            let mut idgen = once(id).chain(&mut idgen);
+                            let synonym = synonym.into_iter().map(|s| {
+                                let id = idgen.next().unwrap();
+                                let kind = QueryKind::NonTolerant(s);
+                                Operation::Query(Query { id, prefix: false, exact, kind })
+                            });
+                            group_alts.push(create_operation(synonym, Operation::And));
+                        }
+
+                        let id = idgen.next().unwrap();
+                        let concat = words.concat();
+                        mapper.declare(range.clone(), id, &[&concat]);
+                        group_alts.push(Operation::non_tolerant(id, is_last, &concat));
+                    }
+                }
+
+                group_ops.push(create_operation(group_alts, Operation::Or));
+
+                if !tail.is_empty() {
+                    let tail_ops = create_inner(reader, ctx, mapper, tail)?;
+                    group_ops.push(create_operation(tail_ops, Operation::Or));
+                }
+
+                alts.push(create_operation(group_ops, Operation::And));
+            }
+        }
+
+        Ok(alts)
+    }
+
+    let alternatives = create_inner(reader, ctx, &mut mapper, &words)?;
+    let operation = Operation::Or(alternatives);
+    let mapping = mapper.mapping();
+
+    Ok((operation, mapping))
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
+pub struct PostingsKey<'o> {
+    pub query: &'o Query,
+    pub input: Vec<u8>,
+    pub distance: u8,
+    pub is_exact: bool,
+}
+
+pub type Postings<'o, 'txn> = HashMap<PostingsKey<'o>, Cow<'txn, Set<DocIndex>>>;
+pub type Cache<'o, 'txn> = HashMap<&'o Operation, Cow<'txn, Set<DocumentId>>>;
+
+pub struct QueryResult<'o, 'txn> {
+    pub docids: Cow<'txn, Set<DocumentId>>,
+    pub queries: Postings<'o, 'txn>,
+}
+
+pub fn traverse_query_tree<'o, 'txn>(
+    reader: &'txn heed::RoTxn<MainT>,
+    ctx: &Context,
+    tree: &'o Operation,
+) -> MResult<QueryResult<'o, 'txn>>
+{
+    fn execute_and<'o, 'txn>(
+        reader: &'txn heed::RoTxn<MainT>,
+        ctx: &Context,
+        cache: &mut Cache<'o, 'txn>,
+        postings: &mut Postings<'o, 'txn>,
+        depth: usize,
+        operations: &'o [Operation],
+    ) -> MResult<Cow<'txn, Set<DocumentId>>>
+    {
+        debug!("{:1$}AND", "", depth * 2);
+
+        let before = Instant::now();
+        let mut results = Vec::new();
+
+        for op in operations {
+            if cache.get(op).is_none() {
+                let docids = match op {
+                    Operation::And(ops) => execute_and(reader, ctx, cache, postings, depth + 1, &ops)?,
+                    Operation::Or(ops) => execute_or(reader, ctx, cache, postings, depth + 1, &ops)?,
+                    Operation::Query(query) => execute_query(reader, ctx, postings, depth + 1, &query)?,
+                };
+                cache.insert(op, docids);
+            }
+        }
+
+        for op in operations {
+            if let Some(docids) = cache.get(op) {
+                results.push(docids.as_ref());
+            }
+        }
+
+        let op = sdset::multi::Intersection::new(results);
+        let docids = op.into_set_buf();
+
+        debug!("{:3$}--- AND fetched {} documents in {:.02?}", "", docids.len(), before.elapsed(), depth * 2);
+
+        Ok(Cow::Owned(docids))
+    }
+
+    fn execute_or<'o, 'txn>(
+        reader: &'txn heed::RoTxn<MainT>,
+        ctx: &Context,
+        cache: &mut Cache<'o, 'txn>,
+        postings: &mut Postings<'o, 'txn>,
+        depth: usize,
+        operations: &'o [Operation],
+    ) -> MResult<Cow<'txn, Set<DocumentId>>>
+    {
+        debug!("{:1$}OR", "", depth * 2);
+
+        let before = Instant::now();
+        let mut results = Vec::new();
+
+        for op in operations {
+            if cache.get(op).is_none() {
+                let docids = match op {
+                    Operation::And(ops) => execute_and(reader, ctx, cache, postings, depth + 1, &ops)?,
+                    Operation::Or(ops) => execute_or(reader, ctx, cache, postings, depth + 1, &ops)?,
+                    Operation::Query(query) => execute_query(reader, ctx, postings, depth + 1, &query)?,
+                };
+                cache.insert(op, docids);
+            }
+        }
+
+        for op in operations {
+            if let Some(docids) = cache.get(op) {
+                results.push(docids.as_ref());
+            }
+        }
+
+        let op = sdset::multi::Union::new(results);
+        let docids = op.into_set_buf();
+
+        debug!("{:3$}--- OR fetched {} documents in {:.02?}", "", docids.len(), before.elapsed(), depth * 2);
+
+        Ok(Cow::Owned(docids))
+    }
+
+    fn execute_query<'o, 'txn>(
+        reader: &'txn heed::RoTxn<MainT>,
+        ctx: &Context,
+        postings: &mut Postings<'o, 'txn>,
+        depth: usize,
+        query: &'o Query,
+    ) -> MResult<Cow<'txn, Set<DocumentId>>>
+    {
+        let before = Instant::now();
+
+        let Query { prefix, kind, exact, .. } = query;
+        let docids: Cow<Set<_>> = match kind {
+            QueryKind::Tolerant(word) => {
+                if *prefix && word.len() <= 2 {
+                    let prefix = {
+                        let mut array = [0; 4];
+                        let bytes = word.as_bytes();
+                        array[..bytes.len()].copy_from_slice(bytes);
+                        array
+                    };
+
+                    // We retrieve the cached postings lists for all
+                    // the words that starts with this short prefix.
+                    let result = ctx.prefix_postings_lists.prefix_postings_list(reader, prefix)?.unwrap_or_default();
+                    let key = PostingsKey { query, input: word.clone().into_bytes(), distance: 0, is_exact: false };
+                    postings.insert(key, result.matches);
+                    let prefix_docids = &result.docids;
+
+                    // We retrieve the exact postings list for the prefix,
+                    // because we must consider these matches as exact.
+                    let result = ctx.postings_lists.postings_list(reader, word.as_bytes())?.unwrap_or_default();
+                    let key = PostingsKey { query, input: word.clone().into_bytes(), distance: 0, is_exact: true };
+                    postings.insert(key, result.matches);
+                    let exact_docids = &result.docids;
+
+                    let before = Instant::now();
+                    let docids = sdset::duo::Union::new(prefix_docids, exact_docids).into_set_buf();
+                    debug!("{:4$}prefix docids ({} and {}) construction took {:.02?}",
+                        "", prefix_docids.len(), exact_docids.len(), before.elapsed(), depth * 2);
+
+                    Cow::Owned(docids)
+
+                } else {
+                    let dfa = if *prefix { build_prefix_dfa(word) } else { build_dfa(word) };
+
+                    let byte = word.as_bytes()[0];
+                    let mut stream = if byte == u8::max_value() {
+                        ctx.words_set.search(&dfa).ge(&[byte]).into_stream()
+                    } else {
+                        ctx.words_set.search(&dfa).ge(&[byte]).lt(&[byte + 1]).into_stream()
+                    };
+
+                    let before = Instant::now();
+                    let mut results = Vec::new();
+                    while let Some(input) = stream.next() {
+                        if let Some(result) = ctx.postings_lists.postings_list(reader, input)? {
+                            let distance = dfa.eval(input).to_u8();
+                            let is_exact = *exact && distance == 0 && input.len() == word.len();
+                            results.push(result.docids);
+                            let key = PostingsKey { query, input: input.to_owned(), distance, is_exact };
+                            postings.insert(key, result.matches);
+                        }
+                    }
+                    debug!("{:3$}docids retrieval ({:?}) took {:.02?}", "", results.len(), before.elapsed(), depth * 2);
+
+                    let before = Instant::now();
+                    let docids = if results.len() > 10 {
+                        let cap = results.iter().map(|dis| dis.len()).sum();
+                        let mut docids = Vec::with_capacity(cap);
+                        for dis in results {
+                            docids.extend_from_slice(&dis);
+                        }
+                        SetBuf::from_dirty(docids)
+                    } else {
+                        let sets = results.iter().map(AsRef::as_ref).collect();
+                        sdset::multi::Union::new(sets).into_set_buf()
+                    };
+                    debug!("{:2$}docids construction took {:.02?}", "", before.elapsed(), depth * 2);
+
+                    Cow::Owned(docids)
+                }
+            },
+            QueryKind::NonTolerant(word) => {
+                // TODO support prefix and non-prefix exact DFA
+                let dfa = build_exact_dfa(word);
+
+                let byte = word.as_bytes()[0];
+                let mut stream = if byte == u8::max_value() {
+                    ctx.words_set.search(&dfa).ge(&[byte]).into_stream()
+                } else {
+                    ctx.words_set.search(&dfa).ge(&[byte]).lt(&[byte + 1]).into_stream()
+                };
+
+                let before = Instant::now();
+                let mut results = Vec::new();
+                while let Some(input) = stream.next() {
+                    if let Some(result) = ctx.postings_lists.postings_list(reader, input)? {
+                        let distance = dfa.eval(input).to_u8();
+                        results.push(result.docids);
+                        let key = PostingsKey { query, input: input.to_owned(), distance, is_exact: *exact };
+                        postings.insert(key, result.matches);
+                    }
+                }
+                debug!("{:3$}docids retrieval ({:?}) took {:.02?}", "", results.len(), before.elapsed(), depth * 2);
+
+                let before = Instant::now();
+                let docids = if results.len() > 10 {
+                    let cap = results.iter().map(|dis| dis.len()).sum();
+                    let mut docids = Vec::with_capacity(cap);
+                    for dis in results {
+                        docids.extend_from_slice(&dis);
+                    }
+                    SetBuf::from_dirty(docids)
+                } else {
+                    let sets = results.iter().map(AsRef::as_ref).collect();
+                    sdset::multi::Union::new(sets).into_set_buf()
+                };
+                debug!("{:2$}docids construction took {:.02?}", "", before.elapsed(), depth * 2);
+
+                Cow::Owned(docids)
+            },
+            QueryKind::Phrase(words) => {
+                // TODO support prefix and non-prefix exact DFA
+                if let [first, second] = words.as_slice() {
+                    let first = ctx.postings_lists.postings_list(reader, first.as_bytes())?.unwrap_or_default();
+                    let second = ctx.postings_lists.postings_list(reader, second.as_bytes())?.unwrap_or_default();
+
+                    let iter = merge_join_by(first.matches.as_slice(), second.matches.as_slice(), |a, b| {
+                        let x = (a.document_id, a.attribute, (a.word_index as u32) + 1);
+                        let y = (b.document_id, b.attribute, b.word_index as u32);
+                        x.cmp(&y)
+                    });
+
+                    let matches: Vec<_> = iter
+                        .filter_map(EitherOrBoth::both)
+                        .flat_map(|(a, b)| once(*a).chain(Some(*b)))
+                        .collect();
+
+                    let before = Instant::now();
+                    let mut docids: Vec<_> = matches.iter().map(|m| m.document_id).collect();
+                    docids.dedup();
+                    let docids = SetBuf::new(docids).unwrap();
+                    debug!("{:2$}docids construction took {:.02?}", "", before.elapsed(), depth * 2);
+
+                    let matches = Cow::Owned(SetBuf::from_dirty(matches));
+                    let key = PostingsKey { query, input: vec![], distance: 0, is_exact: true };
+                    postings.insert(key, matches);
+
+                    Cow::Owned(docids)
+                } else {
+                    debug!("{:2$}{:?} skipped", "", words, depth * 2);
+                    Cow::default()
+                }
+            },
+        };
+
+        debug!("{:4$}{:?} fetched {:?} documents in {:.02?}", "", query, docids.len(), before.elapsed(), depth * 2);
+        Ok(docids)
+    }
+
+    let mut cache = Cache::new();
+    let mut postings = Postings::new();
+
+    let docids = match tree {
+        Operation::And(ops) => execute_and(reader, ctx, &mut cache, &mut postings, 0, &ops)?,
+        Operation::Or(ops) => execute_or(reader, ctx, &mut cache, &mut postings, 0, &ops)?,
+        Operation::Query(query) => execute_query(reader, ctx, &mut postings, 0, &query)?,
+    };
+
+    Ok(QueryResult { docids, queries: postings })
+}

meilisearch-core/src/query_words_mapper.rs

@@ -0,0 +1,415 @@
+use std::collections::HashMap;
+use std::iter::FromIterator;
+use std::ops::Range;
+use intervaltree::{Element, IntervalTree};
+
+pub type QueryId = usize;
+
+pub struct QueryWordsMapper {
+    originals: Vec<String>,
+    mappings: HashMap<QueryId, (Range<usize>, Vec<String>)>,
+}
+
+impl QueryWordsMapper {
+    pub fn new<I, A>(originals: I) -> QueryWordsMapper
+    where I: IntoIterator<Item = A>,
+          A: ToString,
+    {
+        let originals = originals.into_iter().map(|s| s.to_string()).collect();
+        QueryWordsMapper { originals, mappings: HashMap::new() }
+    }
+
+    pub fn declare<I, A>(&mut self, range: Range<usize>, id: QueryId, replacement: I)
+    where I: IntoIterator<Item = A>,
+          A: ToString,
+    {
+        assert!(range.len() != 0);
+        assert!(self.originals.get(range.clone()).is_some());
+        assert!(id >= self.originals.len());
+
+        let replacement: Vec<_> = replacement.into_iter().map(|s| s.to_string()).collect();
+
+        assert!(!replacement.is_empty());
+
+        // We detect words at the end and at the front of the
+        // replacement that are common with the originals:
+        //
+        //     x a b c d e f g
+        //       ^^^/   \^^^
+        //     a b x c d k j e f
+        //     ^^^           ^^^
+        //
+
+        let left = &self.originals[..range.start];
+        let right = &self.originals[range.end..];
+
+        let common_left = longest_common_prefix(left, &replacement);
+        let common_right = longest_common_prefix(&replacement, right);
+
+        for i in 0..common_left {
+            let range = range.start - common_left + i..range.start - common_left + i + 1;
+            let replacement = vec![replacement[i].clone()];
+            self.mappings.insert(id + i, (range, replacement));
+        }
+
+        {
+            let replacement = replacement[common_left..replacement.len() - common_right].iter().cloned().collect();
+            self.mappings.insert(id + common_left, (range.clone(), replacement));
+        }
+
+        for i in 0..common_right {
+            let id = id + replacement.len() - common_right + i;
+            let range = range.end + i..range.end + i + 1;
+            let replacement = vec![replacement[replacement.len() - common_right + i].clone()];
+            self.mappings.insert(id, (range, replacement));
+        }
+    }
+
+    pub fn mapping(self) -> HashMap<QueryId, Range<usize>> {
+        let mappings = self.mappings.into_iter().map(|(i, (r, v))| (r, (i, v)));
+        let intervals = IntervalTree::from_iter(mappings);
+
+        let mut output = HashMap::new();
+        let mut offset = 0;
+
+        // We map each original word to the biggest number of
+        // associated words.
+        for i in 0..self.originals.len() {
+            let max = intervals.query_point(i)
+                .filter_map(|e| {
+                    if e.range.end - 1 == i {
+                        let len = e.value.1.iter().skip(i - e.range.start).count();
+                        if len != 0 { Some(len) } else { None }
+                    } else { None }
+                })
+                .max()
+                .unwrap_or(1);
+
+            let range = i + offset..i + offset + max;
+            output.insert(i, range);
+            offset += max - 1;
+        }
+
+        // We retrieve the range that each original word
+        // is mapped to and apply it to each of the words.
+        for i in 0..self.originals.len() {
+
+            let iter = intervals.query_point(i).filter(|e| e.range.end - 1 == i);
+            for Element { range, value: (id, words) } in iter {
+
+                // We ask for the complete range mapped to the area we map.
+                let start = output.get(&range.start).map(|r| r.start).unwrap_or(range.start);
+                let end = output.get(&(range.end - 1)).map(|r| r.end).unwrap_or(range.end);
+                let range = start..end;
+
+                // We map each query id to one word until the last,
+                // we map it to the remainings words.
+                let add = range.len() - words.len();
+                for (j, x) in range.take(words.len()).enumerate() {
+                    let add = if j == words.len() - 1 { add } else { 0 }; // is last?
+                    let range = x..x + 1 + add;
+                    output.insert(id + j, range);
+                }
+            }
+        }
+
+        output
+    }
+}
+
+fn longest_common_prefix<T: Eq + std::fmt::Debug>(a: &[T], b: &[T]) -> usize {
+    let mut best = None;
+    for i in (0..a.len()).rev() {
+        let count = a[i..].iter().zip(b).take_while(|(a, b)| a == b).count();
+        best = match best {
+            Some(old) if count > old => Some(count),
+            Some(_) => break,
+            None => Some(count),
+        };
+    }
+    best.unwrap_or(0)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn original_unmodified() {
+        let query = ["new", "york", "city", "subway"];
+        //             0       1       2        3
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // new york = new york city
+        builder.declare(0..2, 4, &["new", "york", "city"]);
+        //                    ^      4       5       6
+
+        // new = new york city
+        builder.declare(0..1, 7, &["new", "york", "city"]);
+        //                    ^      7       8       9
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..1); // new
+        assert_eq!(mapping[&1], 1..2); // york
+        assert_eq!(mapping[&2], 2..3); // city
+        assert_eq!(mapping[&3], 3..4); // subway
+
+        assert_eq!(mapping[&4], 0..1); // new
+        assert_eq!(mapping[&5], 1..2); // york
+        assert_eq!(mapping[&6], 2..3); // city
+
+        assert_eq!(mapping[&7], 0..1); // new
+        assert_eq!(mapping[&8], 1..2); // york
+        assert_eq!(mapping[&9], 2..3); // city
+    }
+
+    #[test]
+    fn original_unmodified2() {
+        let query = ["new", "york", "city", "subway"];
+        //             0       1       2        3
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // city subway = new york city underground train
+        builder.declare(2..4, 4, &["new", "york", "city", "underground", "train"]);
+        //                    ^      4      5       6           7           8
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..1); // new
+        assert_eq!(mapping[&1], 1..2); // york
+        assert_eq!(mapping[&2], 2..3); // city
+        assert_eq!(mapping[&3], 3..5); // subway
+
+        assert_eq!(mapping[&4], 0..1); // new
+        assert_eq!(mapping[&5], 1..2); // york
+        assert_eq!(mapping[&6], 2..3); // city
+        assert_eq!(mapping[&7], 3..4); // underground
+        assert_eq!(mapping[&8], 4..5); // train
+    }
+
+    #[test]
+    fn original_unmodified3() {
+        let query = ["a", "b", "x", "x", "a", "b", "c", "d", "e", "f", "g"];
+        //            0    1    2    3    4    5    6    7    8    9    10
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // c d = a b x c d k j e f
+        builder.declare(6..8, 11, &["a", "b", "x", "c", "d", "k", "j", "e", "f"]);
+        //                    ^^    11   12   13   14   15   16   17   18   19
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0],  0..1); // a
+        assert_eq!(mapping[&1],  1..2); // b
+        assert_eq!(mapping[&2],  2..3); // x
+        assert_eq!(mapping[&3],  3..4); // x
+        assert_eq!(mapping[&4],  4..5); // a
+        assert_eq!(mapping[&5],  5..6); // b
+        assert_eq!(mapping[&6],  6..7); // c
+        assert_eq!(mapping[&7],  7..11); // d
+        assert_eq!(mapping[&8],  11..12); // e
+        assert_eq!(mapping[&9],  12..13); // f
+        assert_eq!(mapping[&10], 13..14); // g
+
+        assert_eq!(mapping[&11], 4..5); // a
+        assert_eq!(mapping[&12], 5..6); // b
+        assert_eq!(mapping[&13], 6..7); // x
+        assert_eq!(mapping[&14], 7..8); // c
+        assert_eq!(mapping[&15], 8..9); // d
+        assert_eq!(mapping[&16], 9..10); // k
+        assert_eq!(mapping[&17], 10..11); // j
+        assert_eq!(mapping[&18], 11..12); // e
+        assert_eq!(mapping[&19], 12..13); // f
+    }
+
+    #[test]
+    fn simple_growing() {
+        let query = ["new", "york", "subway"];
+        //             0       1        2
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // new york = new york city
+        builder.declare(0..2, 3, &["new", "york", "city"]);
+        //                    ^      3       4       5
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..1); // new
+        assert_eq!(mapping[&1], 1..3); // york
+        assert_eq!(mapping[&2], 3..4); // subway
+        assert_eq!(mapping[&3], 0..1); // new
+        assert_eq!(mapping[&4], 1..2); // york
+        assert_eq!(mapping[&5], 2..3); // city
+    }
+
+    #[test]
+    fn same_place_growings() {
+        let query = ["NY", "subway"];
+        //             0       1
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // NY = new york
+        builder.declare(0..1, 2, &["new", "york"]);
+        //                    ^      2       3
+
+        // NY = new york city
+        builder.declare(0..1, 4, &["new", "york", "city"]);
+        //                    ^      4       5       6
+
+        // NY = NYC
+        builder.declare(0..1, 7, &["NYC"]);
+        //                    ^      7
+
+        // NY = new york city
+        builder.declare(0..1, 8, &["new", "york", "city"]);
+        //                    ^      8       9      10
+
+        // subway = underground train
+        builder.declare(1..2, 11, &["underground", "train"]);
+        //                    ^          11          12
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..3); // NY
+        assert_eq!(mapping[&1], 3..5); // subway
+        assert_eq!(mapping[&2], 0..1); // new
+        assert_eq!(mapping[&3], 1..3); // york
+        assert_eq!(mapping[&4], 0..1); // new
+        assert_eq!(mapping[&5], 1..2); // york
+        assert_eq!(mapping[&6], 2..3); // city
+        assert_eq!(mapping[&7], 0..3); // NYC
+        assert_eq!(mapping[&8], 0..1); // new
+        assert_eq!(mapping[&9], 1..2); // york
+        assert_eq!(mapping[&10], 2..3); // city
+        assert_eq!(mapping[&11], 3..4); // underground
+        assert_eq!(mapping[&12], 4..5); // train
+    }
+
+    #[test]
+    fn bigger_growing() {
+        let query = ["NYC", "subway"];
+        //             0        1
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // NYC = new york city
+        builder.declare(0..1, 2, &["new", "york", "city"]);
+        //                    ^      2       3       4
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..3); // NYC
+        assert_eq!(mapping[&1], 3..4); // subway
+        assert_eq!(mapping[&2], 0..1); // new
+        assert_eq!(mapping[&3], 1..2); // york
+        assert_eq!(mapping[&4], 2..3); // city
+    }
+
+    #[test]
+    fn middle_query_growing() {
+        let query = ["great", "awesome", "NYC", "subway"];
+        //              0         1        2        3
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // NYC = new york city
+        builder.declare(2..3, 4, &["new", "york", "city"]);
+        //                    ^      4       5       6
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..1); // great
+        assert_eq!(mapping[&1], 1..2); // awesome
+        assert_eq!(mapping[&2], 2..5); // NYC
+        assert_eq!(mapping[&3], 5..6); // subway
+        assert_eq!(mapping[&4], 2..3); // new
+        assert_eq!(mapping[&5], 3..4); // york
+        assert_eq!(mapping[&6], 4..5); // city
+    }
+
+    #[test]
+    fn end_query_growing() {
+        let query = ["NYC", "subway"];
+        //             0        1
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // NYC = new york city
+        builder.declare(1..2, 2, &["underground", "train"]);
+        //                    ^         2            3
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..1); // NYC
+        assert_eq!(mapping[&1], 1..3); // subway
+        assert_eq!(mapping[&2], 1..2); // underground
+        assert_eq!(mapping[&3], 2..3); // train
+    }
+
+    #[test]
+    fn multiple_growings() {
+        let query = ["great", "awesome", "NYC", "subway"];
+        //              0         1        2        3
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // NYC = new york city
+        builder.declare(2..3, 4, &["new", "york", "city"]);
+        //                    ^      4       5       6
+
+        // subway = underground train
+        builder.declare(3..4, 7, &["underground", "train"]);
+        //                    ^          7           8
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..1); // great
+        assert_eq!(mapping[&1], 1..2); // awesome
+        assert_eq!(mapping[&2], 2..5); // NYC
+        assert_eq!(mapping[&3], 5..7); // subway
+        assert_eq!(mapping[&4], 2..3); // new
+        assert_eq!(mapping[&5], 3..4); // york
+        assert_eq!(mapping[&6], 4..5); // city
+        assert_eq!(mapping[&7], 5..6); // underground
+        assert_eq!(mapping[&8], 6..7); // train
+    }
+
+    #[test]
+    fn multiple_probable_growings() {
+        let query = ["great", "awesome", "NYC", "subway"];
+        //              0         1        2        3
+        let mut builder = QueryWordsMapper::new(&query);
+
+        // NYC = new york city
+        builder.declare(2..3, 4, &["new", "york", "city"]);
+        //                    ^      4       5       6
+
+        // subway = underground train
+        builder.declare(3..4, 7, &["underground", "train"]);
+        //                    ^          7           8
+
+        // great awesome = good
+        builder.declare(0..2, 9, &["good"]);
+        //                    ^       9
+
+        // awesome NYC = NY
+        builder.declare(1..3, 10, &["NY"]);
+        //                    ^^     10
+
+        // NYC subway = metro
+        builder.declare(2..4, 11, &["metro"]);
+        //                    ^^      11
+
+        let mapping = builder.mapping();
+
+        assert_eq!(mapping[&0], 0..1); // great
+        assert_eq!(mapping[&1], 1..2); // awesome
+        assert_eq!(mapping[&2], 2..5); // NYC
+        assert_eq!(mapping[&3], 5..7); // subway
+        assert_eq!(mapping[&4], 2..3); // new
+        assert_eq!(mapping[&5], 3..4); // york
+        assert_eq!(mapping[&6], 4..5); // city
+        assert_eq!(mapping[&7], 5..6); // underground
+        assert_eq!(mapping[&8], 6..7); // train
+        assert_eq!(mapping[&9], 0..2); // good
+        assert_eq!(mapping[&10], 1..5); // NY
+        assert_eq!(mapping[&11], 2..7); // metro
+    }
+}

meilisearch-core/src/ranked_map.rs

@@ -0,0 +1,41 @@
+use std::io::{Read, Write};
+
+use hashbrown::HashMap;
+use meilisearch_schema::FieldId;
+use serde::{Deserialize, Serialize};
+
+use crate::{DocumentId, Number};
+
+#[derive(Debug, Default, Clone, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(transparent)]
+pub struct RankedMap(HashMap<(DocumentId, FieldId), Number>);
+
+impl RankedMap {
+    pub fn len(&self) -> usize {
+        self.0.len()
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.0.is_empty()
+    }
+
+    pub fn insert(&mut self, document: DocumentId, field: FieldId, number: Number) {
+        self.0.insert((document, field), number);
+    }
+
+    pub fn remove(&mut self, document: DocumentId, field: FieldId) {
+        self.0.remove(&(document, field));
+    }
+
+    pub fn get(&self, document: DocumentId, field: FieldId) -> Option<Number> {
+        self.0.get(&(document, field)).cloned()
+    }
+
+    pub fn read_from_bin<R: Read>(reader: R) -> bincode::Result<RankedMap> {
+        bincode::deserialize_from(reader).map(RankedMap)
+    }
+
+    pub fn write_to_bin<W: Write>(&self, writer: W) -> bincode::Result<()> {
+        bincode::serialize_into(writer, &self.0)
+    }
+}

meilisearch-core/src/raw_document.rs

@@ -0,0 +1,51 @@
+use compact_arena::SmallArena;
+use sdset::SetBuf;
+use crate::DocIndex;
+use crate::bucket_sort::{SimpleMatch, BareMatch, PostingsListView};
+use crate::reordered_attrs::ReorderedAttrs;
+
+pub struct RawDocument<'a, 'tag> {
+    pub id: crate::DocumentId,
+    pub bare_matches: &'a mut [BareMatch<'tag>],
+    pub processed_matches: Vec<SimpleMatch>,
+    /// The list of minimum `distance` found
+    pub processed_distances: Vec<Option<u8>>,
+    /// Does this document contains a field
+    /// with one word that is exactly matching
+    pub contains_one_word_field: bool,
+}
+
+impl<'a, 'tag> RawDocument<'a, 'tag> {
+    pub fn new<'txn>(
+        bare_matches: &'a mut [BareMatch<'tag>],
+        postings_lists: &mut SmallArena<'tag, PostingsListView<'txn>>,
+        searchable_attrs: Option<&ReorderedAttrs>,
+    ) -> RawDocument<'a, 'tag>
+    {
+        if let Some(reordered_attrs) = searchable_attrs {
+            for bm in bare_matches.iter() {
+                let postings_list = &postings_lists[bm.postings_list];
+
+                let mut rewritten = Vec::new();
+                for di in postings_list.iter() {
+                    if let Some(attribute) = reordered_attrs.get(di.attribute) {
+                        rewritten.push(DocIndex { attribute, ..*di });
+                    }
+                }
+
+                let new_postings = SetBuf::from_dirty(rewritten);
+                postings_lists[bm.postings_list].rewrite_with(new_postings);
+            }
+        }
+
+        bare_matches.sort_unstable_by_key(|m| m.query_index);
+
+        RawDocument {
+            id: bare_matches[0].document_id,
+            bare_matches,
+            processed_matches: Vec::new(),
+            processed_distances: Vec::new(),
+            contains_one_word_field: false,
+        }
+    }
+}

meilisearch-core/src/raw_indexer.rs

@@ -0,0 +1,272 @@
+use std::collections::{BTreeMap, HashMap};
+use std::convert::TryFrom;
+
+use crate::{DocIndex, DocumentId};
+use deunicode::deunicode_with_tofu;
+use meilisearch_schema::IndexedPos;
+use meilisearch_tokenizer::{is_cjk, SeqTokenizer, Token, Tokenizer};
+use sdset::SetBuf;
+
+const WORD_LENGTH_LIMIT: usize = 80;
+
+type Word = Vec<u8>; // TODO make it be a SmallVec
+
+pub struct RawIndexer {
+    word_limit: usize, // the maximum number of indexed words
+    stop_words: fst::Set,
+    words_doc_indexes: BTreeMap<Word, Vec<DocIndex>>,
+    docs_words: HashMap<DocumentId, Vec<Word>>,
+}
+
+pub struct Indexed {
+    pub words_doc_indexes: BTreeMap<Word, SetBuf<DocIndex>>,
+    pub docs_words: HashMap<DocumentId, fst::Set>,
+}
+
+impl RawIndexer {
+    pub fn new(stop_words: fst::Set) -> RawIndexer {
+        RawIndexer::with_word_limit(stop_words, 1000)
+    }
+
+    pub fn with_word_limit(stop_words: fst::Set, limit: usize) -> RawIndexer {
+        RawIndexer {
+            word_limit: limit,
+            stop_words,
+            words_doc_indexes: BTreeMap::new(),
+            docs_words: HashMap::new(),
+        }
+    }
+
+    pub fn index_text(&mut self, id: DocumentId, indexed_pos: IndexedPos, text: &str) -> usize {
+        let mut number_of_words = 0;
+
+        for token in Tokenizer::new(text) {
+            let must_continue = index_token(
+                token,
+                id,
+                indexed_pos,
+                self.word_limit,
+                &self.stop_words,
+                &mut self.words_doc_indexes,
+                &mut self.docs_words,
+            );
+
+            number_of_words += 1;
+
+            if !must_continue {
+                break;
+            }
+        }
+
+        number_of_words
+    }
+
+    pub fn index_text_seq<'a, I>(&mut self, id: DocumentId, indexed_pos: IndexedPos, iter: I)
+    where
+        I: IntoIterator<Item = &'a str>,
+    {
+        let iter = iter.into_iter();
+        for token in SeqTokenizer::new(iter) {
+            let must_continue = index_token(
+                token,
+                id,
+                indexed_pos,
+                self.word_limit,
+                &self.stop_words,
+                &mut self.words_doc_indexes,
+                &mut self.docs_words,
+            );
+
+            if !must_continue {
+                break;
+            }
+        }
+    }
+
+    pub fn build(self) -> Indexed {
+        let words_doc_indexes = self
+            .words_doc_indexes
+            .into_iter()
+            .map(|(word, indexes)| (word, SetBuf::from_dirty(indexes)))
+            .collect();
+
+        let docs_words = self
+            .docs_words
+            .into_iter()
+            .map(|(id, mut words)| {
+                words.sort_unstable();
+                words.dedup();
+                (id, fst::Set::from_iter(words).unwrap())
+            })
+            .collect();
+
+        Indexed {
+            words_doc_indexes,
+            docs_words,
+        }
+    }
+}
+
+fn index_token(
+    token: Token,
+    id: DocumentId,
+    indexed_pos: IndexedPos,
+    word_limit: usize,
+    stop_words: &fst::Set,
+    words_doc_indexes: &mut BTreeMap<Word, Vec<DocIndex>>,
+    docs_words: &mut HashMap<DocumentId, Vec<Word>>,
+) -> bool {
+    if token.word_index >= word_limit {
+        return false;
+    }
+
+    let lower = token.word.to_lowercase();
+    let token = Token {
+        word: &lower,
+        ..token
+    };
+
+    if !stop_words.contains(&token.word) {
+        match token_to_docindex(id, indexed_pos, token) {
+            Some(docindex) => {
+                let word = Vec::from(token.word);
+
+                if word.len() <= WORD_LENGTH_LIMIT {
+                    words_doc_indexes
+                        .entry(word.clone())
+                        .or_insert_with(Vec::new)
+                        .push(docindex);
+                    docs_words.entry(id).or_insert_with(Vec::new).push(word);
+
+                    if !lower.contains(is_cjk) {
+                        let unidecoded = deunicode_with_tofu(&lower, "");
+                        if unidecoded != lower && !unidecoded.is_empty() {
+                            let word = Vec::from(unidecoded);
+                            if word.len() <= WORD_LENGTH_LIMIT {
+                                words_doc_indexes
+                                    .entry(word.clone())
+                                    .or_insert_with(Vec::new)
+                                    .push(docindex);
+                                docs_words.entry(id).or_insert_with(Vec::new).push(word);
+                            }
+                        }
+                    }
+                }
+            }
+            None => return false,
+        }
+    }
+
+    true
+}
+
+fn token_to_docindex(id: DocumentId, indexed_pos: IndexedPos, token: Token) -> Option<DocIndex> {
+    let word_index = u16::try_from(token.word_index).ok()?;
+    let char_index = u16::try_from(token.char_index).ok()?;
+    let char_length = u16::try_from(token.word.chars().count()).ok()?;
+
+    let docindex = DocIndex {
+        document_id: id,
+        attribute: indexed_pos.0,
+        word_index,
+        char_index,
+        char_length,
+    };
+
+    Some(docindex)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use meilisearch_schema::IndexedPos;
+
+    #[test]
+    fn strange_apostrophe() {
+        let mut indexer = RawIndexer::new(fst::Set::default());
+
+        let docid = DocumentId(0);
+        let indexed_pos = IndexedPos(0);
+        let text = "Zut, l’aspirateur, j’ai oublié de l’éteindre !";
+        indexer.index_text(docid, indexed_pos, text);
+
+        let Indexed {
+            words_doc_indexes, ..
+        } = indexer.build();
+
+        assert!(words_doc_indexes.get(&b"l"[..]).is_some());
+        assert!(words_doc_indexes.get(&b"aspirateur"[..]).is_some());
+        assert!(words_doc_indexes.get(&b"ai"[..]).is_some());
+        assert!(words_doc_indexes.get(&b"eteindre"[..]).is_some());
+        assert!(words_doc_indexes
+            .get(&"éteindre".to_owned().into_bytes())
+            .is_some());
+    }
+
+    #[test]
+    fn strange_apostrophe_in_sequence() {
+        let mut indexer = RawIndexer::new(fst::Set::default());
+
+        let docid = DocumentId(0);
+        let indexed_pos = IndexedPos(0);
+        let text = vec!["Zut, l’aspirateur, j’ai oublié de l’éteindre !"];
+        indexer.index_text_seq(docid, indexed_pos, text);
+
+        let Indexed {
+            words_doc_indexes, ..
+        } = indexer.build();
+
+        assert!(words_doc_indexes.get(&b"l"[..]).is_some());
+        assert!(words_doc_indexes.get(&b"aspirateur"[..]).is_some());
+        assert!(words_doc_indexes.get(&b"ai"[..]).is_some());
+        assert!(words_doc_indexes.get(&b"eteindre"[..]).is_some());
+        assert!(words_doc_indexes
+            .get(&"éteindre".to_owned().into_bytes())
+            .is_some());
+    }
+
+    #[test]
+    fn basic_stop_words() {
+        let stop_words = sdset::SetBuf::from_dirty(vec!["l", "j", "ai", "de"]);
+        let stop_words = fst::Set::from_iter(stop_words).unwrap();
+
+        let mut indexer = RawIndexer::new(stop_words);
+
+        let docid = DocumentId(0);
+        let indexed_pos = IndexedPos(0);
+        let text = "Zut, l’aspirateur, j’ai oublié de l’éteindre !";
+        indexer.index_text(docid, indexed_pos, text);
+
+        let Indexed {
+            words_doc_indexes, ..
+        } = indexer.build();
+
+        assert!(words_doc_indexes.get(&b"l"[..]).is_none());
+        assert!(words_doc_indexes.get(&b"aspirateur"[..]).is_some());
+        assert!(words_doc_indexes.get(&b"j"[..]).is_none());
+        assert!(words_doc_indexes.get(&b"ai"[..]).is_none());
+        assert!(words_doc_indexes.get(&b"de"[..]).is_none());
+        assert!(words_doc_indexes.get(&b"eteindre"[..]).is_some());
+        assert!(words_doc_indexes
+            .get(&"éteindre".to_owned().into_bytes())
+            .is_some());
+    }
+
+    #[test]
+    fn no_empty_unidecode() {
+        let mut indexer = RawIndexer::new(fst::Set::default());
+
+        let docid = DocumentId(0);
+        let indexed_pos = IndexedPos(0);
+        let text = "🇯🇵";
+        indexer.index_text(docid, indexed_pos, text);
+
+        let Indexed {
+            words_doc_indexes, ..
+        } = indexer.build();
+
+        assert!(words_doc_indexes
+            .get(&"🇯🇵".to_owned().into_bytes())
+            .is_some());
+    }
+}

meilisearch-core/src/reordered_attrs.rs

@@ -0,0 +1,31 @@
+use std::cmp;
+
+#[derive(Default, Clone)]
+pub struct ReorderedAttrs {
+    reorders: Vec<Option<u16>>,
+    reverse: Vec<u16>,
+}
+
+impl ReorderedAttrs {
+    pub fn new() -> ReorderedAttrs {
+        ReorderedAttrs { reorders: Vec::new(), reverse: Vec::new() }
+    }
+
+    pub fn insert_attribute(&mut self, attribute: u16) {
+        let new_len = cmp::max(attribute as usize + 1, self.reorders.len());
+        self.reorders.resize(new_len, None);
+        self.reorders[attribute as usize] = Some(self.reverse.len() as u16);
+        self.reverse.push(attribute);
+    }
+
+    pub fn get(&self, attribute: u16) -> Option<u16> {
+        match self.reorders.get(attribute as usize)? {
+            Some(attribute) => Some(*attribute),
+            None => None,
+        }
+    }
+
+    pub fn reverse(&self, attribute: u16) -> Option<u16> {
+        self.reverse.get(attribute as usize).copied()
+    }
+}

meilisearch-core/src/serde/convert_to_number.rs

@@ -0,0 +1,198 @@
+use std::str::FromStr;
+
+use ordered_float::OrderedFloat;
+use serde::ser;
+use serde::Serialize;
+
+use super::SerializerError;
+use crate::Number;
+
+pub struct ConvertToNumber;
+
+impl ser::Serializer for ConvertToNumber {
+    type Ok = Number;
+    type Error = SerializerError;
+    type SerializeSeq = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTuple = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleStruct = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleVariant = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeMap = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeStruct = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeStructVariant = ser::Impossible<Self::Ok, Self::Error>;
+
+    fn serialize_bool(self, value: bool) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Unsigned(u64::from(value)))
+    }
+
+    fn serialize_char(self, _value: char) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnrankableType { type_name: "char" })
+    }
+
+    fn serialize_i8(self, value: i8) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Signed(i64::from(value)))
+    }
+
+    fn serialize_i16(self, value: i16) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Signed(i64::from(value)))
+    }
+
+    fn serialize_i32(self, value: i32) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Signed(i64::from(value)))
+    }
+
+    fn serialize_i64(self, value: i64) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Signed(value))
+    }
+
+    fn serialize_u8(self, value: u8) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Unsigned(u64::from(value)))
+    }
+
+    fn serialize_u16(self, value: u16) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Unsigned(u64::from(value)))
+    }
+
+    fn serialize_u32(self, value: u32) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Unsigned(u64::from(value)))
+    }
+
+    fn serialize_u64(self, value: u64) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Unsigned(value))
+    }
+
+    fn serialize_f32(self, value: f32) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Float(OrderedFloat(f64::from(value))))
+    }
+
+    fn serialize_f64(self, value: f64) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::Float(OrderedFloat(value)))
+    }
+
+    fn serialize_str(self, value: &str) -> Result<Self::Ok, Self::Error> {
+        Ok(Number::from_str(value)?)
+    }
+
+    fn serialize_bytes(self, _v: &[u8]) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnrankableType { type_name: "&[u8]" })
+    }
+
+    fn serialize_none(self) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_some<T: ?Sized>(self, _value: &T) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        Err(SerializerError::UnrankableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_unit(self) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnrankableType { type_name: "()" })
+    }
+
+    fn serialize_unit_struct(self, _name: &'static str) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "unit struct",
+        })
+    }
+
+    fn serialize_unit_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+    ) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "unit variant",
+        })
+    }
+
+    fn serialize_newtype_struct<T: ?Sized>(
+        self,
+        _name: &'static str,
+        value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        value.serialize(self)
+    }
+
+    fn serialize_newtype_variant<T: ?Sized>(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        Err(SerializerError::UnrankableType {
+            type_name: "newtype variant",
+        })
+    }
+
+    fn serialize_seq(self, _len: Option<usize>) -> Result<Self::SerializeSeq, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "sequence",
+        })
+    }
+
+    fn serialize_tuple(self, _len: usize) -> Result<Self::SerializeTuple, Self::Error> {
+        Err(SerializerError::UnrankableType { type_name: "tuple" })
+    }
+
+    fn serialize_tuple_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleStruct, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "tuple struct",
+        })
+    }
+
+    fn serialize_tuple_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleVariant, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "tuple variant",
+        })
+    }
+
+    fn serialize_map(self, _len: Option<usize>) -> Result<Self::SerializeMap, Self::Error> {
+        Err(SerializerError::UnrankableType { type_name: "map" })
+    }
+
+    fn serialize_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStruct, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "struct",
+        })
+    }
+
+    fn serialize_struct_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStructVariant, Self::Error> {
+        Err(SerializerError::UnrankableType {
+            type_name: "struct variant",
+        })
+    }
+}

meilisearch-core/src/serde/convert_to_string.rs

@@ -0,0 +1,279 @@
+use serde::ser;
+use serde::Serialize;
+
+use super::SerializerError;
+
+pub struct ConvertToString;
+
+impl ser::Serializer for ConvertToString {
+    type Ok = String;
+    type Error = SerializerError;
+    type SerializeSeq = SeqConvertToString;
+    type SerializeTuple = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleStruct = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleVariant = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeMap = MapConvertToString;
+    type SerializeStruct = StructConvertToString;
+    type SerializeStructVariant = ser::Impossible<Self::Ok, Self::Error>;
+
+    fn serialize_bool(self, value: bool) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_char(self, value: char) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_i8(self, value: i8) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_i16(self, value: i16) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_i32(self, value: i32) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_i64(self, value: i64) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_u8(self, value: u8) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_u16(self, value: u16) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_u32(self, value: u32) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_u64(self, value: u64) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_f32(self, value: f32) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_f64(self, value: f64) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_str(self, value: &str) -> Result<Self::Ok, Self::Error> {
+        Ok(value.to_string())
+    }
+
+    fn serialize_bytes(self, _v: &[u8]) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "&[u8]" })
+    }
+
+    fn serialize_none(self) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_some<T: ?Sized>(self, _value: &T) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        Err(SerializerError::UnserializableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_unit(self) -> Result<Self::Ok, Self::Error> {
+        Ok(String::new())
+    }
+
+    fn serialize_unit_struct(self, _name: &'static str) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "unit struct",
+        })
+    }
+
+    fn serialize_unit_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+    ) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "unit variant",
+        })
+    }
+
+    fn serialize_newtype_struct<T: ?Sized>(
+        self,
+        _name: &'static str,
+        value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        value.serialize(self)
+    }
+
+    fn serialize_newtype_variant<T: ?Sized>(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        Err(SerializerError::UnserializableType {
+            type_name: "newtype variant",
+        })
+    }
+
+    fn serialize_seq(self, _len: Option<usize>) -> Result<Self::SerializeSeq, Self::Error> {
+        Ok(SeqConvertToString {
+            text: String::new(),
+        })
+    }
+
+    fn serialize_tuple(self, _len: usize) -> Result<Self::SerializeTuple, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "tuple" })
+    }
+
+    fn serialize_tuple_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleStruct, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "tuple struct",
+        })
+    }
+
+    fn serialize_tuple_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleVariant, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "tuple variant",
+        })
+    }
+
+    fn serialize_map(self, _len: Option<usize>) -> Result<Self::SerializeMap, Self::Error> {
+        Ok(MapConvertToString {
+            text: String::new(),
+        })
+    }
+
+    fn serialize_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStruct, Self::Error> {
+        Ok(StructConvertToString {
+            text: String::new(),
+        })
+    }
+
+    fn serialize_struct_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStructVariant, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "struct variant",
+        })
+    }
+}
+
+pub struct MapConvertToString {
+    text: String,
+}
+
+impl ser::SerializeMap for MapConvertToString {
+    type Ok = String;
+    type Error = SerializerError;
+
+    fn serialize_key<T: ?Sized>(&mut self, key: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let text = key.serialize(ConvertToString)?;
+        self.text.push_str(&text);
+        self.text.push_str(" ");
+        Ok(())
+    }
+
+    fn serialize_value<T: ?Sized>(&mut self, value: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let text = value.serialize(ConvertToString)?;
+        self.text.push_str(&text);
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        Ok(self.text)
+    }
+}
+
+pub struct StructConvertToString {
+    text: String,
+}
+
+impl ser::SerializeStruct for StructConvertToString {
+    type Ok = String;
+    type Error = SerializerError;
+
+    fn serialize_field<T: ?Sized>(
+        &mut self,
+        key: &'static str,
+        value: &T,
+    ) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let value = value.serialize(ConvertToString)?;
+        self.text.push_str(key);
+        self.text.push_str(" ");
+        self.text.push_str(&value);
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        Ok(self.text)
+    }
+}
+
+pub struct SeqConvertToString {
+    text: String,
+}
+
+impl ser::SerializeSeq for SeqConvertToString {
+    type Ok = String;
+    type Error = SerializerError;
+
+    fn serialize_element<T: ?Sized>(&mut self, key: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let text = key.serialize(ConvertToString)?;
+        self.text.push_str(&text);
+        self.text.push_str(" ");
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        Ok(self.text)
+    }
+}

meilisearch-core/src/serde/deserializer.rs

@@ -0,0 +1,161 @@
+use std::collections::HashSet;
+use std::io::Cursor;
+use std::{error::Error, fmt};
+
+use meilisearch_schema::{Schema, FieldId};
+use serde::{de, forward_to_deserialize_any};
+use serde_json::de::IoRead as SerdeJsonIoRead;
+use serde_json::Deserializer as SerdeJsonDeserializer;
+use serde_json::Error as SerdeJsonError;
+
+use crate::database::MainT;
+use crate::store::DocumentsFields;
+use crate::DocumentId;
+
+#[derive(Debug)]
+pub enum DeserializerError {
+    SerdeJson(SerdeJsonError),
+    Zlmdb(heed::Error),
+    Custom(String),
+}
+
+impl de::Error for DeserializerError {
+    fn custom<T: fmt::Display>(msg: T) -> Self {
+        DeserializerError::Custom(msg.to_string())
+    }
+}
+
+impl fmt::Display for DeserializerError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        match self {
+            DeserializerError::SerdeJson(e) => write!(f, "serde json related error: {}", e),
+            DeserializerError::Zlmdb(e) => write!(f, "heed related error: {}", e),
+            DeserializerError::Custom(s) => f.write_str(s),
+        }
+    }
+}
+
+impl Error for DeserializerError {}
+
+impl From<SerdeJsonError> for DeserializerError {
+    fn from(error: SerdeJsonError) -> DeserializerError {
+        DeserializerError::SerdeJson(error)
+    }
+}
+
+impl From<heed::Error> for DeserializerError {
+    fn from(error: heed::Error) -> DeserializerError {
+        DeserializerError::Zlmdb(error)
+    }
+}
+
+pub struct Deserializer<'a> {
+    pub document_id: DocumentId,
+    pub reader: &'a heed::RoTxn<MainT>,
+    pub documents_fields: DocumentsFields,
+    pub schema: &'a Schema,
+    pub fields: Option<&'a HashSet<FieldId>>,
+}
+
+impl<'de, 'a, 'b> de::Deserializer<'de> for &'b mut Deserializer<'a> {
+    type Error = DeserializerError;
+
+    fn deserialize_any<V>(self, visitor: V) -> Result<V::Value, Self::Error>
+    where
+        V: de::Visitor<'de>,
+    {
+        self.deserialize_option(visitor)
+    }
+
+    fn deserialize_option<V>(self, visitor: V) -> Result<V::Value, Self::Error>
+    where
+        V: de::Visitor<'de>,
+    {
+        self.deserialize_map(visitor)
+    }
+
+    fn deserialize_map<V>(self, visitor: V) -> Result<V::Value, Self::Error>
+    where
+        V: de::Visitor<'de>,
+    {
+        let mut error = None;
+
+        let iter = self
+            .documents_fields
+            .document_fields(self.reader, self.document_id)?
+            .filter_map(|result| {
+                let (attr, value) = match result {
+                    Ok(value) => value,
+                    Err(e) => {
+                        error = Some(e);
+                        return None;
+                    }
+                };
+
+                let is_displayed = self.schema.is_displayed(attr);
+                if is_displayed && self.fields.map_or(true, |f| f.contains(&attr)) {
+                    if let Some(attribute_name) = self.schema.name(attr) {
+                        let cursor = Cursor::new(value.to_owned());
+                        let ioread = SerdeJsonIoRead::new(cursor);
+                        let value = Value(SerdeJsonDeserializer::new(ioread));
+
+                        Some((attribute_name, value))
+                    } else {
+                        None
+                    }
+                } else {
+                    None
+                }
+            });
+
+        let mut iter = iter.peekable();
+
+        let result = match iter.peek() {
+            Some(_) => {
+                let map_deserializer = de::value::MapDeserializer::new(iter);
+                visitor
+                    .visit_some(map_deserializer)
+                    .map_err(DeserializerError::from)
+            }
+            None => visitor.visit_none(),
+        };
+
+        match error.take() {
+            Some(error) => Err(error.into()),
+            None => result,
+        }
+    }
+
+    forward_to_deserialize_any! {
+        bool i8 i16 i32 i64 i128 u8 u16 u32 u64 u128 f32 f64 char str string
+        bytes byte_buf unit unit_struct newtype_struct seq tuple
+        tuple_struct struct enum identifier ignored_any
+    }
+}
+
+struct Value(SerdeJsonDeserializer<SerdeJsonIoRead<Cursor<Vec<u8>>>>);
+
+impl<'de> de::IntoDeserializer<'de, SerdeJsonError> for Value {
+    type Deserializer = Self;
+
+    fn into_deserializer(self) -> Self::Deserializer {
+        self
+    }
+}
+
+impl<'de> de::Deserializer<'de> for Value {
+    type Error = SerdeJsonError;
+
+    fn deserialize_any<V>(mut self, visitor: V) -> Result<V::Value, Self::Error>
+    where
+        V: de::Visitor<'de>,
+    {
+        self.0.deserialize_any(visitor)
+    }
+
+    forward_to_deserialize_any! {
+        bool i8 i16 i32 i64 i128 u8 u16 u32 u64 u128 f32 f64 char str string
+        bytes byte_buf option unit unit_struct newtype_struct seq tuple
+        tuple_struct map struct enum identifier ignored_any
+    }
+}

meilisearch-core/src/serde/extract_document_id.rs

@@ -0,0 +1,310 @@
+use std::hash::{Hash, Hasher};
+
+use crate::DocumentId;
+use serde::{ser, Serialize};
+use serde_json::{Value, Number};
+use siphasher::sip::SipHasher;
+
+use super::{ConvertToString, SerializerError};
+
+pub fn extract_document_id<D>(
+    primary_key: &str,
+    document: &D,
+) -> Result<Option<DocumentId>, SerializerError>
+where
+    D: serde::Serialize,
+{
+    let serializer = ExtractDocumentId { primary_key };
+    document.serialize(serializer)
+}
+
+fn validate_number(value: &Number) -> Option<String> {
+    if value.is_f64() {
+        return None
+    }
+    Some(value.to_string())
+}
+
+fn validate_string(value: &str) -> Option<String> {
+    if value.chars().all(|x| x.is_ascii_alphanumeric() || x == '-' || x == '_') {
+        Some(value.to_string())
+    } else {
+        None
+    }
+}
+
+pub fn value_to_string(value: &Value) -> Option<String> {
+    match value {
+        Value::Null => None,
+        Value::Bool(_) => None,
+        Value::Number(value) => validate_number(value),
+        Value::String(value) => validate_string(value),
+        Value::Array(_) => None,
+        Value::Object(_) => None,
+    }
+}
+
+pub fn compute_document_id<H: Hash>(t: H) -> DocumentId {
+    let mut s = SipHasher::new();
+    t.hash(&mut s);
+    let hash = s.finish();
+    DocumentId(hash)
+}
+
+struct ExtractDocumentId<'a> {
+    primary_key: &'a str,
+}
+
+impl<'a> ser::Serializer for ExtractDocumentId<'a> {
+    type Ok = Option<DocumentId>;
+    type Error = SerializerError;
+    type SerializeSeq = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTuple = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleStruct = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleVariant = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeMap = ExtractDocumentIdMapSerializer<'a>;
+    type SerializeStruct = ExtractDocumentIdStructSerializer<'a>;
+    type SerializeStructVariant = ser::Impossible<Self::Ok, Self::Error>;
+
+    forward_to_unserializable_type! {
+        bool => serialize_bool,
+        char => serialize_char,
+
+        i8  => serialize_i8,
+        i16 => serialize_i16,
+        i32 => serialize_i32,
+        i64 => serialize_i64,
+
+        u8  => serialize_u8,
+        u16 => serialize_u16,
+        u32 => serialize_u32,
+        u64 => serialize_u64,
+
+        f32 => serialize_f32,
+        f64 => serialize_f64,
+    }
+
+    fn serialize_str(self, _value: &str) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "str" })
+    }
+
+    fn serialize_bytes(self, _value: &[u8]) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "&[u8]" })
+    }
+
+    fn serialize_none(self) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_some<T: ?Sized>(self, _value: &T) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        Err(SerializerError::UnserializableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_unit(self) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "()" })
+    }
+
+    fn serialize_unit_struct(self, _name: &'static str) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "unit struct",
+        })
+    }
+
+    fn serialize_unit_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+    ) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "unit variant",
+        })
+    }
+
+    fn serialize_newtype_struct<T: ?Sized>(
+        self,
+        _name: &'static str,
+        value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        value.serialize(self)
+    }
+
+    fn serialize_newtype_variant<T: ?Sized>(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: Serialize,
+    {
+        Err(SerializerError::UnserializableType {
+            type_name: "newtype variant",
+        })
+    }
+
+    fn serialize_seq(self, _len: Option<usize>) -> Result<Self::SerializeSeq, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "sequence",
+        })
+    }
+
+    fn serialize_tuple(self, _len: usize) -> Result<Self::SerializeTuple, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "tuple" })
+    }
+
+    fn serialize_tuple_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleStruct, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "tuple struct",
+        })
+    }
+
+    fn serialize_tuple_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleVariant, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "tuple variant",
+        })
+    }
+
+    fn serialize_map(self, _len: Option<usize>) -> Result<Self::SerializeMap, Self::Error> {
+        let serializer = ExtractDocumentIdMapSerializer {
+            primary_key: self.primary_key,
+            document_id: None,
+            current_key_name: None,
+        };
+
+        Ok(serializer)
+    }
+
+    fn serialize_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStruct, Self::Error> {
+        let serializer = ExtractDocumentIdStructSerializer {
+            primary_key: self.primary_key,
+            document_id: None,
+        };
+
+        Ok(serializer)
+    }
+
+    fn serialize_struct_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStructVariant, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "struct variant",
+        })
+    }
+}
+
+pub struct ExtractDocumentIdMapSerializer<'a> {
+    primary_key: &'a str,
+    document_id: Option<DocumentId>,
+    current_key_name: Option<String>,
+}
+
+impl<'a> ser::SerializeMap for ExtractDocumentIdMapSerializer<'a> {
+    type Ok = Option<DocumentId>;
+    type Error = SerializerError;
+
+    fn serialize_key<T: ?Sized>(&mut self, key: &T) -> Result<(), Self::Error>
+    where
+        T: Serialize,
+    {
+        let key = key.serialize(ConvertToString)?;
+        self.current_key_name = Some(key);
+        Ok(())
+    }
+
+    fn serialize_value<T: ?Sized>(&mut self, value: &T) -> Result<(), Self::Error>
+    where
+        T: Serialize,
+    {
+        let key = self.current_key_name.take().unwrap();
+        self.serialize_entry(&key, value)
+    }
+
+    fn serialize_entry<K: ?Sized, V: ?Sized>(
+        &mut self,
+        key: &K,
+        value: &V,
+    ) -> Result<(), Self::Error>
+    where
+        K: Serialize,
+        V: Serialize,
+    {
+        let key = key.serialize(ConvertToString)?;
+
+        if self.primary_key == key {
+            let value = serde_json::to_string(value).and_then(|s| serde_json::from_str(&s))?;
+            match value_to_string(&value).map(|s| compute_document_id(&s)) {
+                Some(document_id) => self.document_id = Some(document_id),
+                None => return Err(SerializerError::InvalidDocumentIdType),
+            }
+        }
+
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        Ok(self.document_id)
+    }
+}
+
+pub struct ExtractDocumentIdStructSerializer<'a> {
+    primary_key: &'a str,
+    document_id: Option<DocumentId>,
+}
+
+impl<'a> ser::SerializeStruct for ExtractDocumentIdStructSerializer<'a> {
+    type Ok = Option<DocumentId>;
+    type Error = SerializerError;
+
+    fn serialize_field<T: ?Sized>(
+        &mut self,
+        key: &'static str,
+        value: &T,
+    ) -> Result<(), Self::Error>
+    where
+        T: Serialize,
+    {
+        if self.primary_key == key {
+            let value = serde_json::to_string(value).and_then(|s| serde_json::from_str(&s))?;
+            match value_to_string(&value).map(compute_document_id) {
+                Some(document_id) => self.document_id = Some(document_id),
+                None => return Err(SerializerError::InvalidDocumentIdType),
+            }
+        }
+
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        Ok(self.document_id)
+    }
+}

meilisearch-core/src/serde/indexer.rs

@@ -0,0 +1,362 @@
+use meilisearch_schema::IndexedPos;
+use serde::ser;
+use serde::Serialize;
+
+use super::{ConvertToString, SerializerError};
+use crate::raw_indexer::RawIndexer;
+use crate::DocumentId;
+
+pub struct Indexer<'a> {
+    pub pos: IndexedPos,
+    pub indexer: &'a mut RawIndexer,
+    pub document_id: DocumentId,
+}
+
+impl<'a> ser::Serializer for Indexer<'a> {
+    type Ok = Option<usize>;
+    type Error = SerializerError;
+    type SerializeSeq = SeqIndexer<'a>;
+    type SerializeTuple = TupleIndexer<'a>;
+    type SerializeTupleStruct = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleVariant = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeMap = MapIndexer<'a>;
+    type SerializeStruct = StructIndexer<'a>;
+    type SerializeStructVariant = ser::Impossible<Self::Ok, Self::Error>;
+
+    fn serialize_bool(self, _value: bool) -> Result<Self::Ok, Self::Error> {
+        Ok(None)
+    }
+
+    fn serialize_char(self, value: char) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_i8(self, value: i8) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_i16(self, value: i16) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_i32(self, value: i32) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_i64(self, value: i64) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_u8(self, value: u8) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_u16(self, value: u16) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_u32(self, value: u32) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_u64(self, value: u64) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_f32(self, value: f32) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_f64(self, value: f64) -> Result<Self::Ok, Self::Error> {
+        let text = value.serialize(ConvertToString)?;
+        self.serialize_str(&text)
+    }
+
+    fn serialize_str(self, text: &str) -> Result<Self::Ok, Self::Error> {
+        let number_of_words = self
+            .indexer
+            .index_text(self.document_id, self.pos, text);
+        Ok(Some(number_of_words))
+    }
+
+    fn serialize_bytes(self, _v: &[u8]) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnindexableType { type_name: "&[u8]" })
+    }
+
+    fn serialize_none(self) -> Result<Self::Ok, Self::Error> {
+        Ok(None)
+    }
+
+    fn serialize_some<T: ?Sized>(self, value: &T) -> Result<Self::Ok, Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let text = value.serialize(ConvertToString)?;
+        let number_of_words = self
+            .indexer
+            .index_text(self.document_id, self.pos, &text);
+        Ok(Some(number_of_words))
+    }
+
+    fn serialize_unit(self) -> Result<Self::Ok, Self::Error> {
+        Ok(None)
+    }
+
+    fn serialize_unit_struct(self, _name: &'static str) -> Result<Self::Ok, Self::Error> {
+        Ok(None)
+    }
+
+    fn serialize_unit_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+    ) -> Result<Self::Ok, Self::Error> {
+        Ok(None)
+    }
+
+    fn serialize_newtype_struct<T: ?Sized>(
+        self,
+        _name: &'static str,
+        value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        value.serialize(self)
+    }
+
+    fn serialize_newtype_variant<T: ?Sized>(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        Err(SerializerError::UnindexableType {
+            type_name: "newtype variant",
+        })
+    }
+
+    fn serialize_seq(self, _len: Option<usize>) -> Result<Self::SerializeSeq, Self::Error> {
+        let indexer = SeqIndexer {
+            pos: self.pos,
+            document_id: self.document_id,
+            indexer: self.indexer,
+            texts: Vec::new(),
+        };
+
+        Ok(indexer)
+    }
+
+    fn serialize_tuple(self, _len: usize) -> Result<Self::SerializeTuple, Self::Error> {
+        let indexer = TupleIndexer {
+            pos: self.pos,
+            document_id: self.document_id,
+            indexer: self.indexer,
+            texts: Vec::new(),
+        };
+
+        Ok(indexer)
+    }
+
+    fn serialize_tuple_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleStruct, Self::Error> {
+        Err(SerializerError::UnindexableType {
+            type_name: "tuple struct",
+        })
+    }
+
+    fn serialize_tuple_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleVariant, Self::Error> {
+        Err(SerializerError::UnindexableType {
+            type_name: "tuple variant",
+        })
+    }
+
+    fn serialize_map(self, _len: Option<usize>) -> Result<Self::SerializeMap, Self::Error> {
+        let indexer = MapIndexer {
+            pos: self.pos,
+            document_id: self.document_id,
+            indexer: self.indexer,
+            texts: Vec::new(),
+        };
+
+        Ok(indexer)
+    }
+
+    fn serialize_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStruct, Self::Error> {
+        let indexer = StructIndexer {
+            pos: self.pos,
+            document_id: self.document_id,
+            indexer: self.indexer,
+            texts: Vec::new(),
+        };
+
+        Ok(indexer)
+    }
+
+    fn serialize_struct_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStructVariant, Self::Error> {
+        Err(SerializerError::UnindexableType {
+            type_name: "struct variant",
+        })
+    }
+}
+
+pub struct SeqIndexer<'a> {
+    pos: IndexedPos,
+    document_id: DocumentId,
+    indexer: &'a mut RawIndexer,
+    texts: Vec<String>,
+}
+
+impl<'a> ser::SerializeSeq for SeqIndexer<'a> {
+    type Ok = Option<usize>;
+    type Error = SerializerError;
+
+    fn serialize_element<T: ?Sized>(&mut self, value: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let text = value.serialize(ConvertToString)?;
+        self.texts.push(text);
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        let texts = self.texts.iter().map(String::as_str);
+        self.indexer
+            .index_text_seq(self.document_id, self.pos, texts);
+        Ok(None)
+    }
+}
+
+pub struct MapIndexer<'a> {
+    pos: IndexedPos,
+    document_id: DocumentId,
+    indexer: &'a mut RawIndexer,
+    texts: Vec<String>,
+}
+
+impl<'a> ser::SerializeMap for MapIndexer<'a> {
+    type Ok = Option<usize>;
+    type Error = SerializerError;
+
+    fn serialize_key<T: ?Sized>(&mut self, key: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let text = key.serialize(ConvertToString)?;
+        self.texts.push(text);
+        Ok(())
+    }
+
+    fn serialize_value<T: ?Sized>(&mut self, value: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let text = value.serialize(ConvertToString)?;
+        self.texts.push(text);
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        let texts = self.texts.iter().map(String::as_str);
+        self.indexer
+            .index_text_seq(self.document_id, self.pos, texts);
+        Ok(None)
+    }
+}
+
+pub struct StructIndexer<'a> {
+    pos: IndexedPos,
+    document_id: DocumentId,
+    indexer: &'a mut RawIndexer,
+    texts: Vec<String>,
+}
+
+impl<'a> ser::SerializeStruct for StructIndexer<'a> {
+    type Ok = Option<usize>;
+    type Error = SerializerError;
+
+    fn serialize_field<T: ?Sized>(
+        &mut self,
+        key: &'static str,
+        value: &T,
+    ) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let key_text = key.to_owned();
+        let value_text = value.serialize(ConvertToString)?;
+        self.texts.push(key_text);
+        self.texts.push(value_text);
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        let texts = self.texts.iter().map(String::as_str);
+        self.indexer
+            .index_text_seq(self.document_id, self.pos, texts);
+        Ok(None)
+    }
+}
+
+pub struct TupleIndexer<'a> {
+    pos: IndexedPos,
+    document_id: DocumentId,
+    indexer: &'a mut RawIndexer,
+    texts: Vec<String>,
+}
+
+impl<'a> ser::SerializeTuple for TupleIndexer<'a> {
+    type Ok = Option<usize>;
+    type Error = SerializerError;
+
+    fn serialize_element<T: ?Sized>(&mut self, value: &T) -> Result<(), Self::Error>
+    where
+        T: Serialize,
+    {
+        let text = value.serialize(ConvertToString)?;
+        self.texts.push(text);
+        Ok(())
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        let texts = self.texts.iter().map(String::as_str);
+        self.indexer
+            .index_text_seq(self.document_id, self.pos, texts);
+        Ok(None)
+    }
+}

meilisearch-core/src/serde/mod.rs

@@ -0,0 +1,112 @@
+macro_rules! forward_to_unserializable_type {
+    ($($ty:ident => $se_method:ident,)*) => {
+        $(
+            fn $se_method(self, _v: $ty) -> Result<Self::Ok, Self::Error> {
+                Err(SerializerError::UnserializableType { type_name: "$ty" })
+            }
+        )*
+    }
+}
+
+mod convert_to_number;
+mod convert_to_string;
+mod deserializer;
+mod extract_document_id;
+mod indexer;
+mod serializer;
+
+pub use self::convert_to_number::ConvertToNumber;
+pub use self::convert_to_string::ConvertToString;
+pub use self::deserializer::{Deserializer, DeserializerError};
+pub use self::extract_document_id::{compute_document_id, extract_document_id, value_to_string};
+pub use self::indexer::Indexer;
+pub use self::serializer::{serialize_value, serialize_value_with_id, Serializer};
+
+use std::{error::Error, fmt};
+
+use serde::ser;
+use serde_json::Error as SerdeJsonError;
+use meilisearch_schema::Error as SchemaError;
+
+use crate::ParseNumberError;
+
+#[derive(Debug)]
+pub enum SerializerError {
+    DocumentIdNotFound,
+    InvalidDocumentIdType,
+    Zlmdb(heed::Error),
+    SerdeJson(SerdeJsonError),
+    ParseNumber(ParseNumberError),
+    Schema(SchemaError),
+    UnserializableType { type_name: &'static str },
+    UnindexableType { type_name: &'static str },
+    UnrankableType { type_name: &'static str },
+    Custom(String),
+}
+
+impl ser::Error for SerializerError {
+    fn custom<T: fmt::Display>(msg: T) -> Self {
+        SerializerError::Custom(msg.to_string())
+    }
+}
+
+impl fmt::Display for SerializerError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        match self {
+            SerializerError::DocumentIdNotFound => {
+                f.write_str("serialized document does not have an id according to the schema")
+            }
+            SerializerError::InvalidDocumentIdType => {
+                f.write_str("a document primary key can be of type integer or string only composed of alphanumeric characters, hyphens (-) and underscores (_).")
+            }
+            SerializerError::Zlmdb(e) => write!(f, "heed related error: {}", e),
+            SerializerError::SerdeJson(e) => write!(f, "serde json error: {}", e),
+            SerializerError::ParseNumber(e) => {
+                write!(f, "error while trying to parse a number: {}", e)
+            }
+            SerializerError::Schema(e) => write!(f, "impossible to update schema: {}", e),
+            SerializerError::UnserializableType { type_name } => {
+                write!(f, "{} is not a serializable type", type_name)
+            }
+            SerializerError::UnindexableType { type_name } => {
+                write!(f, "{} is not an indexable type", type_name)
+            }
+            SerializerError::UnrankableType { type_name } => {
+                write!(f, "{} types can not be used for ranking", type_name)
+            }
+            SerializerError::Custom(s) => f.write_str(s),
+        }
+    }
+}
+
+impl Error for SerializerError {}
+
+impl From<String> for SerializerError {
+    fn from(value: String) -> SerializerError {
+        SerializerError::Custom(value)
+    }
+}
+
+impl From<SerdeJsonError> for SerializerError {
+    fn from(error: SerdeJsonError) -> SerializerError {
+        SerializerError::SerdeJson(error)
+    }
+}
+
+impl From<heed::Error> for SerializerError {
+    fn from(error: heed::Error) -> SerializerError {
+        SerializerError::Zlmdb(error)
+    }
+}
+
+impl From<ParseNumberError> for SerializerError {
+    fn from(error: ParseNumberError) -> SerializerError {
+        SerializerError::ParseNumber(error)
+    }
+}
+
+impl From<SchemaError> for SerializerError {
+    fn from(error: SchemaError) -> SerializerError {
+        SerializerError::Schema(error)
+   }
+}

meilisearch-core/src/serde/serializer.rs

@@ -0,0 +1,361 @@
+use meilisearch_schema::{Schema, FieldId};
+use serde::ser;
+
+use crate::database::MainT;
+use crate::raw_indexer::RawIndexer;
+use crate::store::{DocumentsFields, DocumentsFieldsCounts};
+use crate::{DocumentId, RankedMap};
+
+use super::{ConvertToNumber, ConvertToString, Indexer, SerializerError};
+
+pub struct Serializer<'a, 'b> {
+    pub txn: &'a mut heed::RwTxn<'b, MainT>,
+    pub schema: &'a mut Schema,
+    pub document_store: DocumentsFields,
+    pub document_fields_counts: DocumentsFieldsCounts,
+    pub indexer: &'a mut RawIndexer,
+    pub ranked_map: &'a mut RankedMap,
+    pub document_id: DocumentId,
+}
+
+impl<'a, 'b> ser::Serializer for Serializer<'a, 'b> {
+    type Ok = ();
+    type Error = SerializerError;
+    type SerializeSeq = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTuple = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleStruct = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeTupleVariant = ser::Impossible<Self::Ok, Self::Error>;
+    type SerializeMap = MapSerializer<'a, 'b>;
+    type SerializeStruct = StructSerializer<'a, 'b>;
+    type SerializeStructVariant = ser::Impossible<Self::Ok, Self::Error>;
+
+    forward_to_unserializable_type! {
+        bool => serialize_bool,
+        char => serialize_char,
+
+        i8  => serialize_i8,
+        i16 => serialize_i16,
+        i32 => serialize_i32,
+        i64 => serialize_i64,
+
+        u8  => serialize_u8,
+        u16 => serialize_u16,
+        u32 => serialize_u32,
+        u64 => serialize_u64,
+
+        f32 => serialize_f32,
+        f64 => serialize_f64,
+    }
+
+    fn serialize_str(self, _v: &str) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "str" })
+    }
+
+    fn serialize_bytes(self, _v: &[u8]) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "&[u8]" })
+    }
+
+    fn serialize_none(self) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_some<T: ?Sized>(self, _value: &T) -> Result<Self::Ok, Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        Err(SerializerError::UnserializableType {
+            type_name: "Option",
+        })
+    }
+
+    fn serialize_unit(self) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "()" })
+    }
+
+    fn serialize_unit_struct(self, _name: &'static str) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "unit struct",
+        })
+    }
+
+    fn serialize_unit_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+    ) -> Result<Self::Ok, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "unit variant",
+        })
+    }
+
+    fn serialize_newtype_struct<T: ?Sized>(
+        self,
+        _name: &'static str,
+        value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        value.serialize(self)
+    }
+
+    fn serialize_newtype_variant<T: ?Sized>(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _value: &T,
+    ) -> Result<Self::Ok, Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        Err(SerializerError::UnserializableType {
+            type_name: "newtype variant",
+        })
+    }
+
+    fn serialize_seq(self, _len: Option<usize>) -> Result<Self::SerializeSeq, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "sequence",
+        })
+    }
+
+    fn serialize_tuple(self, _len: usize) -> Result<Self::SerializeTuple, Self::Error> {
+        Err(SerializerError::UnserializableType { type_name: "tuple" })
+    }
+
+    fn serialize_tuple_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleStruct, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "tuple struct",
+        })
+    }
+
+    fn serialize_tuple_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeTupleVariant, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "tuple variant",
+        })
+    }
+
+    fn serialize_map(self, _len: Option<usize>) -> Result<Self::SerializeMap, Self::Error> {
+        Ok(MapSerializer {
+            txn: self.txn,
+            schema: self.schema,
+            document_id: self.document_id,
+            document_store: self.document_store,
+            document_fields_counts: self.document_fields_counts,
+            indexer: self.indexer,
+            ranked_map: self.ranked_map,
+            current_key_name: None,
+        })
+    }
+
+    fn serialize_struct(
+        self,
+        _name: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStruct, Self::Error> {
+        Ok(StructSerializer {
+            txn: self.txn,
+            schema: self.schema,
+            document_id: self.document_id,
+            document_store: self.document_store,
+            document_fields_counts: self.document_fields_counts,
+            indexer: self.indexer,
+            ranked_map: self.ranked_map,
+        })
+    }
+
+    fn serialize_struct_variant(
+        self,
+        _name: &'static str,
+        _variant_index: u32,
+        _variant: &'static str,
+        _len: usize,
+    ) -> Result<Self::SerializeStructVariant, Self::Error> {
+        Err(SerializerError::UnserializableType {
+            type_name: "struct variant",
+        })
+    }
+}
+
+pub struct MapSerializer<'a, 'b> {
+    txn: &'a mut heed::RwTxn<'b, MainT>,
+    schema: &'a mut Schema,
+    document_id: DocumentId,
+    document_store: DocumentsFields,
+    document_fields_counts: DocumentsFieldsCounts,
+    indexer: &'a mut RawIndexer,
+    ranked_map: &'a mut RankedMap,
+    current_key_name: Option<String>,
+}
+
+impl<'a, 'b> ser::SerializeMap for MapSerializer<'a, 'b> {
+    type Ok = ();
+    type Error = SerializerError;
+
+    fn serialize_key<T: ?Sized>(&mut self, key: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let key = key.serialize(ConvertToString)?;
+        self.current_key_name = Some(key);
+        Ok(())
+    }
+
+    fn serialize_value<T: ?Sized>(&mut self, value: &T) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        let key = self.current_key_name.take().unwrap();
+        self.serialize_entry(&key, value)
+    }
+
+    fn serialize_entry<K: ?Sized, V: ?Sized>(
+        &mut self,
+        key: &K,
+        value: &V,
+    ) -> Result<(), Self::Error>
+    where
+        K: ser::Serialize,
+        V: ser::Serialize,
+    {
+        let key = key.serialize(ConvertToString)?;
+        serialize_value(
+            self.txn,
+            key.as_str(),
+            self.schema,
+            self.document_id,
+            self.document_store,
+            self.document_fields_counts,
+            self.indexer,
+            self.ranked_map,
+            value,
+        )
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        Ok(())
+    }
+}
+
+pub struct StructSerializer<'a, 'b> {
+    txn: &'a mut heed::RwTxn<'b, MainT>,
+    schema: &'a mut Schema,
+    document_id: DocumentId,
+    document_store: DocumentsFields,
+    document_fields_counts: DocumentsFieldsCounts,
+    indexer: &'a mut RawIndexer,
+    ranked_map: &'a mut RankedMap,
+}
+
+impl<'a, 'b> ser::SerializeStruct for StructSerializer<'a, 'b> {
+    type Ok = ();
+    type Error = SerializerError;
+
+    fn serialize_field<T: ?Sized>(
+        &mut self,
+        key: &'static str,
+        value: &T,
+    ) -> Result<(), Self::Error>
+    where
+        T: ser::Serialize,
+    {
+        serialize_value(
+            self.txn,
+            key,
+            self.schema,
+            self.document_id,
+            self.document_store,
+            self.document_fields_counts,
+            self.indexer,
+            self.ranked_map,
+            value,
+        )
+    }
+
+    fn end(self) -> Result<Self::Ok, Self::Error> {
+        Ok(())
+    }
+}
+
+pub fn serialize_value<'a, T: ?Sized>(
+    txn: &mut heed::RwTxn<MainT>,
+    attribute: &str,
+    schema: &'a mut Schema,
+    document_id: DocumentId,
+    document_store: DocumentsFields,
+    documents_fields_counts: DocumentsFieldsCounts,
+    indexer: &mut RawIndexer,
+    ranked_map: &mut RankedMap,
+    value: &T,
+) -> Result<(), SerializerError>
+where
+    T: ser::Serialize,
+{
+    let field_id = schema.insert_and_index(&attribute)?;
+    serialize_value_with_id(
+        txn,
+        field_id,
+        schema,
+        document_id,
+        document_store,
+        documents_fields_counts,
+        indexer,
+        ranked_map,
+        value,
+    )
+}
+
+pub fn serialize_value_with_id<'a, T: ?Sized>(
+    txn: &mut heed::RwTxn<MainT>,
+    field_id: FieldId,
+    schema: &'a Schema,
+    document_id: DocumentId,
+    document_store: DocumentsFields,
+    documents_fields_counts: DocumentsFieldsCounts,
+    indexer: &mut RawIndexer,
+    ranked_map: &mut RankedMap,
+    value: &T,
+) -> Result<(), SerializerError>
+where
+    T: ser::Serialize,
+{
+    let serialized = serde_json::to_vec(value)?;
+    document_store.put_document_field(txn, document_id, field_id, &serialized)?;
+
+    if let Some(indexed_pos) = schema.is_indexed(field_id) {
+        let indexer = Indexer {
+            pos: *indexed_pos,
+            indexer,
+            document_id,
+        };
+        if let Some(number_of_words) = value.serialize(indexer)? {
+            documents_fields_counts.put_document_field_count(
+                txn,
+                document_id,
+                *indexed_pos,
+                number_of_words as u16,
+            )?;
+        }
+    }
+
+    if schema.is_ranked(field_id) {
+        let number = value.serialize(ConvertToNumber).unwrap_or_default();
+        ranked_map.insert(document_id, field_id, number);
+    }
+
+    Ok(())
+}

meilisearch-core/src/settings.rs

@@ -0,0 +1,184 @@
+use std::collections::{BTreeMap, BTreeSet, HashSet};
+use std::str::FromStr;
+use std::iter::IntoIterator;
+
+use serde::{Deserialize, Deserializer, Serialize};
+use once_cell::sync::Lazy;
+
+use self::RankingRule::*;
+
+pub const DEFAULT_RANKING_RULES: [RankingRule; 6] = [Typo, Words, Proximity, Attribute, WordsPosition, Exactness];
+
+static RANKING_RULE_REGEX: Lazy<regex::Regex> = Lazy::new(|| {
+    let regex = regex::Regex::new(r"(asc|desc)\(([a-zA-Z0-9-_]*)\)").unwrap();
+    regex
+});
+
+#[derive(Default, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase", deny_unknown_fields)]
+pub struct Settings {
+    #[serde(default, deserialize_with = "deserialize_some")]
+    pub ranking_rules: Option<Option<Vec<String>>>,
+    #[serde(default, deserialize_with = "deserialize_some")]
+    pub distinct_attribute: Option<Option<String>>,
+    #[serde(default, deserialize_with = "deserialize_some")]
+    pub searchable_attributes: Option<Option<Vec<String>>>,
+    #[serde(default, deserialize_with = "deserialize_some")]
+    pub displayed_attributes: Option<Option<HashSet<String>>>,
+    #[serde(default, deserialize_with = "deserialize_some")]
+    pub stop_words: Option<Option<BTreeSet<String>>>,
+    #[serde(default, deserialize_with = "deserialize_some")]
+    pub synonyms: Option<Option<BTreeMap<String, Vec<String>>>>,
+    #[serde(default, deserialize_with = "deserialize_some")]
+    pub accept_new_fields: Option<Option<bool>>,
+}
+
+// Any value that is present is considered Some value, including null.
+fn deserialize_some<'de, T, D>(deserializer: D) -> Result<Option<T>, D::Error>
+    where T: Deserialize<'de>,
+          D: Deserializer<'de>
+{
+    Deserialize::deserialize(deserializer).map(Some)
+}
+
+impl Settings {
+    pub fn into_update(&self) -> Result<SettingsUpdate, RankingRuleConversionError> {
+        let settings = self.clone();
+
+        let ranking_rules = match settings.ranking_rules {
+            Some(Some(rules)) => UpdateState::Update(RankingRule::from_iter(rules.iter())?),
+            Some(None) => UpdateState::Clear,
+            None => UpdateState::Nothing,
+        };
+
+        Ok(SettingsUpdate {
+            ranking_rules,
+            distinct_attribute: settings.distinct_attribute.into(),
+            primary_key: UpdateState::Nothing,
+            searchable_attributes: settings.searchable_attributes.into(),
+            displayed_attributes: settings.displayed_attributes.into(),
+            stop_words: settings.stop_words.into(),
+            synonyms: settings.synonyms.into(),
+            accept_new_fields: settings.accept_new_fields.into(),
+        })
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub enum UpdateState<T> {
+    Update(T),
+    Clear,
+    Nothing,
+}
+
+impl <T> From<Option<Option<T>>> for UpdateState<T> {
+    fn from(opt: Option<Option<T>>) -> UpdateState<T> {
+        match opt {
+            Some(Some(t)) => UpdateState::Update(t),
+            Some(None) => UpdateState::Clear,
+            None => UpdateState::Nothing,
+        }
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct RankingRuleConversionError;
+
+impl std::fmt::Display for RankingRuleConversionError {
+    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        write!(f, "impossible to convert into RankingRule")
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub enum RankingRule {
+    Typo,
+    Words,
+    Proximity,
+    Attribute,
+    WordsPosition,
+    Exactness,
+    Asc(String),
+    Desc(String),
+}
+
+impl std::fmt::Display for RankingRule {
+    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        match self {
+            RankingRule::Typo => f.write_str("typo"),
+            RankingRule::Words => f.write_str("words"),
+            RankingRule::Proximity => f.write_str("proximity"),
+            RankingRule::Attribute => f.write_str("attribute"),
+            RankingRule::WordsPosition => f.write_str("wordsPosition"),
+            RankingRule::Exactness => f.write_str("exactness"),
+            RankingRule::Asc(field) => write!(f, "asc({})", field),
+            RankingRule::Desc(field) => write!(f, "desc({})", field),
+        }
+    }
+}
+
+impl FromStr for RankingRule {
+    type Err = RankingRuleConversionError;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let rule = match s {
+            "typo" => RankingRule::Typo,
+            "words" => RankingRule::Words,
+            "proximity" => RankingRule::Proximity,
+            "attribute" => RankingRule::Attribute,
+            "wordsPosition" => RankingRule::WordsPosition,
+            "exactness" => RankingRule::Exactness,
+            _ => {
+                let captures = RANKING_RULE_REGEX.captures(s).ok_or(RankingRuleConversionError)?;
+                match (captures.get(1).map(|m| m.as_str()), captures.get(2)) {
+                    (Some("asc"), Some(field)) => RankingRule::Asc(field.as_str().to_string()),
+                    (Some("desc"), Some(field)) => RankingRule::Desc(field.as_str().to_string()),
+                    _ => return Err(RankingRuleConversionError)
+                }
+            }
+        };
+        Ok(rule)
+    }
+}
+
+impl RankingRule {
+    pub fn field(&self) -> Option<&str> {
+        match self {
+            RankingRule::Asc(field) | RankingRule::Desc(field) => Some(field),
+            _ => None,
+        }
+    }
+
+    pub fn from_iter(rules: impl IntoIterator<Item = impl AsRef<str>>) -> Result<Vec<RankingRule>, RankingRuleConversionError> {
+        rules.into_iter()
+            .map(|s| RankingRule::from_str(s.as_ref()))
+            .collect()
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct SettingsUpdate {
+    pub ranking_rules: UpdateState<Vec<RankingRule>>,
+    pub distinct_attribute: UpdateState<String>,
+    pub primary_key: UpdateState<String>,
+    pub searchable_attributes: UpdateState<Vec<String>>,
+    pub displayed_attributes: UpdateState<HashSet<String>>,
+    pub stop_words: UpdateState<BTreeSet<String>>,
+    pub synonyms: UpdateState<BTreeMap<String, Vec<String>>>,
+    pub accept_new_fields: UpdateState<bool>,
+}
+
+impl Default for SettingsUpdate {
+    fn default() -> Self {
+        Self {
+            ranking_rules: UpdateState::Nothing,
+            distinct_attribute: UpdateState::Nothing,
+            primary_key: UpdateState::Nothing,
+            searchable_attributes: UpdateState::Nothing,
+            displayed_attributes: UpdateState::Nothing,
+            stop_words: UpdateState::Nothing,
+            synonyms: UpdateState::Nothing,
+            accept_new_fields: UpdateState::Nothing,
+        }
+    }
+}

meilisearch-core/src/store/docs_words.rs

@@ -0,0 +1,50 @@
+use super::BEU64;
+use crate::database::MainT;
+use crate::DocumentId;
+use heed::types::{ByteSlice, OwnedType};
+use heed::Result as ZResult;
+use std::sync::Arc;
+
+#[derive(Copy, Clone)]
+pub struct DocsWords {
+    pub(crate) docs_words: heed::Database<OwnedType<BEU64>, ByteSlice>,
+}
+
+impl DocsWords {
+    pub fn put_doc_words(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        document_id: DocumentId,
+        words: &fst::Set,
+    ) -> ZResult<()> {
+        let document_id = BEU64::new(document_id.0);
+        let bytes = words.as_fst().as_bytes();
+        self.docs_words.put(writer, &document_id, bytes)
+    }
+
+    pub fn del_doc_words(self, writer: &mut heed::RwTxn<MainT>, document_id: DocumentId) -> ZResult<bool> {
+        let document_id = BEU64::new(document_id.0);
+        self.docs_words.delete(writer, &document_id)
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.docs_words.clear(writer)
+    }
+
+    pub fn doc_words(
+        self,
+        reader: &heed::RoTxn<MainT>,
+        document_id: DocumentId,
+    ) -> ZResult<Option<fst::Set>> {
+        let document_id = BEU64::new(document_id.0);
+        match self.docs_words.get(reader, &document_id)? {
+            Some(bytes) => {
+                let len = bytes.len();
+                let bytes = Arc::new(bytes.to_owned());
+                let fst = fst::raw::Fst::from_shared_bytes(bytes, 0, len).unwrap();
+                Ok(Some(fst::Set::from(fst)))
+            }
+            None => Ok(None),
+        }
+    }
+}

meilisearch-core/src/store/documents_fields.rs

@@ -0,0 +1,79 @@
+use heed::types::{ByteSlice, OwnedType};
+use crate::database::MainT;
+use heed::Result as ZResult;
+use meilisearch_schema::FieldId;
+
+use super::DocumentFieldStoredKey;
+use crate::DocumentId;
+
+#[derive(Copy, Clone)]
+pub struct DocumentsFields {
+    pub(crate) documents_fields: heed::Database<OwnedType<DocumentFieldStoredKey>, ByteSlice>,
+}
+
+impl DocumentsFields {
+    pub fn put_document_field(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        document_id: DocumentId,
+        field: FieldId,
+        value: &[u8],
+    ) -> ZResult<()> {
+        let key = DocumentFieldStoredKey::new(document_id, field);
+        self.documents_fields.put(writer, &key, value)
+    }
+
+    pub fn del_all_document_fields(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        document_id: DocumentId,
+    ) -> ZResult<usize> {
+        let start = DocumentFieldStoredKey::new(document_id, FieldId::min());
+        let end = DocumentFieldStoredKey::new(document_id, FieldId::max());
+        self.documents_fields.delete_range(writer, &(start..=end))
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.documents_fields.clear(writer)
+    }
+
+    pub fn document_attribute<'txn>(
+        self,
+        reader: &'txn heed::RoTxn<MainT>,
+        document_id: DocumentId,
+        field: FieldId,
+    ) -> ZResult<Option<&'txn [u8]>> {
+        let key = DocumentFieldStoredKey::new(document_id, field);
+        self.documents_fields.get(reader, &key)
+    }
+
+    pub fn document_fields<'txn>(
+        self,
+        reader: &'txn heed::RoTxn<MainT>,
+        document_id: DocumentId,
+    ) -> ZResult<DocumentFieldsIter<'txn>> {
+        let start = DocumentFieldStoredKey::new(document_id, FieldId::min());
+        let end = DocumentFieldStoredKey::new(document_id, FieldId::max());
+        let iter = self.documents_fields.range(reader, &(start..=end))?;
+        Ok(DocumentFieldsIter { iter })
+    }
+}
+
+pub struct DocumentFieldsIter<'txn> {
+    iter: heed::RoRange<'txn, OwnedType<DocumentFieldStoredKey>, ByteSlice>,
+}
+
+impl<'txn> Iterator for DocumentFieldsIter<'txn> {
+    type Item = ZResult<(FieldId, &'txn [u8])>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match self.iter.next() {
+            Some(Ok((key, bytes))) => {
+                let field_id = FieldId(key.field_id.get());
+                Some(Ok((field_id, bytes)))
+            }
+            Some(Err(e)) => Some(Err(e)),
+            None => None,
+        }
+    }
+}

meilisearch-core/src/store/documents_fields_counts.rs

@@ -0,0 +1,142 @@
+use super::DocumentFieldIndexedKey;
+use crate::database::MainT;
+use crate::DocumentId;
+use heed::types::OwnedType;
+use heed::Result as ZResult;
+use meilisearch_schema::IndexedPos;
+
+#[derive(Copy, Clone)]
+pub struct DocumentsFieldsCounts {
+    pub(crate) documents_fields_counts: heed::Database<OwnedType<DocumentFieldIndexedKey>, OwnedType<u16>>,
+}
+
+impl DocumentsFieldsCounts {
+    pub fn put_document_field_count(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        document_id: DocumentId,
+        attribute: IndexedPos,
+        value: u16,
+    ) -> ZResult<()> {
+        let key = DocumentFieldIndexedKey::new(document_id, attribute);
+        self.documents_fields_counts.put(writer, &key, &value)
+    }
+
+    pub fn del_all_document_fields_counts(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        document_id: DocumentId,
+    ) -> ZResult<usize> {
+        let start = DocumentFieldIndexedKey::new(document_id, IndexedPos::min());
+        let end = DocumentFieldIndexedKey::new(document_id, IndexedPos::max());
+        self.documents_fields_counts.delete_range(writer, &(start..=end))
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.documents_fields_counts.clear(writer)
+    }
+
+    pub fn document_field_count(
+        self,
+        reader: &heed::RoTxn<MainT>,
+        document_id: DocumentId,
+        attribute: IndexedPos,
+    ) -> ZResult<Option<u16>> {
+        let key = DocumentFieldIndexedKey::new(document_id, attribute);
+        match self.documents_fields_counts.get(reader, &key)? {
+            Some(count) => Ok(Some(count)),
+            None => Ok(None),
+        }
+    }
+
+    pub fn document_fields_counts<'txn>(
+        self,
+        reader: &'txn heed::RoTxn<MainT>,
+        document_id: DocumentId,
+    ) -> ZResult<DocumentFieldsCountsIter<'txn>> {
+        let start = DocumentFieldIndexedKey::new(document_id, IndexedPos::min());
+        let end = DocumentFieldIndexedKey::new(document_id, IndexedPos::max());
+        let iter = self.documents_fields_counts.range(reader, &(start..=end))?;
+        Ok(DocumentFieldsCountsIter { iter })
+    }
+
+    pub fn documents_ids<'txn>(self, reader: &'txn heed::RoTxn<MainT>) -> ZResult<DocumentsIdsIter<'txn>> {
+        let iter = self.documents_fields_counts.iter(reader)?;
+        Ok(DocumentsIdsIter {
+            last_seen_id: None,
+            iter,
+        })
+    }
+
+    pub fn all_documents_fields_counts<'txn>(
+        self,
+        reader: &'txn heed::RoTxn<MainT>,
+    ) -> ZResult<AllDocumentsFieldsCountsIter<'txn>> {
+        let iter = self.documents_fields_counts.iter(reader)?;
+        Ok(AllDocumentsFieldsCountsIter { iter })
+    }
+}
+
+pub struct DocumentFieldsCountsIter<'txn> {
+    iter: heed::RoRange<'txn, OwnedType<DocumentFieldIndexedKey>, OwnedType<u16>>,
+}
+
+impl Iterator for DocumentFieldsCountsIter<'_> {
+    type Item = ZResult<(IndexedPos, u16)>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match self.iter.next() {
+            Some(Ok((key, count))) => {
+                let indexed_pos = IndexedPos(key.indexed_pos.get());
+                Some(Ok((indexed_pos, count)))
+            }
+            Some(Err(e)) => Some(Err(e)),
+            None => None,
+        }
+    }
+}
+
+pub struct DocumentsIdsIter<'txn> {
+    last_seen_id: Option<DocumentId>,
+    iter: heed::RoIter<'txn, OwnedType<DocumentFieldIndexedKey>, OwnedType<u16>>,
+}
+
+impl Iterator for DocumentsIdsIter<'_> {
+    type Item = ZResult<DocumentId>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        for result in &mut self.iter {
+            match result {
+                Ok((key, _)) => {
+                    let document_id = DocumentId(key.docid.get());
+                    if Some(document_id) != self.last_seen_id {
+                        self.last_seen_id = Some(document_id);
+                        return Some(Ok(document_id));
+                    }
+                }
+                Err(e) => return Some(Err(e)),
+            }
+        }
+        None
+    }
+}
+
+pub struct AllDocumentsFieldsCountsIter<'txn> {
+    iter: heed::RoIter<'txn, OwnedType<DocumentFieldIndexedKey>, OwnedType<u16>>,
+}
+
+impl Iterator for AllDocumentsFieldsCountsIter<'_> {
+    type Item = ZResult<(DocumentId, IndexedPos, u16)>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match self.iter.next() {
+            Some(Ok((key, count))) => {
+                let docid = DocumentId(key.docid.get());
+                let indexed_pos = IndexedPos(key.indexed_pos.get());
+                Some(Ok((docid, indexed_pos, count)))
+            }
+            Some(Err(e)) => Some(Err(e)),
+            None => None,
+        }
+    }
+}

meilisearch-core/src/store/main.rs

@@ -0,0 +1,226 @@
+use std::sync::Arc;
+use std::collections::HashMap;
+
+use chrono::{DateTime, Utc};
+use heed::types::{ByteSlice, OwnedType, SerdeBincode, Str};
+use heed::Result as ZResult;
+use meilisearch_schema::Schema;
+
+use crate::database::MainT;
+use crate::RankedMap;
+use crate::settings::RankingRule;
+
+const CREATED_AT_KEY: &str = "created-at";
+const RANKING_RULES_KEY: &str = "ranking-rules";
+const DISTINCT_ATTRIBUTE_KEY: &str = "distinct-attribute";
+const STOP_WORDS_KEY: &str = "stop-words";
+const SYNONYMS_KEY: &str = "synonyms";
+const CUSTOMS_KEY: &str = "customs";
+const FIELDS_FREQUENCY_KEY: &str = "fields-frequency";
+const NAME_KEY: &str = "name";
+const NUMBER_OF_DOCUMENTS_KEY: &str = "number-of-documents";
+const RANKED_MAP_KEY: &str = "ranked-map";
+const SCHEMA_KEY: &str = "schema";
+const UPDATED_AT_KEY: &str = "updated-at";
+const WORDS_KEY: &str = "words";
+
+pub type FreqsMap = HashMap<String, usize>;
+type SerdeFreqsMap = SerdeBincode<FreqsMap>;
+type SerdeDatetime = SerdeBincode<DateTime<Utc>>;
+
+#[derive(Copy, Clone)]
+pub struct Main {
+    pub(crate) main: heed::PolyDatabase,
+}
+
+impl Main {
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.main.clear(writer)
+    }
+
+    pub fn put_name(self, writer: &mut heed::RwTxn<MainT>, name: &str) -> ZResult<()> {
+        self.main.put::<_, Str, Str>(writer, NAME_KEY, name)
+    }
+
+    pub fn name(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<String>> {
+        Ok(self
+            .main
+            .get::<_, Str, Str>(reader, NAME_KEY)?
+            .map(|name| name.to_owned()))
+    }
+
+    pub fn put_created_at(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.main
+            .put::<_, Str, SerdeDatetime>(writer, CREATED_AT_KEY, &Utc::now())
+    }
+
+    pub fn created_at(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<DateTime<Utc>>> {
+        self.main.get::<_, Str, SerdeDatetime>(reader, CREATED_AT_KEY)
+    }
+
+    pub fn put_updated_at(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.main
+            .put::<_, Str, SerdeDatetime>(writer, UPDATED_AT_KEY, &Utc::now())
+    }
+
+    pub fn updated_at(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<DateTime<Utc>>> {
+        self.main.get::<_, Str, SerdeDatetime>(reader, UPDATED_AT_KEY)
+    }
+
+    pub fn put_words_fst(self, writer: &mut heed::RwTxn<MainT>, fst: &fst::Set) -> ZResult<()> {
+        let bytes = fst.as_fst().as_bytes();
+        self.main.put::<_, Str, ByteSlice>(writer, WORDS_KEY, bytes)
+    }
+
+    pub unsafe fn static_words_fst(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<fst::Set>> {
+        match self.main.get::<_, Str, ByteSlice>(reader, WORDS_KEY)? {
+            Some(bytes) => {
+                let bytes: &'static [u8] = std::mem::transmute(bytes);
+                let set = fst::Set::from_static_slice(bytes).unwrap();
+                Ok(Some(set))
+            }
+            None => Ok(None),
+        }
+    }
+
+    pub fn words_fst(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<fst::Set>> {
+        match self.main.get::<_, Str, ByteSlice>(reader, WORDS_KEY)? {
+            Some(bytes) => {
+                let len = bytes.len();
+                let bytes = Arc::new(bytes.to_owned());
+                let fst = fst::raw::Fst::from_shared_bytes(bytes, 0, len).unwrap();
+                Ok(Some(fst::Set::from(fst)))
+            }
+            None => Ok(None),
+        }
+    }
+
+    pub fn put_schema(self, writer: &mut heed::RwTxn<MainT>, schema: &Schema) -> ZResult<()> {
+        self.main.put::<_, Str, SerdeBincode<Schema>>(writer, SCHEMA_KEY, schema)
+    }
+
+    pub fn schema(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<Schema>> {
+        self.main.get::<_, Str, SerdeBincode<Schema>>(reader, SCHEMA_KEY)
+    }
+
+    pub fn delete_schema(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<bool> {
+        self.main.delete::<_, Str>(writer, SCHEMA_KEY)
+    }
+
+    pub fn put_ranked_map(self, writer: &mut heed::RwTxn<MainT>, ranked_map: &RankedMap) -> ZResult<()> {
+        self.main.put::<_, Str, SerdeBincode<RankedMap>>(writer, RANKED_MAP_KEY, &ranked_map)
+    }
+
+    pub fn ranked_map(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<RankedMap>> {
+        self.main.get::<_, Str, SerdeBincode<RankedMap>>(reader, RANKED_MAP_KEY)
+    }
+
+    pub fn put_synonyms_fst(self, writer: &mut heed::RwTxn<MainT>, fst: &fst::Set) -> ZResult<()> {
+        let bytes = fst.as_fst().as_bytes();
+        self.main.put::<_, Str, ByteSlice>(writer, SYNONYMS_KEY, bytes)
+    }
+
+    pub fn synonyms_fst(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<fst::Set>> {
+        match self.main.get::<_, Str, ByteSlice>(reader, SYNONYMS_KEY)? {
+            Some(bytes) => {
+                let len = bytes.len();
+                let bytes = Arc::new(bytes.to_owned());
+                let fst = fst::raw::Fst::from_shared_bytes(bytes, 0, len).unwrap();
+                Ok(Some(fst::Set::from(fst)))
+            }
+            None => Ok(None),
+        }
+    }
+
+    pub fn put_stop_words_fst(self, writer: &mut heed::RwTxn<MainT>, fst: &fst::Set) -> ZResult<()> {
+        let bytes = fst.as_fst().as_bytes();
+        self.main.put::<_, Str, ByteSlice>(writer, STOP_WORDS_KEY, bytes)
+    }
+
+    pub fn stop_words_fst(self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<fst::Set>> {
+        match self.main.get::<_, Str, ByteSlice>(reader, STOP_WORDS_KEY)? {
+            Some(bytes) => {
+                let len = bytes.len();
+                let bytes = Arc::new(bytes.to_owned());
+                let fst = fst::raw::Fst::from_shared_bytes(bytes, 0, len).unwrap();
+                Ok(Some(fst::Set::from(fst)))
+            }
+            None => Ok(None),
+        }
+    }
+
+    pub fn put_number_of_documents<F>(self, writer: &mut heed::RwTxn<MainT>, f: F) -> ZResult<u64>
+    where
+        F: Fn(u64) -> u64,
+    {
+        let new = self.number_of_documents(&*writer).map(f)?;
+        self.main
+            .put::<_, Str, OwnedType<u64>>(writer, NUMBER_OF_DOCUMENTS_KEY, &new)?;
+        Ok(new)
+    }
+
+    pub fn number_of_documents(self, reader: &heed::RoTxn<MainT>) -> ZResult<u64> {
+        match self
+            .main
+            .get::<_, Str, OwnedType<u64>>(reader, NUMBER_OF_DOCUMENTS_KEY)?
+        {
+            Some(value) => Ok(value),
+            None => Ok(0),
+        }
+    }
+
+    pub fn put_fields_frequency(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        fields_frequency: &FreqsMap,
+    ) -> ZResult<()> {
+        self.main
+            .put::<_, Str, SerdeFreqsMap>(writer, FIELDS_FREQUENCY_KEY, fields_frequency)
+    }
+
+    pub fn fields_frequency(&self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<FreqsMap>> {
+        match self
+            .main
+            .get::<_, Str, SerdeFreqsMap>(reader, FIELDS_FREQUENCY_KEY)?
+        {
+            Some(freqs) => Ok(Some(freqs)),
+            None => Ok(None),
+        }
+    }
+
+    pub fn ranking_rules(&self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<Vec<RankingRule>>> {
+        self.main.get::<_, Str, SerdeBincode<Vec<RankingRule>>>(reader, RANKING_RULES_KEY)
+    }
+
+    pub fn put_ranking_rules(self, writer: &mut heed::RwTxn<MainT>, value: &[RankingRule]) -> ZResult<()> {
+        self.main.put::<_, Str, SerdeBincode<Vec<RankingRule>>>(writer, RANKING_RULES_KEY, &value.to_vec())
+    }
+
+    pub fn delete_ranking_rules(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<bool> {
+        self.main.delete::<_, Str>(writer, RANKING_RULES_KEY)
+    }
+
+    pub fn distinct_attribute(&self, reader: &heed::RoTxn<MainT>) -> ZResult<Option<String>> {
+        if let Some(value) = self.main.get::<_, Str, Str>(reader, DISTINCT_ATTRIBUTE_KEY)? {
+            return Ok(Some(value.to_owned()))
+        }
+        return Ok(None)
+    }
+
+    pub fn put_distinct_attribute(self, writer: &mut heed::RwTxn<MainT>, value: &str) -> ZResult<()> {
+        self.main.put::<_, Str, Str>(writer, DISTINCT_ATTRIBUTE_KEY, value)
+    }
+
+    pub fn delete_distinct_attribute(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<bool> {
+        self.main.delete::<_, Str>(writer, DISTINCT_ATTRIBUTE_KEY)
+    }
+
+    pub fn put_customs(self, writer: &mut heed::RwTxn<MainT>, customs: &[u8]) -> ZResult<()> {
+        self.main
+            .put::<_, Str, ByteSlice>(writer, CUSTOMS_KEY, customs)
+    }
+
+    pub fn customs<'txn>(self, reader: &'txn heed::RoTxn<MainT>) -> ZResult<Option<&'txn [u8]>> {
+        self.main.get::<_, Str, ByteSlice>(reader, CUSTOMS_KEY)
+    }
+}

meilisearch-core/src/store/mod.rs

@@ -0,0 +1,518 @@
+mod docs_words;
+mod prefix_documents_cache;
+mod prefix_postings_lists_cache;
+mod documents_fields;
+mod documents_fields_counts;
+mod main;
+mod postings_lists;
+mod synonyms;
+mod updates;
+mod updates_results;
+
+pub use self::docs_words::DocsWords;
+pub use self::prefix_documents_cache::PrefixDocumentsCache;
+pub use self::prefix_postings_lists_cache::PrefixPostingsListsCache;
+pub use self::documents_fields::{DocumentFieldsIter, DocumentsFields};
+pub use self::documents_fields_counts::{
+    DocumentFieldsCountsIter, DocumentsFieldsCounts, DocumentsIdsIter,
+};
+pub use self::main::Main;
+pub use self::postings_lists::PostingsLists;
+pub use self::synonyms::Synonyms;
+pub use self::updates::Updates;
+pub use self::updates_results::UpdatesResults;
+
+use std::borrow::Cow;
+use std::collections::HashSet;
+use std::convert::TryInto;
+use std::{mem, ptr};
+
+use heed::Result as ZResult;
+use heed::{BytesEncode, BytesDecode};
+use meilisearch_schema::{IndexedPos, FieldId};
+use sdset::{Set, SetBuf};
+use serde::de::{self, Deserialize};
+use zerocopy::{AsBytes, FromBytes};
+
+use crate::criterion::Criteria;
+use crate::database::{MainT, UpdateT};
+use crate::database::{UpdateEvent, UpdateEventsEmitter};
+use crate::serde::Deserializer;
+use crate::settings::SettingsUpdate;
+use crate::{query_builder::QueryBuilder, update, DocIndex, DocumentId, Error, MResult};
+
+type BEU64 = zerocopy::U64<byteorder::BigEndian>;
+type BEU16 = zerocopy::U16<byteorder::BigEndian>;
+
+#[derive(Debug, Copy, Clone, AsBytes, FromBytes)]
+#[repr(C)]
+pub struct DocumentFieldIndexedKey {
+    docid: BEU64,
+    indexed_pos: BEU16,
+}
+
+impl DocumentFieldIndexedKey {
+    fn new(docid: DocumentId, indexed_pos: IndexedPos) -> DocumentFieldIndexedKey {
+        DocumentFieldIndexedKey {
+            docid: BEU64::new(docid.0),
+            indexed_pos: BEU16::new(indexed_pos.0),
+        }
+    }
+}
+
+#[derive(Debug, Copy, Clone, AsBytes, FromBytes)]
+#[repr(C)]
+pub struct DocumentFieldStoredKey {
+    docid: BEU64,
+    field_id: BEU16,
+}
+
+impl DocumentFieldStoredKey {
+    fn new(docid: DocumentId, field_id: FieldId) -> DocumentFieldStoredKey {
+        DocumentFieldStoredKey {
+            docid: BEU64::new(docid.0),
+            field_id: BEU16::new(field_id.0),
+        }
+    }
+}
+
+#[derive(Default, Debug)]
+pub struct Postings<'a> {
+    pub docids: Cow<'a, Set<DocumentId>>,
+    pub matches: Cow<'a, Set<DocIndex>>,
+}
+
+pub struct PostingsCodec;
+
+impl<'a> BytesEncode<'a> for PostingsCodec {
+    type EItem = Postings<'a>;
+
+    fn bytes_encode(item: &'a Self::EItem) -> Option<Cow<'a, [u8]>> {
+        let u64_size = mem::size_of::<u64>();
+        let docids_size = item.docids.len() * mem::size_of::<DocumentId>();
+        let matches_size = item.matches.len() * mem::size_of::<DocIndex>();
+
+        let mut buffer = Vec::with_capacity(u64_size + docids_size + matches_size);
+
+        let docids_len = item.docids.len();
+        buffer.extend_from_slice(&docids_len.to_be_bytes());
+        buffer.extend_from_slice(item.docids.as_bytes());
+        buffer.extend_from_slice(item.matches.as_bytes());
+
+        Some(Cow::Owned(buffer))
+    }
+}
+
+fn aligned_to(bytes: &[u8], align: usize) -> bool {
+    (bytes as *const _ as *const () as usize) % align == 0
+}
+
+fn from_bytes_to_set<'a, T: 'a>(bytes: &'a [u8]) -> Option<Cow<'a, Set<T>>>
+where T: Clone + FromBytes
+{
+    match zerocopy::LayoutVerified::<_, [T]>::new_slice(bytes) {
+        Some(layout) => Some(Cow::Borrowed(Set::new_unchecked(layout.into_slice()))),
+        None => {
+            let len = bytes.len();
+            let elem_size = mem::size_of::<T>();
+
+            // ensure that it is the alignment that is wrong
+            // and the length is valid
+            if len % elem_size == 0 && !aligned_to(bytes, mem::align_of::<T>()) {
+                let elems = len / elem_size;
+                let mut vec = Vec::<T>::with_capacity(elems);
+
+                unsafe {
+                    let dst = vec.as_mut_ptr() as *mut u8;
+                    ptr::copy_nonoverlapping(bytes.as_ptr(), dst, len);
+                    vec.set_len(elems);
+                }
+
+                return Some(Cow::Owned(SetBuf::new_unchecked(vec)));
+            }
+
+            None
+        }
+    }
+}
+
+impl<'a> BytesDecode<'a> for PostingsCodec {
+    type DItem = Postings<'a>;
+
+    fn bytes_decode(bytes: &'a [u8]) -> Option<Self::DItem> {
+        let u64_size = mem::size_of::<u64>();
+        let docid_size = mem::size_of::<DocumentId>();
+
+        let (len_bytes, bytes) = bytes.split_at(u64_size);
+        let docids_len = len_bytes.try_into().ok().map(u64::from_be_bytes)? as usize;
+        let docids_size = docids_len * docid_size;
+
+        let docids_bytes = &bytes[..docids_size];
+        let matches_bytes = &bytes[docids_size..];
+
+        let docids = from_bytes_to_set(docids_bytes)?;
+        let matches = from_bytes_to_set(matches_bytes)?;
+
+        Some(Postings { docids, matches })
+    }
+}
+
+fn main_name(name: &str) -> String {
+    format!("store-{}", name)
+}
+
+fn postings_lists_name(name: &str) -> String {
+    format!("store-{}-postings-lists", name)
+}
+
+fn documents_fields_name(name: &str) -> String {
+    format!("store-{}-documents-fields", name)
+}
+
+fn documents_fields_counts_name(name: &str) -> String {
+    format!("store-{}-documents-fields-counts", name)
+}
+
+fn synonyms_name(name: &str) -> String {
+    format!("store-{}-synonyms", name)
+}
+
+fn docs_words_name(name: &str) -> String {
+    format!("store-{}-docs-words", name)
+}
+
+fn prefix_documents_cache_name(name: &str) -> String {
+    format!("store-{}-prefix-documents-cache", name)
+}
+
+fn prefix_postings_lists_cache_name(name: &str) -> String {
+    format!("store-{}-prefix-postings-lists-cache", name)
+}
+
+fn updates_name(name: &str) -> String {
+    format!("store-{}-updates", name)
+}
+
+fn updates_results_name(name: &str) -> String {
+    format!("store-{}-updates-results", name)
+}
+
+#[derive(Clone)]
+pub struct Index {
+    pub main: Main,
+    pub postings_lists: PostingsLists,
+    pub documents_fields: DocumentsFields,
+    pub documents_fields_counts: DocumentsFieldsCounts,
+    pub synonyms: Synonyms,
+    pub docs_words: DocsWords,
+    pub prefix_documents_cache: PrefixDocumentsCache,
+    pub prefix_postings_lists_cache: PrefixPostingsListsCache,
+
+    pub updates: Updates,
+    pub updates_results: UpdatesResults,
+    pub(crate) updates_notifier: UpdateEventsEmitter,
+}
+
+impl Index {
+    pub fn document<T: de::DeserializeOwned>(
+        &self,
+        reader: &heed::RoTxn<MainT>,
+        attributes: Option<&HashSet<&str>>,
+        document_id: DocumentId,
+    ) -> MResult<Option<T>> {
+        let schema = self.main.schema(reader)?;
+        let schema = schema.ok_or(Error::SchemaMissing)?;
+
+        let attributes = match attributes {
+            Some(attributes) => Some(attributes.iter().filter_map(|name| schema.id(*name)).collect()),
+            None => None,
+        };
+
+        let mut deserializer = Deserializer {
+            document_id,
+            reader,
+            documents_fields: self.documents_fields,
+            schema: &schema,
+            fields: attributes.as_ref(),
+        };
+
+        Ok(Option::<T>::deserialize(&mut deserializer)?)
+    }
+
+    pub fn document_attribute<T: de::DeserializeOwned>(
+        &self,
+        reader: &heed::RoTxn<MainT>,
+        document_id: DocumentId,
+        attribute: FieldId,
+    ) -> MResult<Option<T>> {
+        let bytes = self
+            .documents_fields
+            .document_attribute(reader, document_id, attribute)?;
+        match bytes {
+            Some(bytes) => Ok(Some(serde_json::from_slice(bytes)?)),
+            None => Ok(None),
+        }
+    }
+
+    pub fn document_attribute_bytes<'txn>(
+        &self,
+        reader: &'txn heed::RoTxn<MainT>,
+        document_id: DocumentId,
+        attribute: FieldId,
+    ) -> MResult<Option<&'txn [u8]>> {
+        let bytes = self
+            .documents_fields
+            .document_attribute(reader, document_id, attribute)?;
+        match bytes {
+            Some(bytes) => Ok(Some(bytes)),
+            None => Ok(None),
+        }
+    }
+
+    pub fn customs_update(&self, writer: &mut heed::RwTxn<UpdateT>, customs: Vec<u8>) -> ZResult<u64> {
+        let _ = self.updates_notifier.send(UpdateEvent::NewUpdate);
+        update::push_customs_update(writer, self.updates, self.updates_results, customs)
+    }
+
+    pub fn settings_update(&self, writer: &mut heed::RwTxn<UpdateT>, update: SettingsUpdate) -> ZResult<u64> {
+        let _ = self.updates_notifier.send(UpdateEvent::NewUpdate);
+        update::push_settings_update(writer, self.updates, self.updates_results, update)
+    }
+
+    pub fn documents_addition<D>(&self) -> update::DocumentsAddition<D> {
+        update::DocumentsAddition::new(
+            self.updates,
+            self.updates_results,
+            self.updates_notifier.clone(),
+        )
+    }
+
+    pub fn documents_partial_addition<D>(&self) -> update::DocumentsAddition<D> {
+        update::DocumentsAddition::new_partial(
+            self.updates,
+            self.updates_results,
+            self.updates_notifier.clone(),
+        )
+    }
+
+    pub fn documents_deletion(&self) -> update::DocumentsDeletion {
+        update::DocumentsDeletion::new(
+            self.updates,
+            self.updates_results,
+            self.updates_notifier.clone(),
+        )
+    }
+
+    pub fn clear_all(&self, writer: &mut heed::RwTxn<UpdateT>) -> MResult<u64> {
+        let _ = self.updates_notifier.send(UpdateEvent::NewUpdate);
+        update::push_clear_all(writer, self.updates, self.updates_results)
+    }
+
+    pub fn current_update_id(&self, reader: &heed::RoTxn<UpdateT>) -> MResult<Option<u64>> {
+        match self.updates.last_update(reader)? {
+            Some((id, _)) => Ok(Some(id)),
+            None => Ok(None),
+        }
+    }
+
+    pub fn update_status(
+        &self,
+        reader: &heed::RoTxn<UpdateT>,
+        update_id: u64,
+    ) -> MResult<Option<update::UpdateStatus>> {
+        update::update_status(reader, self.updates, self.updates_results, update_id)
+    }
+
+    pub fn all_updates_status(&self, reader: &heed::RoTxn<UpdateT>) -> MResult<Vec<update::UpdateStatus>> {
+        let mut updates = Vec::new();
+        let mut last_update_result_id = 0;
+
+        // retrieve all updates results
+        if let Some((last_id, _)) = self.updates_results.last_update(reader)? {
+            updates.reserve(last_id as usize);
+
+            for id in 0..=last_id {
+                if let Some(update) = self.update_status(reader, id)? {
+                    updates.push(update);
+                    last_update_result_id = id + 1;
+                }
+            }
+        }
+
+        // retrieve all enqueued updates
+        if let Some((last_id, _)) = self.updates.last_update(reader)? {
+            for id in last_update_result_id..=last_id {
+                if let Some(update) = self.update_status(reader, id)? {
+                    updates.push(update);
+                }
+            }
+        }
+
+        Ok(updates)
+    }
+
+    pub fn query_builder(&self) -> QueryBuilder {
+        QueryBuilder::new(
+            self.main,
+            self.postings_lists,
+            self.documents_fields_counts,
+            self.synonyms,
+            self.prefix_documents_cache,
+            self.prefix_postings_lists_cache,
+        )
+    }
+
+    pub fn query_builder_with_criteria<'c, 'f, 'd>(
+        &self,
+        criteria: Criteria<'c>,
+    ) -> QueryBuilder<'c, 'f, 'd> {
+        QueryBuilder::with_criteria(
+            self.main,
+            self.postings_lists,
+            self.documents_fields_counts,
+            self.synonyms,
+            self.prefix_documents_cache,
+            self.prefix_postings_lists_cache,
+            criteria,
+        )
+    }
+}
+
+pub fn create(
+    env: &heed::Env,
+    update_env: &heed::Env,
+    name: &str,
+    updates_notifier: UpdateEventsEmitter,
+) -> MResult<Index> {
+    // create all the store names
+    let main_name = main_name(name);
+    let postings_lists_name = postings_lists_name(name);
+    let documents_fields_name = documents_fields_name(name);
+    let documents_fields_counts_name = documents_fields_counts_name(name);
+    let synonyms_name = synonyms_name(name);
+    let docs_words_name = docs_words_name(name);
+    let prefix_documents_cache_name = prefix_documents_cache_name(name);
+    let prefix_postings_lists_cache_name = prefix_postings_lists_cache_name(name);
+    let updates_name = updates_name(name);
+    let updates_results_name = updates_results_name(name);
+
+    // open all the stores
+    let main = env.create_poly_database(Some(&main_name))?;
+    let postings_lists = env.create_database(Some(&postings_lists_name))?;
+    let documents_fields = env.create_database(Some(&documents_fields_name))?;
+    let documents_fields_counts = env.create_database(Some(&documents_fields_counts_name))?;
+    let synonyms = env.create_database(Some(&synonyms_name))?;
+    let docs_words = env.create_database(Some(&docs_words_name))?;
+    let prefix_documents_cache = env.create_database(Some(&prefix_documents_cache_name))?;
+    let prefix_postings_lists_cache = env.create_database(Some(&prefix_postings_lists_cache_name))?;
+    let updates = update_env.create_database(Some(&updates_name))?;
+    let updates_results = update_env.create_database(Some(&updates_results_name))?;
+
+    Ok(Index {
+        main: Main { main },
+        postings_lists: PostingsLists { postings_lists },
+        documents_fields: DocumentsFields { documents_fields },
+        documents_fields_counts: DocumentsFieldsCounts { documents_fields_counts },
+        synonyms: Synonyms { synonyms },
+        docs_words: DocsWords { docs_words },
+        prefix_postings_lists_cache: PrefixPostingsListsCache { prefix_postings_lists_cache },
+        prefix_documents_cache: PrefixDocumentsCache { prefix_documents_cache },
+        updates: Updates { updates },
+        updates_results: UpdatesResults { updates_results },
+        updates_notifier,
+    })
+}
+
+pub fn open(
+    env: &heed::Env,
+    update_env: &heed::Env,
+    name: &str,
+    updates_notifier: UpdateEventsEmitter,
+) -> MResult<Option<Index>> {
+    // create all the store names
+    let main_name = main_name(name);
+    let postings_lists_name = postings_lists_name(name);
+    let documents_fields_name = documents_fields_name(name);
+    let documents_fields_counts_name = documents_fields_counts_name(name);
+    let synonyms_name = synonyms_name(name);
+    let docs_words_name = docs_words_name(name);
+    let prefix_documents_cache_name = prefix_documents_cache_name(name);
+    let prefix_postings_lists_cache_name = prefix_postings_lists_cache_name(name);
+    let updates_name = updates_name(name);
+    let updates_results_name = updates_results_name(name);
+
+    // open all the stores
+    let main = match env.open_poly_database(Some(&main_name))? {
+        Some(main) => main,
+        None => return Ok(None),
+    };
+    let postings_lists = match env.open_database(Some(&postings_lists_name))? {
+        Some(postings_lists) => postings_lists,
+        None => return Ok(None),
+    };
+    let documents_fields = match env.open_database(Some(&documents_fields_name))? {
+        Some(documents_fields) => documents_fields,
+        None => return Ok(None),
+    };
+    let documents_fields_counts = match env.open_database(Some(&documents_fields_counts_name))? {
+        Some(documents_fields_counts) => documents_fields_counts,
+        None => return Ok(None),
+    };
+    let synonyms = match env.open_database(Some(&synonyms_name))? {
+        Some(synonyms) => synonyms,
+        None => return Ok(None),
+    };
+    let docs_words = match env.open_database(Some(&docs_words_name))? {
+        Some(docs_words) => docs_words,
+        None => return Ok(None),
+    };
+    let prefix_documents_cache = match env.open_database(Some(&prefix_documents_cache_name))? {
+        Some(prefix_documents_cache) => prefix_documents_cache,
+        None => return Ok(None),
+    };
+    let prefix_postings_lists_cache = match env.open_database(Some(&prefix_postings_lists_cache_name))? {
+        Some(prefix_postings_lists_cache) => prefix_postings_lists_cache,
+        None => return Ok(None),
+    };
+    let updates = match update_env.open_database(Some(&updates_name))? {
+        Some(updates) => updates,
+        None => return Ok(None),
+    };
+    let updates_results = match update_env.open_database(Some(&updates_results_name))? {
+        Some(updates_results) => updates_results,
+        None => return Ok(None),
+    };
+
+    Ok(Some(Index {
+        main: Main { main },
+        postings_lists: PostingsLists { postings_lists },
+        documents_fields: DocumentsFields { documents_fields },
+        documents_fields_counts: DocumentsFieldsCounts { documents_fields_counts },
+        synonyms: Synonyms { synonyms },
+        docs_words: DocsWords { docs_words },
+        prefix_documents_cache: PrefixDocumentsCache { prefix_documents_cache },
+        prefix_postings_lists_cache: PrefixPostingsListsCache { prefix_postings_lists_cache },
+        updates: Updates { updates },
+        updates_results: UpdatesResults { updates_results },
+        updates_notifier,
+    }))
+}
+
+pub fn clear(
+    writer: &mut heed::RwTxn<MainT>,
+    update_writer: &mut heed::RwTxn<UpdateT>,
+    index: &Index,
+) -> MResult<()> {
+    // clear all the stores
+    index.main.clear(writer)?;
+    index.postings_lists.clear(writer)?;
+    index.documents_fields.clear(writer)?;
+    index.documents_fields_counts.clear(writer)?;
+    index.synonyms.clear(writer)?;
+    index.docs_words.clear(writer)?;
+    index.prefix_documents_cache.clear(writer)?;
+    index.prefix_postings_lists_cache.clear(writer)?;
+    index.updates.clear(update_writer)?;
+    index.updates_results.clear(update_writer)?;
+    Ok(())
+}

meilisearch-core/src/store/postings_lists.rs

@@ -0,0 +1,47 @@
+use std::borrow::Cow;
+
+use heed::Result as ZResult;
+use heed::types::ByteSlice;
+use sdset::{Set, SetBuf};
+use slice_group_by::GroupBy;
+
+use crate::database::MainT;
+use crate::DocIndex;
+use crate::store::{Postings, PostingsCodec};
+
+#[derive(Copy, Clone)]
+pub struct PostingsLists {
+    pub(crate) postings_lists: heed::Database<ByteSlice, PostingsCodec>,
+}
+
+impl PostingsLists {
+    pub fn put_postings_list(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        word: &[u8],
+        matches: &Set<DocIndex>,
+    ) -> ZResult<()> {
+        let docids = matches.linear_group_by_key(|m| m.document_id).map(|g| g[0].document_id).collect();
+        let docids = Cow::Owned(SetBuf::new_unchecked(docids));
+        let matches = Cow::Borrowed(matches);
+        let postings = Postings { docids, matches };
+
+        self.postings_lists.put(writer, word, &postings)
+    }
+
+    pub fn del_postings_list(self, writer: &mut heed::RwTxn<MainT>, word: &[u8]) -> ZResult<bool> {
+        self.postings_lists.delete(writer, word)
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.postings_lists.clear(writer)
+    }
+
+    pub fn postings_list<'txn>(
+        self,
+        reader: &'txn heed::RoTxn<MainT>,
+        word: &[u8],
+    ) -> ZResult<Option<Postings<'txn>>> {
+        self.postings_lists.get(reader, word)
+    }
+}

meilisearch-core/src/store/prefix_documents_cache.rs

@@ -0,0 +1,80 @@
+use std::borrow::Cow;
+
+use heed::types::{OwnedType, CowSlice};
+use heed::Result as ZResult;
+use zerocopy::{AsBytes, FromBytes};
+
+use super::BEU64;
+use crate::{DocumentId, Highlight};
+use crate::database::MainT;
+
+#[derive(Debug, Copy, Clone, AsBytes, FromBytes)]
+#[repr(C)]
+pub struct PrefixKey {
+    prefix: [u8; 4],
+    index: BEU64,
+    docid: BEU64,
+}
+
+impl PrefixKey {
+    pub fn new(prefix: [u8; 4], index: u64, docid: u64) -> PrefixKey {
+        PrefixKey {
+            prefix,
+            index: BEU64::new(index),
+            docid: BEU64::new(docid),
+        }
+    }
+}
+
+#[derive(Copy, Clone)]
+pub struct PrefixDocumentsCache {
+    pub(crate) prefix_documents_cache: heed::Database<OwnedType<PrefixKey>, CowSlice<Highlight>>,
+}
+
+impl PrefixDocumentsCache {
+    pub fn put_prefix_document(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        prefix: [u8; 4],
+        index: usize,
+        docid: DocumentId,
+        highlights: &[Highlight],
+    ) -> ZResult<()> {
+        let key = PrefixKey::new(prefix, index as u64, docid.0);
+        self.prefix_documents_cache.put(writer, &key, highlights)
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.prefix_documents_cache.clear(writer)
+    }
+
+    pub fn prefix_documents<'txn>(
+        self,
+        reader: &'txn heed::RoTxn<MainT>,
+        prefix: [u8; 4],
+    ) -> ZResult<PrefixDocumentsIter<'txn>> {
+        let start = PrefixKey::new(prefix, 0, 0);
+        let end = PrefixKey::new(prefix, u64::max_value(), u64::max_value());
+        let iter = self.prefix_documents_cache.range(reader, &(start..=end))?;
+        Ok(PrefixDocumentsIter { iter })
+    }
+}
+
+pub struct PrefixDocumentsIter<'txn> {
+    iter: heed::RoRange<'txn, OwnedType<PrefixKey>, CowSlice<Highlight>>,
+}
+
+impl<'txn> Iterator for PrefixDocumentsIter<'txn> {
+    type Item = ZResult<(DocumentId, Cow<'txn, [Highlight]>)>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match self.iter.next() {
+            Some(Ok((key, highlights))) => {
+                let docid = DocumentId(key.docid.get());
+                Some(Ok((docid, highlights)))
+            }
+            Some(Err(e)) => Some(Err(e)),
+            None => None,
+        }
+    }
+}

meilisearch-core/src/store/prefix_postings_lists_cache.rs

@@ -0,0 +1,45 @@
+use std::borrow::Cow;
+
+use heed::Result as ZResult;
+use heed::types::OwnedType;
+use sdset::{Set, SetBuf};
+use slice_group_by::GroupBy;
+
+use crate::database::MainT;
+use crate::DocIndex;
+use crate::store::{PostingsCodec, Postings};
+
+#[derive(Copy, Clone)]
+pub struct PrefixPostingsListsCache {
+    pub(crate) prefix_postings_lists_cache: heed::Database<OwnedType<[u8; 4]>, PostingsCodec>,
+}
+
+impl PrefixPostingsListsCache {
+    pub fn put_prefix_postings_list(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        prefix: [u8; 4],
+        matches: &Set<DocIndex>,
+    ) -> ZResult<()>
+    {
+        let docids = matches.linear_group_by_key(|m| m.document_id).map(|g| g[0].document_id).collect();
+        let docids = Cow::Owned(SetBuf::new_unchecked(docids));
+        let matches = Cow::Borrowed(matches);
+        let postings = Postings { docids, matches };
+
+        self.prefix_postings_lists_cache.put(writer, &prefix, &postings)
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.prefix_postings_lists_cache.clear(writer)
+    }
+
+    pub fn prefix_postings_list<'txn>(
+        self,
+        reader: &'txn heed::RoTxn<MainT>,
+        prefix: [u8; 4],
+    ) -> ZResult<Option<Postings<'txn>>>
+    {
+        self.prefix_postings_lists_cache.get(reader, &prefix)
+    }
+}

meilisearch-core/src/store/synonyms.rs

@@ -0,0 +1,41 @@
+use heed::types::ByteSlice;
+use crate::database::MainT;
+use heed::Result as ZResult;
+use std::sync::Arc;
+
+#[derive(Copy, Clone)]
+pub struct Synonyms {
+    pub(crate) synonyms: heed::Database<ByteSlice, ByteSlice>,
+}
+
+impl Synonyms {
+    pub fn put_synonyms(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        word: &[u8],
+        synonyms: &fst::Set,
+    ) -> ZResult<()> {
+        let bytes = synonyms.as_fst().as_bytes();
+        self.synonyms.put(writer, word, bytes)
+    }
+
+    pub fn del_synonyms(self, writer: &mut heed::RwTxn<MainT>, word: &[u8]) -> ZResult<bool> {
+        self.synonyms.delete(writer, word)
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> ZResult<()> {
+        self.synonyms.clear(writer)
+    }
+
+    pub fn synonyms(self, reader: &heed::RoTxn<MainT>, word: &[u8]) -> ZResult<Option<fst::Set>> {
+        match self.synonyms.get(reader, word)? {
+            Some(bytes) => {
+                let len = bytes.len();
+                let bytes = Arc::new(bytes.to_owned());
+                let fst = fst::raw::Fst::from_shared_bytes(bytes, 0, len).unwrap();
+                Ok(Some(fst::Set::from(fst)))
+            }
+            None => Ok(None),
+        }
+    }
+}

meilisearch-core/src/store/updates.rs

@@ -0,0 +1,65 @@
+use super::BEU64;
+use crate::database::UpdateT;
+use crate::update::Update;
+use heed::types::{OwnedType, SerdeJson};
+use heed::Result as ZResult;
+
+#[derive(Copy, Clone)]
+pub struct Updates {
+    pub(crate) updates: heed::Database<OwnedType<BEU64>, SerdeJson<Update>>,
+}
+
+impl Updates {
+    // TODO do not trigger deserialize if possible
+    pub fn last_update(self, reader: &heed::RoTxn<UpdateT>) -> ZResult<Option<(u64, Update)>> {
+        match self.updates.last(reader)? {
+            Some((key, data)) => Ok(Some((key.get(), data))),
+            None => Ok(None),
+        }
+    }
+
+    // TODO do not trigger deserialize if possible
+    pub fn first_update(self, reader: &heed::RoTxn<UpdateT>) -> ZResult<Option<(u64, Update)>> {
+        match self.updates.first(reader)? {
+            Some((key, data)) => Ok(Some((key.get(), data))),
+            None => Ok(None),
+        }
+    }
+
+    // TODO do not trigger deserialize if possible
+    pub fn get(self, reader: &heed::RoTxn<UpdateT>, update_id: u64) -> ZResult<Option<Update>> {
+        let update_id = BEU64::new(update_id);
+        self.updates.get(reader, &update_id)
+    }
+
+    pub fn put_update(
+        self,
+        writer: &mut heed::RwTxn<UpdateT>,
+        update_id: u64,
+        update: &Update,
+    ) -> ZResult<()> {
+        // TODO prefer using serde_json?
+        let update_id = BEU64::new(update_id);
+        self.updates.put(writer, &update_id, update)
+    }
+
+    pub fn del_update(self, writer: &mut heed::RwTxn<UpdateT>, update_id: u64) -> ZResult<bool> {
+        let update_id = BEU64::new(update_id);
+        self.updates.delete(writer, &update_id)
+    }
+
+    pub fn pop_front(self, writer: &mut heed::RwTxn<UpdateT>) -> ZResult<Option<(u64, Update)>> {
+        match self.first_update(writer)? {
+            Some((update_id, update)) => {
+                let key = BEU64::new(update_id);
+                self.updates.delete(writer, &key)?;
+                Ok(Some((update_id, update)))
+            }
+            None => Ok(None),
+        }
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<UpdateT>) -> ZResult<()> {
+        self.updates.clear(writer)
+    }
+}

meilisearch-core/src/store/updates_results.rs

@@ -0,0 +1,45 @@
+use super::BEU64;
+use crate::database::UpdateT;
+use crate::update::ProcessedUpdateResult;
+use heed::types::{OwnedType, SerdeJson};
+use heed::Result as ZResult;
+
+#[derive(Copy, Clone)]
+pub struct UpdatesResults {
+    pub(crate) updates_results: heed::Database<OwnedType<BEU64>, SerdeJson<ProcessedUpdateResult>>,
+}
+
+impl UpdatesResults {
+    pub fn last_update(
+        self,
+        reader: &heed::RoTxn<UpdateT>,
+    ) -> ZResult<Option<(u64, ProcessedUpdateResult)>> {
+        match self.updates_results.last(reader)? {
+            Some((key, data)) => Ok(Some((key.get(), data))),
+            None => Ok(None),
+        }
+    }
+
+    pub fn put_update_result(
+        self,
+        writer: &mut heed::RwTxn<UpdateT>,
+        update_id: u64,
+        update_result: &ProcessedUpdateResult,
+    ) -> ZResult<()> {
+        let update_id = BEU64::new(update_id);
+        self.updates_results.put(writer, &update_id, update_result)
+    }
+
+    pub fn update_result(
+        self,
+        reader: &heed::RoTxn<UpdateT>,
+        update_id: u64,
+    ) -> ZResult<Option<ProcessedUpdateResult>> {
+        let update_id = BEU64::new(update_id);
+        self.updates_results.get(reader, &update_id)
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<UpdateT>) -> ZResult<()> {
+        self.updates_results.clear(writer)
+    }
+}

meilisearch-core/src/update/clear_all.rs

@@ -0,0 +1,32 @@
+use crate::database::{MainT, UpdateT};
+use crate::update::{next_update_id, Update};
+use crate::{store, MResult, RankedMap};
+
+pub fn apply_clear_all(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+) -> MResult<()> {
+    index.main.put_words_fst(writer, &fst::Set::default())?;
+    index.main.put_ranked_map(writer, &RankedMap::default())?;
+    index.main.put_number_of_documents(writer, |_| 0)?;
+    index.documents_fields.clear(writer)?;
+    index.documents_fields_counts.clear(writer)?;
+    index.postings_lists.clear(writer)?;
+    index.docs_words.clear(writer)?;
+    index.prefix_documents_cache.clear(writer)?;
+    index.prefix_postings_lists_cache.clear(writer)?;
+
+    Ok(())
+}
+
+pub fn push_clear_all(
+    writer: &mut heed::RwTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+) -> MResult<u64> {
+    let last_update_id = next_update_id(writer, updates_store, updates_results_store)?;
+    let update = Update::clear_all();
+    updates_store.put_update(writer, last_update_id, &update)?;
+
+    Ok(last_update_id)
+}

meilisearch-core/src/update/customs_update.rs

@@ -0,0 +1,27 @@
+use heed::Result as ZResult;
+
+use crate::database::{MainT, UpdateT};
+use crate::store;
+use crate::update::{next_update_id, Update};
+
+pub fn apply_customs_update(
+    writer: &mut heed::RwTxn<MainT>,
+    main_store: store::Main,
+    customs: &[u8],
+) -> ZResult<()> {
+    main_store.put_customs(writer, customs)
+}
+
+pub fn push_customs_update(
+    writer: &mut heed::RwTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    customs: Vec<u8>,
+) -> ZResult<u64> {
+    let last_update_id = next_update_id(writer, updates_store, updates_results_store)?;
+
+    let update = Update::customs(customs);
+    updates_store.put_update(writer, last_update_id, &update)?;
+
+    Ok(last_update_id)
+}

meilisearch-core/src/update/documents_addition.rs

@@ -0,0 +1,382 @@
+use std::collections::HashMap;
+
+use fst::{set::OpBuilder, SetBuilder};
+use indexmap::IndexMap;
+use sdset::{duo::Union, SetOperation};
+use serde::{Deserialize, Serialize};
+
+use crate::database::{MainT, UpdateT};
+use crate::database::{UpdateEvent, UpdateEventsEmitter};
+use crate::raw_indexer::RawIndexer;
+use crate::serde::{extract_document_id, serialize_value_with_id, Deserializer, Serializer};
+use crate::store;
+use crate::update::{apply_documents_deletion, compute_short_prefixes, next_update_id, Update};
+use crate::{Error, MResult, RankedMap};
+
+pub struct DocumentsAddition<D> {
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    updates_notifier: UpdateEventsEmitter,
+    documents: Vec<D>,
+    is_partial: bool,
+}
+
+impl<D> DocumentsAddition<D> {
+    pub fn new(
+        updates_store: store::Updates,
+        updates_results_store: store::UpdatesResults,
+        updates_notifier: UpdateEventsEmitter,
+    ) -> DocumentsAddition<D> {
+        DocumentsAddition {
+            updates_store,
+            updates_results_store,
+            updates_notifier,
+            documents: Vec::new(),
+            is_partial: false,
+        }
+    }
+
+    pub fn new_partial(
+        updates_store: store::Updates,
+        updates_results_store: store::UpdatesResults,
+        updates_notifier: UpdateEventsEmitter,
+    ) -> DocumentsAddition<D> {
+        DocumentsAddition {
+            updates_store,
+            updates_results_store,
+            updates_notifier,
+            documents: Vec::new(),
+            is_partial: true,
+        }
+    }
+
+    pub fn update_document(&mut self, document: D) {
+        self.documents.push(document);
+    }
+
+    pub fn finalize(self, writer: &mut heed::RwTxn<UpdateT>) -> MResult<u64>
+    where
+        D: serde::Serialize,
+    {
+        let _ = self.updates_notifier.send(UpdateEvent::NewUpdate);
+        let update_id = push_documents_addition(
+            writer,
+            self.updates_store,
+            self.updates_results_store,
+            self.documents,
+            self.is_partial,
+        )?;
+        Ok(update_id)
+    }
+}
+
+impl<D> Extend<D> for DocumentsAddition<D> {
+    fn extend<T: IntoIterator<Item = D>>(&mut self, iter: T) {
+        self.documents.extend(iter)
+    }
+}
+
+pub fn push_documents_addition<D: serde::Serialize>(
+    writer: &mut heed::RwTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    addition: Vec<D>,
+    is_partial: bool,
+) -> MResult<u64> {
+    let mut values = Vec::with_capacity(addition.len());
+    for add in addition {
+        let vec = serde_json::to_vec(&add)?;
+        let add = serde_json::from_slice(&vec)?;
+        values.push(add);
+    }
+
+    let last_update_id = next_update_id(writer, updates_store, updates_results_store)?;
+
+    let update = if is_partial {
+        Update::documents_partial(values)
+    } else {
+        Update::documents_addition(values)
+    };
+
+    updates_store.put_update(writer, last_update_id, &update)?;
+
+    Ok(last_update_id)
+}
+
+pub fn apply_documents_addition<'a, 'b>(
+    writer: &'a mut heed::RwTxn<'b, MainT>,
+    index: &store::Index,
+    addition: Vec<IndexMap<String, serde_json::Value>>,
+) -> MResult<()> {
+    let mut documents_additions = HashMap::new();
+
+    let mut schema = match index.main.schema(writer)? {
+        Some(schema) => schema,
+        None => return Err(Error::SchemaMissing),
+    };
+
+    let primary_key = schema.primary_key().ok_or(Error::MissingPrimaryKey)?;
+
+    // 1. store documents ids for future deletion
+    for document in addition {
+        let document_id = match extract_document_id(&primary_key, &document)? {
+            Some(id) => id,
+            None => return Err(Error::MissingDocumentId),
+        };
+
+        documents_additions.insert(document_id, document);
+    }
+
+    // 2. remove the documents posting lists
+    let number_of_inserted_documents = documents_additions.len();
+    let documents_ids = documents_additions.iter().map(|(id, _)| *id).collect();
+    apply_documents_deletion(writer, index, documents_ids)?;
+
+    let mut ranked_map = match index.main.ranked_map(writer)? {
+        Some(ranked_map) => ranked_map,
+        None => RankedMap::default(),
+    };
+
+    let stop_words = match index.main.stop_words_fst(writer)? {
+        Some(stop_words) => stop_words,
+        None => fst::Set::default(),
+    };
+
+    // 3. index the documents fields in the stores
+    let mut indexer = RawIndexer::new(stop_words);
+
+    for (document_id, document) in documents_additions {
+        let serializer = Serializer {
+            txn: writer,
+            schema: &mut schema,
+            document_store: index.documents_fields,
+            document_fields_counts: index.documents_fields_counts,
+            indexer: &mut indexer,
+            ranked_map: &mut ranked_map,
+            document_id,
+        };
+
+        document.serialize(serializer)?;
+    }
+
+    write_documents_addition_index(
+        writer,
+        index,
+        &ranked_map,
+        number_of_inserted_documents,
+        indexer,
+    )?;
+
+    index.main.put_schema(writer, &schema)?;
+
+    Ok(())
+}
+
+pub fn apply_documents_partial_addition<'a, 'b>(
+    writer: &'a mut heed::RwTxn<'b, MainT>,
+    index: &store::Index,
+    addition: Vec<IndexMap<String, serde_json::Value>>,
+) -> MResult<()> {
+    let mut documents_additions = HashMap::new();
+
+    let mut schema = match index.main.schema(writer)? {
+        Some(schema) => schema,
+        None => return Err(Error::SchemaMissing),
+    };
+
+    let primary_key = schema.primary_key().ok_or(Error::MissingPrimaryKey)?;
+
+    // 1. store documents ids for future deletion
+    for mut document in addition {
+        let document_id = match extract_document_id(&primary_key, &document)? {
+            Some(id) => id,
+            None => return Err(Error::MissingDocumentId),
+        };
+
+        let mut deserializer = Deserializer {
+            document_id,
+            reader: writer,
+            documents_fields: index.documents_fields,
+            schema: &schema,
+            fields: None,
+        };
+
+        // retrieve the old document and
+        // update the new one with missing keys found in the old one
+        let result = Option::<HashMap<String, serde_json::Value>>::deserialize(&mut deserializer)?;
+        if let Some(old_document) = result {
+            for (key, value) in old_document {
+                document.entry(key).or_insert(value);
+            }
+        }
+
+        documents_additions.insert(document_id, document);
+    }
+
+    // 2. remove the documents posting lists
+    let number_of_inserted_documents = documents_additions.len();
+    let documents_ids = documents_additions.iter().map(|(id, _)| *id).collect();
+    apply_documents_deletion(writer, index, documents_ids)?;
+
+    let mut ranked_map = match index.main.ranked_map(writer)? {
+        Some(ranked_map) => ranked_map,
+        None => RankedMap::default(),
+    };
+
+    let stop_words = match index.main.stop_words_fst(writer)? {
+        Some(stop_words) => stop_words,
+        None => fst::Set::default(),
+    };
+
+    // 3. index the documents fields in the stores
+    let mut indexer = RawIndexer::new(stop_words);
+
+    for (document_id, document) in documents_additions {
+        let serializer = Serializer {
+            txn: writer,
+            schema: &mut schema,
+            document_store: index.documents_fields,
+            document_fields_counts: index.documents_fields_counts,
+            indexer: &mut indexer,
+            ranked_map: &mut ranked_map,
+            document_id,
+        };
+
+        document.serialize(serializer)?;
+    }
+
+    write_documents_addition_index(
+        writer,
+        index,
+        &ranked_map,
+        number_of_inserted_documents,
+        indexer,
+    )?;
+
+    index.main.put_schema(writer, &schema)?;
+
+    Ok(())
+}
+
+pub fn reindex_all_documents(writer: &mut heed::RwTxn<MainT>, index: &store::Index) -> MResult<()> {
+    let schema = match index.main.schema(writer)? {
+        Some(schema) => schema,
+        None => return Err(Error::SchemaMissing),
+    };
+
+    let mut ranked_map = RankedMap::default();
+
+    // 1. retrieve all documents ids
+    let mut documents_ids_to_reindex = Vec::new();
+    for result in index.documents_fields_counts.documents_ids(writer)? {
+        let document_id = result?;
+        documents_ids_to_reindex.push(document_id);
+    }
+
+    // 2. remove the documents posting lists
+    index.main.put_words_fst(writer, &fst::Set::default())?;
+    index.main.put_ranked_map(writer, &ranked_map)?;
+    index.main.put_number_of_documents(writer, |_| 0)?;
+    index.postings_lists.clear(writer)?;
+    index.docs_words.clear(writer)?;
+
+    let stop_words = match index.main.stop_words_fst(writer)? {
+        Some(stop_words) => stop_words,
+        None => fst::Set::default(),
+    };
+
+    let number_of_inserted_documents = documents_ids_to_reindex.len();
+    let mut indexer = RawIndexer::new(stop_words);
+    let mut ram_store = HashMap::new();
+
+    for document_id in documents_ids_to_reindex {
+        for result in index.documents_fields.document_fields(writer, document_id)? {
+            let (field_id, bytes) = result?;
+            let value: serde_json::Value = serde_json::from_slice(bytes)?;
+            ram_store.insert((document_id, field_id), value);
+        }
+
+        for ((docid, field_id), value) in ram_store.drain() {
+            serialize_value_with_id(
+                writer,
+                field_id,
+                &schema,
+                docid,
+                index.documents_fields,
+                index.documents_fields_counts,
+                &mut indexer,
+                &mut ranked_map,
+                &value
+            )?;
+        }
+    }
+
+    // 4. write the new index in the main store
+    write_documents_addition_index(
+        writer,
+        index,
+        &ranked_map,
+        number_of_inserted_documents,
+        indexer,
+    )?;
+
+    index.main.put_schema(writer, &schema)?;
+
+    Ok(())
+}
+
+pub fn write_documents_addition_index(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    ranked_map: &RankedMap,
+    number_of_inserted_documents: usize,
+    indexer: RawIndexer,
+) -> MResult<()> {
+    let indexed = indexer.build();
+    let mut delta_words_builder = SetBuilder::memory();
+
+    for (word, delta_set) in indexed.words_doc_indexes {
+        delta_words_builder.insert(&word).unwrap();
+
+        let set = match index.postings_lists.postings_list(writer, &word)? {
+            Some(postings) => Union::new(&postings.matches, &delta_set).into_set_buf(),
+            None => delta_set,
+        };
+
+        index.postings_lists.put_postings_list(writer, &word, &set)?;
+    }
+
+    for (id, words) in indexed.docs_words {
+        index.docs_words.put_doc_words(writer, id, &words)?;
+    }
+
+    let delta_words = delta_words_builder
+        .into_inner()
+        .and_then(fst::Set::from_bytes)
+        .unwrap();
+
+    let words = match index.main.words_fst(writer)? {
+        Some(words) => {
+            let op = OpBuilder::new()
+                .add(words.stream())
+                .add(delta_words.stream())
+                .r#union();
+
+            let mut words_builder = SetBuilder::memory();
+            words_builder.extend_stream(op).unwrap();
+            words_builder
+                .into_inner()
+                .and_then(fst::Set::from_bytes)
+                .unwrap()
+        }
+        None => delta_words,
+    };
+
+    index.main.put_words_fst(writer, &words)?;
+    index.main.put_ranked_map(writer, ranked_map)?;
+    index.main.put_number_of_documents(writer, |old| old + number_of_inserted_documents as u64)?;
+
+    compute_short_prefixes(writer, index)?;
+
+    Ok(())
+}

meilisearch-core/src/update/documents_deletion.rs

@@ -0,0 +1,180 @@
+use std::collections::{BTreeSet, HashMap, HashSet};
+
+use fst::{SetBuilder, Streamer};
+use meilisearch_schema::Schema;
+use sdset::{duo::DifferenceByKey, SetBuf, SetOperation};
+
+use crate::database::{MainT, UpdateT};
+use crate::database::{UpdateEvent, UpdateEventsEmitter};
+use crate::serde::extract_document_id;
+use crate::store;
+use crate::update::{next_update_id, compute_short_prefixes, Update};
+use crate::{DocumentId, Error, MResult, RankedMap};
+
+pub struct DocumentsDeletion {
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    updates_notifier: UpdateEventsEmitter,
+    documents: Vec<DocumentId>,
+}
+
+impl DocumentsDeletion {
+    pub fn new(
+        updates_store: store::Updates,
+        updates_results_store: store::UpdatesResults,
+        updates_notifier: UpdateEventsEmitter,
+    ) -> DocumentsDeletion {
+        DocumentsDeletion {
+            updates_store,
+            updates_results_store,
+            updates_notifier,
+            documents: Vec::new(),
+        }
+    }
+
+    pub fn delete_document_by_id(&mut self, document_id: DocumentId) {
+        self.documents.push(document_id);
+    }
+
+    pub fn delete_document<D>(&mut self, schema: &Schema, document: D) -> MResult<()>
+    where
+        D: serde::Serialize,
+    {
+        let primary_key = schema.primary_key().ok_or(Error::MissingPrimaryKey)?;
+        let document_id = match extract_document_id(&primary_key, &document)? {
+            Some(id) => id,
+            None => return Err(Error::MissingDocumentId),
+        };
+
+        self.delete_document_by_id(document_id);
+
+        Ok(())
+    }
+
+    pub fn finalize(self, writer: &mut heed::RwTxn<UpdateT>) -> MResult<u64> {
+        let _ = self.updates_notifier.send(UpdateEvent::NewUpdate);
+        let update_id = push_documents_deletion(
+            writer,
+            self.updates_store,
+            self.updates_results_store,
+            self.documents,
+        )?;
+        Ok(update_id)
+    }
+}
+
+impl Extend<DocumentId> for DocumentsDeletion {
+    fn extend<T: IntoIterator<Item = DocumentId>>(&mut self, iter: T) {
+        self.documents.extend(iter)
+    }
+}
+
+pub fn push_documents_deletion(
+    writer: &mut heed::RwTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    deletion: Vec<DocumentId>,
+) -> MResult<u64> {
+    let last_update_id = next_update_id(writer, updates_store, updates_results_store)?;
+
+    let update = Update::documents_deletion(deletion);
+    updates_store.put_update(writer, last_update_id, &update)?;
+
+    Ok(last_update_id)
+}
+
+pub fn apply_documents_deletion(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    deletion: Vec<DocumentId>,
+) -> MResult<()> {
+    let idset = SetBuf::from_dirty(deletion);
+
+    let schema = match index.main.schema(writer)? {
+        Some(schema) => schema,
+        None => return Err(Error::SchemaMissing),
+    };
+
+    let mut ranked_map = match index.main.ranked_map(writer)? {
+        Some(ranked_map) => ranked_map,
+        None => RankedMap::default(),
+    };
+
+    // collect the ranked attributes according to the schema
+    let ranked_fields = schema.ranked();
+
+    let mut words_document_ids = HashMap::new();
+    for id in idset {
+        // remove all the ranked attributes from the ranked_map
+        for ranked_attr in ranked_fields {
+            ranked_map.remove(id, *ranked_attr);
+        }
+
+        if let Some(words) = index.docs_words.doc_words(writer, id)? {
+            let mut stream = words.stream();
+            while let Some(word) = stream.next() {
+                let word = word.to_vec();
+                words_document_ids
+                    .entry(word)
+                    .or_insert_with(Vec::new)
+                    .push(id);
+            }
+        }
+    }
+
+    let mut deleted_documents = HashSet::new();
+    let mut removed_words = BTreeSet::new();
+    for (word, document_ids) in words_document_ids {
+        let document_ids = SetBuf::from_dirty(document_ids);
+
+        if let Some(postings) = index.postings_lists.postings_list(writer, &word)? {
+            let op = DifferenceByKey::new(&postings.matches, &document_ids, |d| d.document_id, |id| *id);
+            let doc_indexes = op.into_set_buf();
+
+            if !doc_indexes.is_empty() {
+                index.postings_lists.put_postings_list(writer, &word, &doc_indexes)?;
+            } else {
+                index.postings_lists.del_postings_list(writer, &word)?;
+                removed_words.insert(word);
+            }
+        }
+
+        for id in document_ids {
+            index.documents_fields_counts.del_all_document_fields_counts(writer, id)?;
+            if index.documents_fields.del_all_document_fields(writer, id)? != 0 {
+                deleted_documents.insert(id);
+            }
+        }
+    }
+
+    let deleted_documents_len = deleted_documents.len() as u64;
+    for id in deleted_documents {
+        index.docs_words.del_doc_words(writer, id)?;
+    }
+
+    let removed_words = fst::Set::from_iter(removed_words).unwrap();
+    let words = match index.main.words_fst(writer)? {
+        Some(words_set) => {
+            let op = fst::set::OpBuilder::new()
+                .add(words_set.stream())
+                .add(removed_words.stream())
+                .difference();
+
+            let mut words_builder = SetBuilder::memory();
+            words_builder.extend_stream(op).unwrap();
+            words_builder
+                .into_inner()
+                .and_then(fst::Set::from_bytes)
+                .unwrap()
+        }
+        None => fst::Set::default(),
+    };
+
+    index.main.put_words_fst(writer, &words)?;
+    index.main.put_ranked_map(writer, &ranked_map)?;
+    index.main.put_number_of_documents(writer, |old| old - deleted_documents_len)?;
+
+    compute_short_prefixes(writer, index)?;
+
+    Ok(())
+}

meilisearch-core/src/update/mod.rs

@@ -0,0 +1,361 @@
+mod clear_all;
+mod customs_update;
+mod documents_addition;
+mod documents_deletion;
+mod settings_update;
+
+pub use self::clear_all::{apply_clear_all, push_clear_all};
+pub use self::customs_update::{apply_customs_update, push_customs_update};
+pub use self::documents_addition::{
+    apply_documents_addition, apply_documents_partial_addition, DocumentsAddition,
+};
+pub use self::documents_deletion::{apply_documents_deletion, DocumentsDeletion};
+pub use self::settings_update::{apply_settings_update, push_settings_update};
+
+use std::cmp;
+use std::time::Instant;
+
+use chrono::{DateTime, Utc};
+use fst::{IntoStreamer, Streamer};
+use heed::Result as ZResult;
+use indexmap::IndexMap;
+use log::debug;
+use sdset::Set;
+use serde::{Deserialize, Serialize};
+
+use crate::{store, DocumentId, MResult};
+use crate::database::{MainT, UpdateT};
+use crate::settings::SettingsUpdate;
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Update {
+    data: UpdateData,
+    enqueued_at: DateTime<Utc>,
+}
+
+impl Update {
+    fn clear_all() -> Update {
+        Update {
+            data: UpdateData::ClearAll,
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn customs(data: Vec<u8>) -> Update {
+        Update {
+            data: UpdateData::Customs(data),
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn documents_addition(data: Vec<IndexMap<String, serde_json::Value>>) -> Update {
+        Update {
+            data: UpdateData::DocumentsAddition(data),
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn documents_partial(data: Vec<IndexMap<String, serde_json::Value>>) -> Update {
+        Update {
+            data: UpdateData::DocumentsPartial(data),
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn documents_deletion(data: Vec<DocumentId>) -> Update {
+        Update {
+            data: UpdateData::DocumentsDeletion(data),
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn settings(data: SettingsUpdate) -> Update {
+        Update {
+            data: UpdateData::Settings(data),
+            enqueued_at: Utc::now(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub enum UpdateData {
+    ClearAll,
+    Customs(Vec<u8>),
+    DocumentsAddition(Vec<IndexMap<String, serde_json::Value>>),
+    DocumentsPartial(Vec<IndexMap<String, serde_json::Value>>),
+    DocumentsDeletion(Vec<DocumentId>),
+    Settings(SettingsUpdate)
+}
+
+impl UpdateData {
+    pub fn update_type(&self) -> UpdateType {
+        match self {
+            UpdateData::ClearAll => UpdateType::ClearAll,
+            UpdateData::Customs(_) => UpdateType::Customs,
+            UpdateData::DocumentsAddition(addition) => UpdateType::DocumentsAddition {
+                number: addition.len(),
+            },
+            UpdateData::DocumentsPartial(addition) => UpdateType::DocumentsPartial {
+                number: addition.len(),
+            },
+            UpdateData::DocumentsDeletion(deletion) => UpdateType::DocumentsDeletion {
+                number: deletion.len(),
+            },
+            UpdateData::Settings(update) => UpdateType::Settings {
+                settings: update.clone(),
+            },
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(tag = "name")]
+pub enum UpdateType {
+    ClearAll,
+    Customs,
+    DocumentsAddition { number: usize },
+    DocumentsPartial { number: usize },
+    DocumentsDeletion { number: usize },
+    Settings { settings: SettingsUpdate },
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct ProcessedUpdateResult {
+    pub update_id: u64,
+    #[serde(rename = "type")]
+    pub update_type: UpdateType,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub error: Option<String>,
+    pub duration: f64, // in seconds
+    pub enqueued_at: DateTime<Utc>,
+    pub processed_at: DateTime<Utc>,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct EnqueuedUpdateResult {
+    pub update_id: u64,
+    #[serde(rename = "type")]
+    pub update_type: UpdateType,
+    pub enqueued_at: DateTime<Utc>,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase", tag = "status")]
+pub enum UpdateStatus {
+    Enqueued {
+        #[serde(flatten)]
+        content: EnqueuedUpdateResult,
+    },
+    Failed {
+        #[serde(flatten)]
+        content: ProcessedUpdateResult,
+    },
+    Processed {
+        #[serde(flatten)]
+        content: ProcessedUpdateResult,
+    },
+}
+
+pub fn update_status(
+    update_reader: &heed::RoTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    update_id: u64,
+) -> MResult<Option<UpdateStatus>> {
+    match updates_results_store.update_result(update_reader, update_id)? {
+        Some(result) => {
+            if result.error.is_some() {
+                Ok(Some(UpdateStatus::Failed { content: result }))
+            } else {
+                Ok(Some(UpdateStatus::Processed { content: result }))
+            }
+        },
+        None => match updates_store.get(update_reader, update_id)? {
+            Some(update) => Ok(Some(UpdateStatus::Enqueued {
+                content: EnqueuedUpdateResult {
+                    update_id,
+                    update_type: update.data.update_type(),
+                    enqueued_at: update.enqueued_at,
+                },
+            })),
+            None => Ok(None),
+        },
+    }
+}
+
+pub fn next_update_id(
+    update_writer: &mut heed::RwTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+) -> ZResult<u64> {
+    let last_update = updates_store.last_update(update_writer)?;
+    let last_update = last_update.map(|(n, _)| n);
+
+    let last_update_results_id = updates_results_store.last_update(update_writer)?;
+    let last_update_results_id = last_update_results_id.map(|(n, _)| n);
+
+    let max_update_id = cmp::max(last_update, last_update_results_id);
+    let new_update_id = max_update_id.map_or(0, |n| n + 1);
+
+    Ok(new_update_id)
+}
+
+pub fn update_task<'a, 'b>(
+    writer: &'a mut heed::RwTxn<'b, MainT>,
+    index: &store::Index,
+    update_id: u64,
+    update: Update,
+) -> MResult<ProcessedUpdateResult> {
+    debug!("Processing update number {}", update_id);
+
+    let Update { enqueued_at, data } = update;
+
+    let (update_type, result, duration) = match data {
+        UpdateData::ClearAll => {
+            let start = Instant::now();
+
+            let update_type = UpdateType::ClearAll;
+            let result = apply_clear_all(writer, index);
+
+            (update_type, result, start.elapsed())
+        }
+        UpdateData::Customs(customs) => {
+            let start = Instant::now();
+
+            let update_type = UpdateType::Customs;
+            let result = apply_customs_update(writer, index.main, &customs).map_err(Into::into);
+
+            (update_type, result, start.elapsed())
+        }
+        UpdateData::DocumentsAddition(documents) => {
+            let start = Instant::now();
+
+            let update_type = UpdateType::DocumentsAddition {
+                number: documents.len(),
+            };
+
+            let result = apply_documents_addition(writer, index, documents);
+
+            (update_type, result, start.elapsed())
+        }
+        UpdateData::DocumentsPartial(documents) => {
+            let start = Instant::now();
+
+            let update_type = UpdateType::DocumentsPartial {
+                number: documents.len(),
+            };
+
+            let result = apply_documents_partial_addition(writer, index, documents);
+
+            (update_type, result, start.elapsed())
+        }
+        UpdateData::DocumentsDeletion(documents) => {
+            let start = Instant::now();
+
+            let update_type = UpdateType::DocumentsDeletion {
+                number: documents.len(),
+            };
+
+            let result = apply_documents_deletion(writer, index, documents);
+
+            (update_type, result, start.elapsed())
+        }
+        UpdateData::Settings(settings) => {
+            let start = Instant::now();
+
+            let update_type = UpdateType::Settings {
+                settings: settings.clone(),
+            };
+
+            let result = apply_settings_update(
+                writer,
+                index,
+                settings,
+            );
+
+            (update_type, result, start.elapsed())
+        }
+    };
+
+    debug!(
+        "Processed update number {} {:?} {:?}",
+        update_id, update_type, result
+    );
+
+    let status = ProcessedUpdateResult {
+        update_id,
+        update_type,
+        error: result.map_err(|e| e.to_string()).err(),
+        duration: duration.as_secs_f64(),
+        enqueued_at,
+        processed_at: Utc::now(),
+    };
+
+    Ok(status)
+}
+
+fn compute_short_prefixes(writer: &mut heed::RwTxn<MainT>, index: &store::Index) -> MResult<()> {
+    // retrieve the words fst to compute all those prefixes
+    let words_fst = match index.main.words_fst(writer)? {
+        Some(fst) => fst,
+        None => return Ok(()),
+    };
+
+    // clear the prefixes
+    let pplc_store = index.prefix_postings_lists_cache;
+    pplc_store.clear(writer)?;
+
+    for prefix_len in 1..=2 {
+        // compute prefixes and store those in the PrefixPostingsListsCache store.
+        let mut previous_prefix: Option<([u8; 4], Vec<_>)> = None;
+        let mut stream = words_fst.into_stream();
+        while let Some(input) = stream.next() {
+
+            // We skip the prefixes that are shorter than the current length
+            // we want to cache (<). We must ignore the input when it is exactly the
+            // same word as the prefix because if we match exactly on it we need
+            // to consider it as an exact match and not as a prefix (=).
+            if input.len() <= prefix_len { continue }
+
+            if let Some(postings_list) = index.postings_lists.postings_list(writer, input)?.map(|p| p.matches.into_owned()) {
+                let prefix = &input[..prefix_len];
+
+                let mut arr_prefix = [0; 4];
+                arr_prefix[..prefix_len].copy_from_slice(prefix);
+
+                match previous_prefix {
+                    Some((ref mut prev_prefix, ref mut prev_pl)) if *prev_prefix != arr_prefix => {
+                        prev_pl.sort_unstable();
+                        prev_pl.dedup();
+
+                        if let Ok(prefix) = std::str::from_utf8(&prev_prefix[..prefix_len]) {
+                            debug!("writing the prefix of {:?} of length {}", prefix, prev_pl.len());
+                        }
+
+                        let pls = Set::new_unchecked(&prev_pl);
+                        pplc_store.put_prefix_postings_list(writer, *prev_prefix, &pls)?;
+
+                        *prev_prefix = arr_prefix;
+                        prev_pl.clear();
+                        prev_pl.extend_from_slice(&postings_list);
+                    },
+                    Some((_, ref mut prev_pl)) => prev_pl.extend_from_slice(&postings_list),
+                    None => previous_prefix = Some((arr_prefix, postings_list.to_vec())),
+                }
+            }
+        }
+
+        // write the last prefix postings lists
+        if let Some((prev_prefix, mut prev_pl)) = previous_prefix.take() {
+            prev_pl.sort_unstable();
+            prev_pl.dedup();
+
+            let pls = Set::new_unchecked(&prev_pl);
+            pplc_store.put_prefix_postings_list(writer, prev_prefix, &pls)?;
+        }
+    }
+
+    Ok(())
+}

meilisearch-core/src/update/settings_update.rs

@@ -0,0 +1,301 @@
+use std::collections::{BTreeMap, BTreeSet};
+
+use heed::Result as ZResult;
+use fst::{set::OpBuilder, SetBuilder};
+use sdset::SetBuf;
+use meilisearch_schema::Schema;
+
+use crate::database::{MainT, UpdateT};
+use crate::settings::{UpdateState, SettingsUpdate, RankingRule};
+use crate::update::documents_addition::reindex_all_documents;
+use crate::update::{next_update_id, Update};
+use crate::{store, MResult, Error};
+
+pub fn push_settings_update(
+    writer: &mut heed::RwTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    settings: SettingsUpdate,
+) -> ZResult<u64> {
+    let last_update_id = next_update_id(writer, updates_store, updates_results_store)?;
+
+    let update = Update::settings(settings);
+    updates_store.put_update(writer, last_update_id, &update)?;
+
+    Ok(last_update_id)
+}
+
+pub fn apply_settings_update(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    settings: SettingsUpdate,
+) -> MResult<()> {
+    let mut must_reindex = false;
+
+    let mut schema = match index.main.schema(writer)? {
+        Some(schema) => schema,
+        None => {
+            match settings.primary_key.clone() {
+                UpdateState::Update(id) => Schema::with_primary_key(&id),
+                _ => return Err(Error::MissingPrimaryKey)
+            }
+        }
+    };
+
+    match settings.ranking_rules {
+        UpdateState::Update(v) => {
+            let ranked_field: Vec<&str> = v.iter().filter_map(RankingRule::field).collect();
+            schema.update_ranked(&ranked_field)?;
+            for name in ranked_field {
+                if schema.accept_new_fields() {
+                    schema.set_indexed(name.as_ref())?;
+                    schema.set_displayed(name.as_ref())?;
+                }
+            }
+            index.main.put_ranking_rules(writer, &v)?;
+            must_reindex = true;
+        },
+        UpdateState::Clear => {
+            index.main.delete_ranking_rules(writer)?;
+            schema.clear_ranked();
+            must_reindex = true;
+        },
+        UpdateState::Nothing => (),
+    }
+
+    match settings.distinct_attribute {
+        UpdateState::Update(v) => {
+            index.main.put_distinct_attribute(writer, &v)?;
+        },
+        UpdateState::Clear => {
+            index.main.delete_distinct_attribute(writer)?;
+        },
+        UpdateState::Nothing => (),
+    }
+
+    match settings.accept_new_fields {
+        UpdateState::Update(v) => {
+            schema.set_accept_new_fields(v);
+        },
+        UpdateState::Clear => {
+            schema.set_accept_new_fields(true);
+        },
+        UpdateState::Nothing => (),
+    }
+
+    match settings.searchable_attributes.clone() {
+        UpdateState::Update(v) => {
+            schema.update_indexed(v)?;
+            must_reindex = true;
+        },
+        UpdateState::Clear => {
+            schema.set_all_fields_as_indexed();
+            must_reindex = true;
+        },
+        UpdateState::Nothing => (),
+    }
+    match settings.displayed_attributes.clone() {
+        UpdateState::Update(v) => schema.update_displayed(v)?,
+        UpdateState::Clear => {
+            schema.set_all_fields_as_displayed();
+        },
+        UpdateState::Nothing => (),
+    }
+
+    index.main.put_schema(writer, &schema)?;
+
+    match settings.stop_words {
+        UpdateState::Update(stop_words) => {
+            if apply_stop_words_update(writer, index, stop_words)? {
+                must_reindex = true;
+            }
+        },
+        UpdateState::Clear => {
+            if apply_stop_words_update(writer, index, BTreeSet::new())? {
+                must_reindex = true;
+            }
+        },
+        UpdateState::Nothing => (),
+    }
+
+    match settings.synonyms {
+        UpdateState::Update(synonyms) => apply_synonyms_update(writer, index, synonyms)?,
+        UpdateState::Clear => apply_synonyms_update(writer, index, BTreeMap::new())?,
+        UpdateState::Nothing => (),
+    }
+
+    if must_reindex {
+        reindex_all_documents(writer, index)?;
+    }
+
+    Ok(())
+}
+
+pub fn apply_stop_words_update(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    stop_words: BTreeSet<String>,
+) -> MResult<bool>
+{
+    let mut must_reindex = false;
+
+    let old_stop_words: BTreeSet<String> = index.main
+        .stop_words_fst(writer)?
+        .unwrap_or_default()
+        .stream()
+        .into_strs()?
+        .into_iter()
+        .collect();
+
+    let deletion: BTreeSet<String> = old_stop_words.difference(&stop_words).cloned().collect();
+    let addition: BTreeSet<String> = stop_words.difference(&old_stop_words).cloned().collect();
+
+    if !addition.is_empty() {
+        apply_stop_words_addition(writer, index, addition)?;
+    }
+
+    if !deletion.is_empty() {
+        must_reindex = true;
+        apply_stop_words_deletion(writer, index, deletion)?;
+    }
+
+    if let Some(words_fst) = index.main.words_fst(writer)? {
+        let stop_words = fst::Set::from_iter(stop_words)?;
+        let op = OpBuilder::new()
+            .add(&words_fst)
+            .add(&stop_words)
+            .difference();
+
+        let mut builder = fst::SetBuilder::memory();
+        builder.extend_stream(op)?;
+        let words_fst = builder.into_inner().and_then(fst::Set::from_bytes)?;
+
+        index.main.put_words_fst(writer, &words_fst)?;
+        index.main.put_stop_words_fst(writer, &stop_words)?;
+    }
+
+    Ok(must_reindex)
+}
+
+fn apply_stop_words_addition(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    addition: BTreeSet<String>,
+) -> MResult<()>
+{
+    let main_store = index.main;
+    let postings_lists_store = index.postings_lists;
+
+    let mut stop_words_builder = SetBuilder::memory();
+
+    for word in addition {
+        stop_words_builder.insert(&word)?;
+        // we remove every posting list associated to a new stop word
+        postings_lists_store.del_postings_list(writer, word.as_bytes())?;
+    }
+
+    // create the new delta stop words fst
+    let delta_stop_words = stop_words_builder
+        .into_inner()
+        .and_then(fst::Set::from_bytes)?;
+
+    // we also need to remove all the stop words from the main fst
+    if let Some(word_fst) = main_store.words_fst(writer)? {
+        let op = OpBuilder::new()
+            .add(&word_fst)
+            .add(&delta_stop_words)
+            .difference();
+
+        let mut word_fst_builder = SetBuilder::memory();
+        word_fst_builder.extend_stream(op)?;
+        let word_fst = word_fst_builder
+            .into_inner()
+            .and_then(fst::Set::from_bytes)?;
+
+        main_store.put_words_fst(writer, &word_fst)?;
+    }
+
+    // now we add all of these stop words from the main store
+    let stop_words_fst = main_store.stop_words_fst(writer)?.unwrap_or_default();
+
+    let op = OpBuilder::new()
+        .add(&stop_words_fst)
+        .add(&delta_stop_words)
+        .r#union();
+
+    let mut stop_words_builder = SetBuilder::memory();
+    stop_words_builder.extend_stream(op)?;
+    let stop_words_fst = stop_words_builder
+        .into_inner()
+        .and_then(fst::Set::from_bytes)?;
+
+    main_store.put_stop_words_fst(writer, &stop_words_fst)?;
+
+    Ok(())
+}
+
+fn apply_stop_words_deletion(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    deletion: BTreeSet<String>,
+) -> MResult<()> {
+
+    let mut stop_words_builder = SetBuilder::memory();
+
+    for word in deletion {
+        stop_words_builder.insert(&word)?;
+    }
+
+    // create the new delta stop words fst
+    let delta_stop_words = stop_words_builder
+        .into_inner()
+        .and_then(fst::Set::from_bytes)?;
+
+    // now we delete all of these stop words from the main store
+    let stop_words_fst = index.main.stop_words_fst(writer)?.unwrap_or_default();
+
+    let op = OpBuilder::new()
+        .add(&stop_words_fst)
+        .add(&delta_stop_words)
+        .difference();
+
+    let mut stop_words_builder = SetBuilder::memory();
+    stop_words_builder.extend_stream(op)?;
+    let stop_words_fst = stop_words_builder.into_inner().and_then(fst::Set::from_bytes)?;
+
+    Ok(index.main.put_stop_words_fst(writer, &stop_words_fst)?)
+}
+
+pub fn apply_synonyms_update(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    synonyms: BTreeMap<String, Vec<String>>,
+) -> MResult<()> {
+
+    let main_store = index.main;
+    let synonyms_store = index.synonyms;
+
+    let mut synonyms_builder = SetBuilder::memory();
+    synonyms_store.clear(writer)?;
+    for (word, alternatives) in synonyms.clone() {
+        synonyms_builder.insert(&word)?;
+
+        let alternatives = {
+            let alternatives = SetBuf::from_dirty(alternatives);
+            let mut alternatives_builder = SetBuilder::memory();
+            alternatives_builder.extend_iter(alternatives)?;
+            let bytes = alternatives_builder.into_inner()?;
+            fst::Set::from_bytes(bytes)?
+        };
+
+        synonyms_store.put_synonyms(writer, word.as_bytes(), &alternatives)?;
+    }
+
+    let synonyms_set = synonyms_builder
+        .into_inner()
+        .and_then(fst::Set::from_bytes)?;
+
+    main_store.put_synonyms_fst(writer, &synonyms_set)?;
+
+    Ok(())
+}

meilisearch-http/Cargo.toml

@@ -0,0 +1,61 @@
+[package]
+name = "meilisearch-http"
+description = "MeiliSearch HTTP server"
+version = "0.10.1"
+license = "MIT"
+authors = [
+    "Quentin de Quelen <quentin@dequelen.me>",
+    "Clément Renault <clement@meilisearch.com>",
+]
+edition = "2018"
+
+[[bin]]
+name = "meilisearch"
+path = "src/main.rs"
+
+[dependencies]
+async-std = { version = "1.5.0", features = ["attributes"] }
+chrono = { version = "0.4.11", features = ["serde"] }
+crossbeam-channel = "0.4.2"
+env_logger = "0.7.1"
+futures = "0.3.4"
+heed = "0.7.0"
+http = "0.1.19"
+indexmap = { version = "1.3.2", features = ["serde-1"] }
+log = "0.4.8"
+main_error = "0.1.0"
+meilisearch-core = { path = "../meilisearch-core", version = "0.10.1" }
+meilisearch-schema = { path = "../meilisearch-schema", version = "0.10.1" }
+meilisearch-tokenizer = {path = "../meilisearch-tokenizer", version = "0.10.1"}
+mime = "0.3.16"
+pretty-bytes = "0.2.2"
+rand = "0.7.3"
+rayon = "1.3.0"
+serde = { version = "1.0.105", features = ["derive"] }
+serde_json = { version = "1.0.50", features = ["preserve_order"] }
+serde_qs = "0.5.2"
+sha2 = "0.8.1"
+siphasher = "0.3.2"
+structopt = "0.3.12"
+sysinfo = "0.12.0"
+tide = "0.6.0"
+ureq = { version = "0.12.0", features = ["tls"], default-features = false }
+walkdir = "2.3.1"
+whoami = "0.8.1"
+slice-group-by = "0.2.6"
+
+[dev-dependencies]
+http-service = "0.4.0"
+http-service-mock = "0.4.0"
+tempdir = "0.3.7"
+once_cell = "1.3.1"
+
+[dev-dependencies.assert-json-diff]
+git = "https://github.com/qdequele/assert-json-diff"
+branch = "master"
+
+[build-dependencies]
+vergen = "3.1.0"
+
+[target.'cfg(unix)'.dependencies]
+jemallocator = "0.3.2"

meilisearch-http/build.rs

@@ -0,0 +1,10 @@
+use vergen::{generate_cargo_keys, ConstantsFlags};
+
+fn main() {
+    // Setup the flags, toggling off the 'SEMVER_FROM_CARGO_PKG' flag
+    let mut flags = ConstantsFlags::all();
+    flags.toggle(ConstantsFlags::SEMVER_FROM_CARGO_PKG);
+
+    // Generate the 'cargo:' key output
+    generate_cargo_keys(ConstantsFlags::all()).expect("Unable to generate the cargo keys!");
+}

meilisearch-http/public/interface.html

@@ -0,0 +1,270 @@
+<!DOCTYPE html>
+<html>
+  <head>
+    <meta charset="utf-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1">
+    <link rel="stylesheet" href="/bulma.min.css">
+    <title>MeiliSearch</title>
+    <style>
+      em {
+        color: hsl(204, 86%, 25%);
+        font-style: inherit;
+        background-color: hsl(204, 86%, 88%);
+      }
+
+      #results {
+        max-width: 900px;
+        margin: 20px auto 0 auto;
+        padding: 0;
+      }
+
+      .notification {
+        display: flex;
+        justify-content: center;
+      }
+
+      .level-left {
+        margin-right: 50px;
+      }
+
+      .document {
+        padding: 20px 20px;
+        background-color: #f5f5f5;
+        border-radius: 4px;
+        margin-bottom: 20px;
+        display: flex;
+      }
+
+      .document ol {
+        flex: 0 0 75%;
+        max-width: 75%;
+        padding: 0;
+        margin: 0;
+      }
+
+      .document .image {
+        max-width: 25%;
+        flex: 0 0 25%;
+        padding-left: 30px;
+        box-sizing: border-box;
+      }
+
+      .document .image img {
+        width: 100%;
+      }
+
+      .field {
+        list-style-type: none;
+        display: flex;
+        flex-wrap: wrap;
+      }
+
+      .field:not(:last-child) {
+        margin-bottom: 7px;
+      }
+
+      .attribute {
+        flex: 0 0 25%;
+        max-width: 25%;
+        text-align: right;
+        padding-right: 10px;
+        box-sizing: border-box;
+        text-transform: uppercase;
+        color: rgba(0,0,0,.7);
+      }
+
+      .content {
+        max-width: 75%;
+        flex: 0 0 75%;
+        box-sizing: border-box;
+        padding-left: 10px;
+        color: rgba(0,0,0,.9);
+        overflow-wrap: break-word;
+      }
+    </style>
+  </head>
+  <body>
+
+    <section class="hero is-light">
+      <div class="hero-body">
+        <div class="container">
+          <h1 class="title">
+            Welcome to MeiliSearch
+          </h1>
+          <h2 class="subtitle">
+            This dashboard will help you check the search results with ease.
+          </h2>
+        </div>
+      </div>
+    </section>
+
+    <section class="hero container">
+        <div class="notification" style="border-radius: 0 0 4px 4px;">
+
+          <nav class="level">
+            <!-- Left side -->
+            <div class="level-left">
+              <div class="level-item">
+                <div class="field has-addons has-addons-right">
+                  <p class="control">
+                    <span class="select">
+                      <select id="index">
+                        <!-- indexes names -->
+                      </select>
+                    </span>
+                  </p>
+                  <p class="control">
+                    <input id="search" class="input" type="text" autofocus placeholder="e.g. George Clooney">
+                  </p>
+                </div>
+              </div>
+            </div>
+
+            <!-- Right side -->
+            <nav class="level-right">
+              <div class="level-item has-text-centered">
+                <div>
+                  <p class="heading">Documents</p>
+                  <p id="count" class="title">25</p>
+                </div>
+              </div>
+              <div class="level-item has-text-centered">
+                <div>
+                  <p class="heading">Time Spent</p>
+                  <p id="time" class="title">4ms</p>
+                </div>
+              </div>
+            </nav>
+          </nav>
+
+        </div>
+    </section>
+
+    <section>
+      <ol id="results" class="content">
+        <!-- documents matching resquests -->
+      </ol>
+    </section>
+  </body>
+
+  <script>
+    function sanitizeHTMLEntities(str) {
+        if (str && typeof str === 'string') {
+            str = str.replace(/</g,"&lt;");
+            str = str.replace(/>/g,"&gt;");
+            str = str.replace(/&lt;em&gt;/g,"<em>");
+            str = str.replace(/&lt;\/em&gt;/g,"<\/em>");
+        }
+        return str;
+    }
+
+    function httpGet(theUrl) {
+        var xmlHttp = new XMLHttpRequest();
+        xmlHttp.open("GET", theUrl, false); // false for synchronous request
+        xmlHttp.send(null);
+        return xmlHttp.responseText;
+    }
+
+    let lastRequest = undefined;
+
+    function triggerSearch() {
+        var e = document.getElementById("index");
+        if (e.selectedIndex == -1) { return }
+        var index = e.options[e.selectedIndex].value;
+
+        let theUrl = `${baseUrl}/indexes/${index}/search?q=${search.value}&attributesToHighlight=*`;
+
+        if (lastRequest) { lastRequest.abort() }
+        lastRequest = new XMLHttpRequest();
+
+        lastRequest.open("GET", theUrl, true);
+        lastRequest.onload = function (e) {
+            if (lastRequest.readyState === 4 && lastRequest.status === 200) {
+                let sanitizedResponseText = sanitizeHTMLEntities(lastRequest.responseText);
+                let httpResults = JSON.parse(sanitizedResponseText);
+                results.innerHTML = '';
+
+                let processingTimeMs = httpResults.processingTimeMs;
+                let numberOfDocuments = httpResults.hits.length;
+                time.innerHTML = `${processingTimeMs}ms`;
+                count.innerHTML = `${numberOfDocuments}`;
+
+                for (result of httpResults.hits) {
+                    const element = {...result, ...result._formatted };
+                    delete element._formatted;
+
+                    const elem = document.createElement('li');
+                    elem.classList.add("document");
+
+                    const ol = document.createElement('ol');
+                    let image = undefined;
+
+                    for (const prop in element) {
+                        // Check if property is an image url link.
+                        if (typeof result[prop] === 'string') {
+                            if (image == undefined && result[prop].match(/^(https|http):\/\/.*(jpe?g|png|gif)(\?.*)?$/g)) {
+                                image = result[prop];
+                            }
+                        }
+
+                        const field = document.createElement('li');
+                        field.classList.add("field");
+
+                        const attribute = document.createElement('div');
+                        attribute.classList.add("attribute");
+                        attribute.innerHTML = prop;
+
+                        const content = document.createElement('div');
+                        content.classList.add("content");
+                        if (typeof (element[prop]) === "object") {
+                          content.innerHTML = JSON.stringify(element[prop]);
+                        } else {
+                          content.innerHTML = element[prop];
+                        }
+
+                        field.appendChild(attribute);
+                        field.appendChild(content);
+
+                        ol.appendChild(field);
+                    }
+
+                    elem.appendChild(ol);
+
+                    if (image != undefined) {
+                        const div = document.createElement('div');
+                        div.classList.add("image");
+
+                        const img = document.createElement('img');
+                        img.src = image;
+
+                        div.appendChild(img);
+                        elem.appendChild(div);
+                    }
+
+                    results.appendChild(elem)
+                }
+            } else {
+                console.error(lastRequest.statusText);
+            }
+        };
+        lastRequest.send(null);
+    }
+
+    let baseUrl = window.location.origin;
+    // TODO we must not block here
+    let result = JSON.parse(httpGet(`${baseUrl}/indexes`));
+
+    let select = document.getElementById("index");
+    for (index of result) {
+        const option = document.createElement('option');
+        option.value = index.uid;
+        option.innerHTML = index.name;
+        select.appendChild(option);
+    }
+
+    search.oninput = triggerSearch;
+    select.onchange = triggerSearch;
+
+    triggerSearch();
+  </script>
+</html>

meilisearch-http/src/analytics.rs

@@ -0,0 +1,69 @@
+use std::hash::{Hash, Hasher};
+use std::thread;
+use std::time::{Duration, SystemTime, UNIX_EPOCH};
+
+use log::error;
+use serde::Serialize;
+use serde_qs as qs;
+use siphasher::sip::SipHasher;
+
+const AMPLITUDE_API_KEY: &str = "f7fba398780e06d8fe6666a9be7e3d47";
+
+#[derive(Debug, Serialize)]
+struct Event<'a> {
+    user_id: &'a str,
+    event_type: &'a str,
+    device_id: &'a str,
+    time: u64,
+}
+
+#[derive(Debug, Serialize)]
+struct AmplitudeRequest<'a> {
+    api_key: &'a str,
+    event: &'a str,
+}
+
+pub fn analytics_sender() {
+    let username = whoami::username();
+    let hostname = whoami::hostname();
+    let platform = whoami::platform();
+
+    let uid = username + &hostname + &platform.to_string();
+
+    let mut hasher = SipHasher::new();
+    uid.hash(&mut hasher);
+    let hash = hasher.finish();
+
+    let uid = format!("{:X}", hash);
+    let platform = platform.to_string();
+
+    loop {
+        let n = SystemTime::now().duration_since(UNIX_EPOCH).unwrap();
+        let user_id = &uid;
+        let device_id = &platform;
+        let time = n.as_secs();
+        let event_type = "runtime_tick";
+
+        let event = Event {
+            user_id,
+            event_type,
+            device_id,
+            time,
+        };
+        let event = serde_json::to_string(&event).unwrap();
+
+        let request = AmplitudeRequest {
+            api_key: AMPLITUDE_API_KEY,
+            event: &event,
+        };
+
+        let body = qs::to_string(&request).unwrap();
+        let response = ureq::post("https://api.amplitude.com/httpapi").send_string(&body);
+        if !response.ok() {
+            let body = response.into_string().unwrap();
+            error!("Unsuccessful call to Amplitude: {}", body);
+        }
+
+        thread::sleep(Duration::from_secs(86_400)) // one day
+    }
+}

meilisearch-http/src/data.rs

@@ -0,0 +1,163 @@
+use std::collections::HashMap;
+use std::ops::Deref;
+use std::sync::Arc;
+
+use chrono::{DateTime, Utc};
+use heed::types::{SerdeBincode, Str};
+use log::error;
+use meilisearch_core::{Database, Error as MError, MResult, MainT, UpdateT};
+use sha2::Digest;
+use sysinfo::Pid;
+
+use crate::option::Opt;
+use crate::routes::index::index_update_callback;
+
+const LAST_UPDATE_KEY: &str = "last-update";
+
+type SerdeDatetime = SerdeBincode<DateTime<Utc>>;
+
+#[derive(Clone)]
+pub struct Data {
+    inner: Arc<DataInner>,
+}
+
+impl Deref for Data {
+    type Target = DataInner;
+
+    fn deref(&self) -> &Self::Target {
+        &self.inner
+    }
+}
+
+#[derive(Clone)]
+pub struct DataInner {
+    pub db: Arc<Database>,
+    pub db_path: String,
+    pub api_keys: ApiKeys,
+    pub server_pid: Pid,
+}
+
+#[derive(Default, Clone)]
+pub struct ApiKeys {
+    pub public: Option<String>,
+    pub private: Option<String>,
+    pub master: Option<String>,
+}
+
+impl ApiKeys {
+    pub fn generate_missing_api_keys(&mut self) {
+        if let Some(master_key) = &self.master {
+            if self.private.is_none() {
+                let key = format!("{}-private", master_key);
+                let sha = sha2::Sha256::digest(key.as_bytes());
+                self.private = Some(format!("{:x}", sha));
+            }
+            if self.public.is_none() {
+                let key = format!("{}-public", master_key);
+                let sha = sha2::Sha256::digest(key.as_bytes());
+                self.public = Some(format!("{:x}", sha));
+            }
+        }
+    }
+}
+
+impl DataInner {
+    pub fn is_indexing(&self, reader: &heed::RoTxn<UpdateT>, index: &str) -> MResult<Option<bool>> {
+        match self.db.open_index(&index) {
+            Some(index) => index.current_update_id(&reader).map(|u| Some(u.is_some())),
+            None => Ok(None),
+        }
+    }
+
+    pub fn last_update(&self, reader: &heed::RoTxn<MainT>) -> MResult<Option<DateTime<Utc>>> {
+        match self
+            .db
+            .common_store()
+            .get::<_, Str, SerdeDatetime>(reader, LAST_UPDATE_KEY)?
+        {
+            Some(datetime) => Ok(Some(datetime)),
+            None => Ok(None),
+        }
+    }
+
+    pub fn set_last_update(&self, writer: &mut heed::RwTxn<MainT>) -> MResult<()> {
+        self.db
+            .common_store()
+            .put::<_, Str, SerdeDatetime>(writer, LAST_UPDATE_KEY, &Utc::now())
+            .map_err(Into::into)
+    }
+
+    pub fn compute_stats(&self, writer: &mut heed::RwTxn<MainT>, index_uid: &str) -> MResult<()> {
+        let index = match self.db.open_index(&index_uid) {
+            Some(index) => index,
+            None => {
+                error!("Impossible to retrieve index {}", index_uid);
+                return Ok(());
+            }
+        };
+
+        let schema = match index.main.schema(&writer)? {
+            Some(schema) => schema,
+            None => return Ok(()),
+        };
+
+        let all_documents_fields = index
+            .documents_fields_counts
+            .all_documents_fields_counts(&writer)?;
+
+        // count fields frequencies
+        let mut fields_frequency = HashMap::<_, usize>::new();
+        for result in all_documents_fields {
+            let (_, attr, _) = result?;
+            if let Some(field_id) = schema.indexed_pos_to_field_id(attr) {
+                *fields_frequency.entry(field_id).or_default() += 1;
+            }
+        }
+
+        // convert attributes to their names
+        let frequency: HashMap<_, _> = fields_frequency
+            .into_iter()
+            .filter_map(|(a, c)| schema.name(a).map(|name| (name.to_string(), c)))
+            .collect();
+
+        index
+            .main
+            .put_fields_frequency(writer, &frequency)
+            .map_err(MError::Zlmdb)
+    }
+}
+
+impl Data {
+    pub fn new(opt: Opt) -> Data {
+        let db_path = opt.db_path.clone();
+        let server_pid = sysinfo::get_current_pid().unwrap();
+
+        let db = Arc::new(Database::open_or_create(opt.db_path).unwrap());
+
+        let mut api_keys = ApiKeys {
+            master: opt.master_key.clone(),
+            private: None,
+            public: None,
+        };
+
+        api_keys.generate_missing_api_keys();
+
+        let inner_data = DataInner {
+            db: db.clone(),
+            db_path,
+            api_keys,
+            server_pid,
+        };
+
+        let data = Data {
+            inner: Arc::new(inner_data),
+        };
+
+        let callback_context = data.clone();
+        db.set_update_callback(Box::new(move |index_uid, status| {
+            index_update_callback(&index_uid, &callback_context, status);
+        }));
+
+        data
+    }
+}

meilisearch-http/src/error.rs

@@ -0,0 +1,191 @@
+use std::fmt::Display;
+
+use http::status::StatusCode;
+use log::{error, warn};
+use meilisearch_core::{FstError, HeedError};
+use serde::{Deserialize, Serialize};
+use tide::IntoResponse;
+use tide::Response;
+
+use crate::helpers::meilisearch::Error as SearchError;
+
+pub type SResult<T> = Result<T, ResponseError>;
+
+pub enum ResponseError {
+    Internal(String),
+    BadRequest(String),
+    InvalidToken(String),
+    NotFound(String),
+    IndexNotFound(String),
+    DocumentNotFound(String),
+    MissingHeader(String),
+    FilterParsing(String),
+    BadParameter(String, String),
+    OpenIndex(String),
+    CreateIndex(String),
+    InvalidIndexUid,
+    Maintenance,
+}
+
+impl ResponseError {
+    pub fn internal(message: impl Display) -> ResponseError {
+        ResponseError::Internal(message.to_string())
+    }
+
+    pub fn bad_request(message: impl Display) -> ResponseError {
+        ResponseError::BadRequest(message.to_string())
+    }
+
+    pub fn invalid_token(message: impl Display) -> ResponseError {
+        ResponseError::InvalidToken(message.to_string())
+    }
+
+    pub fn not_found(message: impl Display) -> ResponseError {
+        ResponseError::NotFound(message.to_string())
+    }
+
+    pub fn index_not_found(message: impl Display) -> ResponseError {
+        ResponseError::IndexNotFound(message.to_string())
+    }
+
+    pub fn document_not_found(message: impl Display) -> ResponseError {
+        ResponseError::DocumentNotFound(message.to_string())
+    }
+
+    pub fn missing_header(message: impl Display) -> ResponseError {
+        ResponseError::MissingHeader(message.to_string())
+    }
+
+    pub fn bad_parameter(name: impl Display, message: impl Display) -> ResponseError {
+        ResponseError::BadParameter(name.to_string(), message.to_string())
+    }
+
+    pub fn open_index(message: impl Display) -> ResponseError {
+        ResponseError::OpenIndex(message.to_string())
+    }
+
+    pub fn create_index(message: impl Display) -> ResponseError {
+        ResponseError::CreateIndex(message.to_string())
+    }
+}
+
+impl IntoResponse for ResponseError {
+    fn into_response(self) -> Response {
+        match self {
+            ResponseError::Internal(err) => {
+                error!("internal server error: {}", err);
+                error("Internal server error".to_string(),
+                    StatusCode::INTERNAL_SERVER_ERROR,
+                )
+            }
+            ResponseError::FilterParsing(err) => {
+                warn!("error paring filter: {}", err);
+                error(format!("parsing error: {}", err),
+                StatusCode::BAD_REQUEST)
+            }
+            ResponseError::BadRequest(err) => {
+                warn!("bad request: {}", err);
+                error(err, StatusCode::BAD_REQUEST)
+            }
+            ResponseError::InvalidToken(err) => {
+                error(format!("Invalid API key: {}", err), StatusCode::FORBIDDEN)
+            }
+            ResponseError::NotFound(err) => error(err, StatusCode::NOT_FOUND),
+            ResponseError::IndexNotFound(index) => {
+                error(format!("Index {} not found", index), StatusCode::NOT_FOUND)
+            }
+            ResponseError::DocumentNotFound(id) => error(
+                format!("Document with id {} not found", id),
+                StatusCode::NOT_FOUND,
+            ),
+            ResponseError::MissingHeader(header) => error(
+                format!("Header {} is missing", header),
+                StatusCode::UNAUTHORIZED,
+            ),
+            ResponseError::BadParameter(param, e) => error(
+                format!("Url parameter {} error: {}", param, e),
+                StatusCode::BAD_REQUEST,
+            ),
+            ResponseError::CreateIndex(err) => error(
+                format!("Impossible to create index; {}", err),
+                StatusCode::BAD_REQUEST,
+            ),
+            ResponseError::OpenIndex(err) => error(
+                format!("Impossible to open index; {}", err),
+                StatusCode::BAD_REQUEST,
+            ),
+            ResponseError::InvalidIndexUid => error(
+                "Index must have a valid uid; Index uid can be of type integer or string only composed of alphanumeric characters, hyphens (-) and underscores (_).".to_string(),
+                StatusCode::BAD_REQUEST,
+            ),
+            ResponseError::Maintenance => error(
+                String::from("Server is in maintenance, please try again later"),
+                StatusCode::SERVICE_UNAVAILABLE,
+            ),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize)]
+struct ErrorMessage {
+    message: String,
+}
+
+fn error(message: String, status: StatusCode) -> Response {
+    let message = ErrorMessage { message };
+    tide::Response::new(status.as_u16())
+        .body_json(&message)
+        .unwrap()
+}
+
+impl From<serde_json::Error> for ResponseError {
+    fn from(err: serde_json::Error) -> ResponseError {
+        ResponseError::internal(err)
+    }
+}
+
+impl From<meilisearch_core::Error> for ResponseError {
+    fn from(err: meilisearch_core::Error) -> ResponseError {
+        ResponseError::internal(err)
+    }
+}
+
+impl From<HeedError> for ResponseError {
+    fn from(err: HeedError) -> ResponseError {
+        ResponseError::internal(err)
+    }
+}
+
+impl From<FstError> for ResponseError {
+    fn from(err: FstError) -> ResponseError {
+        ResponseError::internal(err)
+    }
+}
+
+impl From<SearchError> for ResponseError {
+    fn from(err: SearchError) -> ResponseError {
+        match err {
+            SearchError::FilterParsing(s) => ResponseError::FilterParsing(s),
+            _ => ResponseError::internal(err),
+        }
+    }
+}
+
+impl From<meilisearch_core::settings::RankingRuleConversionError> for ResponseError {
+    fn from(err: meilisearch_core::settings::RankingRuleConversionError) -> ResponseError {
+        ResponseError::internal(err)
+    }
+}
+
+pub trait IntoInternalError<T> {
+    fn into_internal_error(self) -> SResult<T>;
+}
+
+impl<T> IntoInternalError<T> for Option<T> {
+    fn into_internal_error(self) -> SResult<T> {
+        match self {
+            Some(value) => Ok(value),
+            None => Err(ResponseError::internal("Heed cannot find requested value")),
+        }
+    }
+}

meilisearch-http/src/helpers/meilisearch.rs

@@ -0,0 +1,697 @@
+use std::cmp::Ordering;
+use std::collections::{HashMap, HashSet};
+use std::convert::From;
+use std::error;
+use std::fmt;
+use std::hash::{Hash, Hasher};
+use std::time::{Duration, Instant};
+
+use indexmap::IndexMap;
+use log::error;
+use meilisearch_core::Filter;
+use meilisearch_core::criterion::*;
+use meilisearch_core::settings::RankingRule;
+use meilisearch_core::{Highlight, Index, MainT, RankedMap};
+use meilisearch_schema::{FieldId, Schema};
+use meilisearch_tokenizer::is_cjk;
+use serde::{Deserialize, Serialize};
+use serde_json::Value;
+use siphasher::sip::SipHasher;
+use slice_group_by::GroupBy;
+
+#[derive(Debug)]
+pub enum Error {
+    SearchDocuments(String),
+    RetrieveDocument(u64, String),
+    DocumentNotFound(u64),
+    CropFieldWrongType(String),
+    FilterParsing(String),
+    AttributeNotFoundOnDocument(String),
+    AttributeNotFoundOnSchema(String),
+    MissingFilterValue,
+    UnknownFilteredAttribute,
+    Internal(String),
+}
+
+impl error::Error for Error {}
+
+impl fmt::Display for Error {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        use Error::*;
+
+        match self {
+            SearchDocuments(err) => write!(f, "impossible to search documents; {}", err),
+            RetrieveDocument(id, err) => write!(
+                f,
+                "impossible to retrieve the document with id: {}; {}",
+                id, err
+            ),
+            DocumentNotFound(id) => write!(f, "document {} not found", id),
+            CropFieldWrongType(field) => {
+                write!(f, "the field {} cannot be cropped it's not a string", field)
+            }
+            AttributeNotFoundOnDocument(field) => {
+                write!(f, "field {} is not found on document", field)
+            }
+            AttributeNotFoundOnSchema(field) => write!(f, "field {} is not found on schema", field),
+            MissingFilterValue => f.write_str("a filter doesn't have a value to compare it with"),
+            UnknownFilteredAttribute => {
+                f.write_str("a filter is specifying an unknown schema attribute")
+            }
+            Internal(err) => write!(f, "internal error; {}", err),
+            FilterParsing(err) => write!(f, "filter parsing error: {}", err),
+        }
+    }
+}
+
+impl From<meilisearch_core::Error> for Error {
+    fn from(error: meilisearch_core::Error) -> Self {
+        use meilisearch_core::pest_error::LineColLocation::*;
+        match error {
+            meilisearch_core::Error::FilterParseError(e) => {
+                let (line, column) = match e.line_col {
+                    Span((line, _), (column, _)) => (line, column),
+                    Pos((line, column)) => (line, column),
+                };
+                let message = format!("parsing error on line {} at column {}: {}", line, column, e.variant.message());
+
+                Error::FilterParsing(message) 
+            },
+            _ => Error::Internal(error.to_string()),
+        }
+    }
+}
+
+impl From<heed::Error> for Error {
+    fn from(error: heed::Error) -> Self {
+        Error::Internal(error.to_string())
+    }
+}
+
+pub trait IndexSearchExt {
+    fn new_search(&self, query: String) -> SearchBuilder;
+}
+
+impl IndexSearchExt for Index {
+    fn new_search(&self, query: String) -> SearchBuilder {
+        SearchBuilder {
+            index: self,
+            query,
+            offset: 0,
+            limit: 20,
+            attributes_to_crop: None,
+            attributes_to_retrieve: None,
+            attributes_to_highlight: None,
+            filters: None,
+            timeout: Duration::from_millis(30),
+            matches: false,
+        }
+    }
+}
+
+pub struct SearchBuilder<'a> {
+    index: &'a Index,
+    query: String,
+    offset: usize,
+    limit: usize,
+    attributes_to_crop: Option<HashMap<String, usize>>,
+    attributes_to_retrieve: Option<HashSet<String>>,
+    attributes_to_highlight: Option<HashSet<String>>,
+    filters: Option<String>,
+    timeout: Duration,
+    matches: bool,
+}
+
+impl<'a> SearchBuilder<'a> {
+    pub fn offset(&mut self, value: usize) -> &SearchBuilder {
+        self.offset = value;
+        self
+    }
+
+    pub fn limit(&mut self, value: usize) -> &SearchBuilder {
+        self.limit = value;
+        self
+    }
+
+    pub fn attributes_to_crop(&mut self, value: HashMap<String, usize>) -> &SearchBuilder {
+        self.attributes_to_crop = Some(value);
+        self
+    }
+
+    pub fn attributes_to_retrieve(&mut self, value: HashSet<String>) -> &SearchBuilder {
+        self.attributes_to_retrieve = Some(value);
+        self
+    }
+
+    pub fn add_retrievable_field(&mut self, value: String) -> &SearchBuilder {
+        let attributes_to_retrieve = self.attributes_to_retrieve.get_or_insert(HashSet::new());
+        attributes_to_retrieve.insert(value);
+        self
+    }
+
+    pub fn attributes_to_highlight(&mut self, value: HashSet<String>) -> &SearchBuilder {
+        self.attributes_to_highlight = Some(value);
+        self
+    }
+
+    pub fn filters(&mut self, value: String) -> &SearchBuilder {
+        self.filters = Some(value);
+        self
+    }
+
+    pub fn timeout(&mut self, value: Duration) -> &SearchBuilder {
+        self.timeout = value;
+        self
+    }
+
+    pub fn get_matches(&mut self) -> &SearchBuilder {
+        self.matches = true;
+        self
+    }
+
+    pub fn search(&self, reader: &heed::RoTxn<MainT>) -> Result<SearchResult, Error> {
+        let schema = self.index.main.schema(reader);
+        let schema = schema.map_err(|e| Error::Internal(e.to_string()))?;
+        let schema = match schema {
+            Some(schema) => schema,
+            None => return Err(Error::Internal(String::from("missing schema"))),
+        };
+
+        let ranked_map = self.index.main.ranked_map(reader);
+        let ranked_map = ranked_map.map_err(|e| Error::Internal(e.to_string()))?;
+        let ranked_map = ranked_map.unwrap_or_default();
+
+        // Change criteria
+        let mut query_builder = match self.get_criteria(reader, &ranked_map, &schema)? {
+            Some(criteria) => self.index.query_builder_with_criteria(criteria),
+            None => self.index.query_builder(),
+        };
+
+        if let Some(filter_expression) = &self.filters {
+            let filter = Filter::parse(filter_expression, &schema)?;
+            query_builder.with_filter(move |id| {
+                let index = &self.index;
+                let reader = &reader;
+                let filter = &filter;
+                match filter.test(reader, index, id) {
+                    Ok(res) => res,
+                    Err(e) => {
+                        log::warn!("unexpected error during filtering: {}", e);
+                        false
+                    }
+                }
+            });
+        }
+
+        query_builder.with_fetch_timeout(self.timeout);
+
+        if let Some(field) = self.index.main.distinct_attribute(reader)? {
+            if let Some(field_id) = schema.id(&field) {
+                query_builder.with_distinct(1, move |id| {
+                    match self.index.document_attribute_bytes(reader, id, field_id) {
+                        Ok(Some(bytes)) => {
+                            let mut s = SipHasher::new();
+                            bytes.hash(&mut s);
+                            Some(s.finish())
+                        }
+                        _ => None,
+                    }
+                });
+            }
+        }
+
+        let start = Instant::now();
+        let result =
+            query_builder.query(reader, &self.query, self.offset..(self.offset + self.limit));
+        let (docs, nb_hits) = result.map_err(|e| Error::SearchDocuments(e.to_string()))?;
+        let time_ms = start.elapsed().as_millis() as usize;
+
+        let mut all_attributes: HashSet<&str> = HashSet::new();
+        let mut all_formatted: HashSet<&str> = HashSet::new();
+
+        match &self.attributes_to_retrieve {
+            Some(to_retrieve) => {
+                all_attributes.extend(to_retrieve.iter().map(String::as_str));
+
+                if let Some(to_highlight) = &self.attributes_to_highlight {
+                    all_formatted.extend(to_highlight.iter().map(String::as_str));
+                }
+
+                if let Some(to_crop) = &self.attributes_to_crop {
+                    all_formatted.extend(to_crop.keys().map(String::as_str));
+                }
+
+                all_attributes.extend(&all_formatted);
+            },
+            None => {
+                all_attributes.extend(schema.displayed_name());
+                // If we specified at least one attribute to highlight or crop then
+                // all available attributes will be returned in the _formatted field.
+                if self.attributes_to_highlight.is_some() || self.attributes_to_crop.is_some() {
+                    all_formatted.extend(all_attributes.iter().cloned());
+                }
+            },
+        }
+
+        let mut hits = Vec::with_capacity(self.limit);
+        for doc in docs {
+            let mut document: IndexMap<String, Value> = self
+                .index
+                .document(reader, Some(&all_attributes), doc.id)
+                .map_err(|e| Error::RetrieveDocument(doc.id.0, e.to_string()))?
+                .ok_or(Error::DocumentNotFound(doc.id.0))?;
+
+            let mut formatted = document.iter()
+                .filter(|(key, _)| all_formatted.contains(key.as_str()))
+                .map(|(k, v)| (k.clone(), v.clone()))
+                .collect();
+
+            let mut matches = doc.highlights.clone();
+
+            // Crops fields if needed
+            if let Some(fields) = &self.attributes_to_crop {
+                crop_document(&mut formatted, &mut matches, &schema, fields);
+            }
+
+            // Transform to readable matches
+            if let Some(attributes_to_highlight) = &self.attributes_to_highlight {
+                let matches = calculate_matches(
+                    matches.clone(),
+                    self.attributes_to_highlight.clone(),
+                    &schema,
+                );
+                formatted = calculate_highlights(&formatted, &matches, attributes_to_highlight);
+            }
+
+            let matches_info = if self.matches {
+                Some(calculate_matches(matches, self.attributes_to_retrieve.clone(), &schema))
+            } else {
+                None
+            };
+
+            if let Some(attributes_to_retrieve) = &self.attributes_to_retrieve {
+                document.retain(|key, _| attributes_to_retrieve.contains(&key.to_string()))
+            }
+
+            let hit = SearchHit {
+                document,
+                formatted,
+                matches_info,
+            };
+
+            hits.push(hit);
+        }
+
+        let results = SearchResult {
+            hits,
+            offset: self.offset,
+            limit: self.limit,
+            nb_hits,
+            exhaustive_nb_hits: false,
+            processing_time_ms: time_ms,
+            query: self.query.to_string(),
+        };
+
+        Ok(results)
+    }
+
+    pub fn get_criteria(
+        &self,
+        reader: &heed::RoTxn<MainT>,
+        ranked_map: &'a RankedMap,
+        schema: &Schema,
+    ) -> Result<Option<Criteria<'a>>, Error> {
+        let ranking_rules = self.index.main.ranking_rules(reader)?;
+
+        if let Some(ranking_rules) = ranking_rules {
+            let mut builder = CriteriaBuilder::with_capacity(7 + ranking_rules.len());
+            for rule in ranking_rules {
+                match rule {
+                    RankingRule::Typo => builder.push(Typo),
+                    RankingRule::Words => builder.push(Words),
+                    RankingRule::Proximity => builder.push(Proximity),
+                    RankingRule::Attribute => builder.push(Attribute),
+                    RankingRule::WordsPosition => builder.push(WordsPosition),
+                    RankingRule::Exactness => builder.push(Exactness),
+                    RankingRule::Asc(field) => {
+                        match SortByAttr::lower_is_better(&ranked_map, &schema, &field) {
+                            Ok(rule) => builder.push(rule),
+                            Err(err) => error!("Error during criteria builder; {:?}", err),
+                        }
+                    }
+                    RankingRule::Desc(field) => {
+                        match SortByAttr::higher_is_better(&ranked_map, &schema, &field) {
+                            Ok(rule) => builder.push(rule),
+                            Err(err) => error!("Error during criteria builder; {:?}", err),
+                        }
+                    }
+                }
+            }
+            builder.push(DocumentId);
+            return Ok(Some(builder.build()));
+        }
+
+        Ok(None)
+    }
+}
+
+#[derive(Debug, Clone, Eq, PartialEq, PartialOrd, Serialize, Deserialize)]
+pub struct MatchPosition {
+    pub start: usize,
+    pub length: usize,
+}
+
+impl Ord for MatchPosition {
+    fn cmp(&self, other: &Self) -> Ordering {
+        match self.start.cmp(&other.start) {
+            Ordering::Equal => self.length.cmp(&other.length),
+            _ => self.start.cmp(&other.start),
+        }
+    }
+}
+
+pub type HighlightInfos = HashMap<String, Value>;
+pub type MatchesInfos = HashMap<String, Vec<MatchPosition>>;
+// pub type RankingInfos = HashMap<String, u64>;
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct SearchHit {
+    #[serde(flatten)]
+    pub document: IndexMap<String, Value>,
+    #[serde(rename = "_formatted", skip_serializing_if = "IndexMap::is_empty")]
+    pub formatted: IndexMap<String, Value>,
+    #[serde(rename = "_matchesInfo", skip_serializing_if = "Option::is_none")]
+    pub matches_info: Option<MatchesInfos>,
+}
+
+#[derive(Debug, Clone, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct SearchResult {
+    pub hits: Vec<SearchHit>,
+    pub offset: usize,
+    pub limit: usize,
+    pub nb_hits: usize,
+    pub exhaustive_nb_hits: bool,
+    pub processing_time_ms: usize,
+    pub query: String,
+}
+
+/// returns the start index and the length on the crop.
+fn aligned_crop(text: &str, match_index: usize, context: usize) -> (usize, usize) {
+    let is_word_component = |c: &char| c.is_alphanumeric() && !is_cjk(*c);
+
+    let word_end_index = |mut index| {
+        if text.chars().nth(index - 1).map_or(false, |c| is_word_component(&c)) {
+            index += text.chars().skip(index).take_while(is_word_component).count();
+        }
+        index
+    };
+
+    if context == 0 {
+        // count need to be at least 1 for cjk queries to return something
+        return (match_index, 1 + text.chars().skip(match_index).take_while(is_word_component).count());
+    }
+    let start = match match_index.saturating_sub(context) {
+        n if n == 0 => n,
+        n => word_end_index(n)
+    };
+    let end = word_end_index(start + 2 * context);
+
+    (start, end - start)
+}
+
+fn crop_text(
+    text: &str,
+    matches: impl IntoIterator<Item = Highlight>,
+    context: usize,
+) -> (String, Vec<Highlight>) {
+    let mut matches = matches.into_iter().peekable();
+
+    let char_index = matches.peek().map(|m| m.char_index as usize).unwrap_or(0);
+    let (start, count) = aligned_crop(text, char_index, context);
+
+    //TODO do something about the double allocation
+    let text = text.chars().skip(start).take(count).collect::<String>().trim().to_string();
+
+    // update matches index to match the new cropped text
+    let matches = matches
+        .take_while(|m| (m.char_index as usize) + (m.char_length as usize) <= start + (context * 2))
+        .map(|match_| Highlight {
+            char_index: match_.char_index - start as u16,
+            ..match_
+        })
+        .collect();
+
+    (text, matches)
+}
+
+fn crop_document(
+    document: &mut IndexMap<String, Value>,
+    matches: &mut Vec<Highlight>,
+    schema: &Schema,
+    fields: &HashMap<String, usize>,
+) {
+    matches.sort_unstable_by_key(|m| (m.char_index, m.char_length));
+
+    for (field, length) in fields {
+        let attribute = match schema.id(field) {
+            Some(attribute) => attribute,
+            None => continue,
+        };
+
+        let selected_matches = matches
+            .iter()
+            .filter(|m| FieldId::new(m.attribute) == attribute)
+            .cloned();
+
+        if let Some(Value::String(ref mut original_text)) = document.get_mut(field) {
+            let (cropped_text, cropped_matches) =
+                crop_text(original_text, selected_matches, *length);
+
+            *original_text = cropped_text;
+
+            matches.retain(|m| FieldId::new(m.attribute) != attribute);
+            matches.extend_from_slice(&cropped_matches);
+        }
+    }
+}
+
+fn calculate_matches(
+    matches: Vec<Highlight>,
+    attributes_to_retrieve: Option<HashSet<String>>,
+    schema: &Schema,
+) -> MatchesInfos {
+    let mut matches_result: HashMap<String, Vec<MatchPosition>> = HashMap::new();
+    for m in matches.iter() {
+        if let Some(attribute) = schema.name(FieldId::new(m.attribute)) {
+            if let Some(attributes_to_retrieve) = attributes_to_retrieve.clone() {
+                if !attributes_to_retrieve.contains(attribute) {
+                    continue;
+                }
+            }
+            if !schema.displayed_name().contains(attribute) {
+                continue;
+            }
+            if let Some(pos) = matches_result.get_mut(attribute) {
+                pos.push(MatchPosition {
+                    start: m.char_index as usize,
+                    length: m.char_length as usize,
+                });
+            } else {
+                let mut positions = Vec::new();
+                positions.push(MatchPosition {
+                    start: m.char_index as usize,
+                    length: m.char_length as usize,
+                });
+                matches_result.insert(attribute.to_string(), positions);
+            }
+        }
+    }
+    for (_, val) in matches_result.iter_mut() {
+        val.sort_unstable();
+        val.dedup();
+    }
+    matches_result
+}
+
+fn calculate_highlights(
+    document: &IndexMap<String, Value>,
+    matches: &MatchesInfos,
+    attributes_to_highlight: &HashSet<String>,
+) -> IndexMap<String, Value> {
+    let mut highlight_result = document.clone();
+
+    for (attribute, matches) in matches.iter() {
+        if attributes_to_highlight.contains(attribute) {
+            if let Some(Value::String(value)) = document.get(attribute) {
+                let value: Vec<_> = value.chars().collect();
+                let mut highlighted_value = String::new();
+                let mut index = 0;
+
+                let longest_matches = matches
+                    .linear_group_by_key(|m| m.start)
+                    .map(|group| group.last().unwrap());
+
+                for m in longest_matches {
+                    if m.start >= index {
+                        let before = value.get(index..m.start);
+                        let highlighted = value.get(m.start..(m.start + m.length));
+                        if let (Some(before), Some(highlighted)) = (before, highlighted) {
+                            highlighted_value.extend(before);
+                            highlighted_value.push_str("<em>");
+                            highlighted_value.extend(highlighted);
+                            highlighted_value.push_str("</em>");
+                            index = m.start + m.length;
+                        } else {
+                            error!("value: {:?}; index: {:?}, match: {:?}", value, index, m);
+                        }
+                    }
+                }
+                highlighted_value.extend(value[index..].iter());
+                highlight_result.insert(attribute.to_string(), Value::String(highlighted_value));
+            };
+        }
+    }
+
+    highlight_result
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn aligned_crops() {
+        let text = r#"En ce début de trentième millénaire, l'Empire n'a jamais été aussi puissant, aussi étendu à travers toute la galaxie. C'est dans sa capitale, Trantor, que l'éminent savant Hari Seldon invente la psychohistoire, une science toute nouvelle, à base de psychologie et de mathématiques, qui lui permet de prédire l'avenir... C'est-à-dire l'effondrement de l'Empire d'ici cinq siècles et au-delà, trente mille années de chaos et de ténèbres. Pour empêcher cette catastrophe et sauver la civilisation, Seldon crée la Fondation."#;
+
+        // simple test
+        let (start, length) = aligned_crop(&text, 6, 2);
+        let cropped =  text.chars().skip(start).take(length).collect::<String>().trim().to_string();
+        assert_eq!("début", cropped);
+
+        // first word test
+        let (start, length) = aligned_crop(&text, 0, 1);
+        let cropped =  text.chars().skip(start).take(length).collect::<String>().trim().to_string();
+        assert_eq!("En", cropped);
+        // last word test
+        let (start, length) = aligned_crop(&text, 510, 2);
+        let cropped =  text.chars().skip(start).take(length).collect::<String>().trim().to_string();
+        assert_eq!("Fondation", cropped);
+
+        // CJK tests
+        let text = "this isのス foo myタイリ test";
+
+        // mixed charset
+        let (start, length) = aligned_crop(&text, 5, 3);
+        let cropped =  text.chars().skip(start).take(length).collect::<String>().trim().to_string();
+        assert_eq!("isのス", cropped);
+
+        // split regular word / CJK word, no space
+        let (start, length) = aligned_crop(&text, 7, 1);
+        let cropped =  text.chars().skip(start).take(length).collect::<String>().trim().to_string();
+        assert_eq!("のス", cropped);
+
+    }
+
+    #[test]
+    fn calculate_matches() {
+        let mut matches = Vec::new();
+        matches.push(Highlight { attribute: 0, char_index: 0, char_length: 3});
+        matches.push(Highlight { attribute: 0, char_index: 0, char_length: 2});
+
+        let mut attributes_to_retrieve: HashSet<String> = HashSet::new();
+        attributes_to_retrieve.insert("title".to_string());
+
+        let schema = Schema::with_primary_key("title");
+
+        let matches_result = super::calculate_matches(matches, Some(attributes_to_retrieve), &schema);
+
+        let mut matches_result_expected: HashMap<String, Vec<MatchPosition>> = HashMap::new();
+
+        let mut positions = Vec::new();
+        positions.push(MatchPosition {
+            start: 0,
+            length: 2,
+        });
+        positions.push(MatchPosition {
+            start: 0,
+            length: 3,
+        });
+        matches_result_expected.insert("title".to_string(), positions);
+
+        assert_eq!(matches_result, matches_result_expected);
+    }
+
+    #[test]
+    fn calculate_highlights() {
+        let data = r#"{
+            "title": "Fondation (Isaac ASIMOV)",
+            "description": "En ce début de trentième millénaire, l'Empire n'a jamais été aussi puissant, aussi étendu à travers toute la galaxie. C'est dans sa capitale, Trantor, que l'éminent savant Hari Seldon invente la psychohistoire, une science toute nouvelle, à base de psychologie et de mathématiques, qui lui permet de prédire l'avenir... C'est-à-dire l'effondrement de l'Empire d'ici cinq siècles et au-delà, trente mille années de chaos et de ténèbres. Pour empêcher cette catastrophe et sauver la civilisation, Seldon crée la Fondation."
+        }"#;
+
+        let document: IndexMap<String, Value> = serde_json::from_str(data).unwrap();
+        let mut attributes_to_highlight = HashSet::new();
+        attributes_to_highlight.insert("title".to_string());
+        attributes_to_highlight.insert("description".to_string());
+
+        let mut matches = HashMap::new();
+
+        let mut m = Vec::new();
+        m.push(MatchPosition {
+            start: 0,
+            length: 9,
+        });
+        matches.insert("title".to_string(), m);
+
+        let mut m = Vec::new();
+        m.push(MatchPosition {
+            start: 510,
+            length: 9,
+        });
+        matches.insert("description".to_string(), m);
+        let result = super::calculate_highlights(&document, &matches, &attributes_to_highlight);
+
+        let mut result_expected = IndexMap::new();
+        result_expected.insert(
+            "title".to_string(),
+            Value::String("<em>Fondation</em> (Isaac ASIMOV)".to_string()),
+        );
+        result_expected.insert("description".to_string(), Value::String("En ce début de trentième millénaire, l'Empire n'a jamais été aussi puissant, aussi étendu à travers toute la galaxie. C'est dans sa capitale, Trantor, que l'éminent savant Hari Seldon invente la psychohistoire, une science toute nouvelle, à base de psychologie et de mathématiques, qui lui permet de prédire l'avenir... C'est-à-dire l'effondrement de l'Empire d'ici cinq siècles et au-delà, trente mille années de chaos et de ténèbres. Pour empêcher cette catastrophe et sauver la civilisation, Seldon crée la <em>Fondation</em>.".to_string()));
+
+        assert_eq!(result, result_expected);
+    }
+
+    #[test]
+    fn highlight_longest_match() {
+        let data = r#"{
+            "title": "Ice"
+        }"#;
+
+        let document: IndexMap<String, Value> = serde_json::from_str(data).unwrap();
+        let mut attributes_to_highlight = HashSet::new();
+        attributes_to_highlight.insert("title".to_string());
+
+        let mut matches = HashMap::new();
+
+        let mut m = Vec::new();
+        m.push(MatchPosition {
+            start: 0,
+            length: 2,
+        });
+        m.push(MatchPosition {
+            start: 0,
+            length: 3,
+        });
+        matches.insert("title".to_string(), m);
+
+        let result = super::calculate_highlights(&document, &matches, &attributes_to_highlight);
+
+        let mut result_expected = IndexMap::new();
+        result_expected.insert(
+            "title".to_string(),
+            Value::String("<em>Ice</em>".to_string()),
+        );
+
+        assert_eq!(result, result_expected);
+    }
+}

meilisearch-http/src/helpers/mod.rs

@@ -0,0 +1,2 @@
+pub mod meilisearch;
+pub mod tide;

meilisearch-http/src/helpers/tide.rs

@@ -0,0 +1,83 @@
+use crate::error::{ResponseError, SResult};
+use crate::Data;
+use meilisearch_core::Index;
+use tide::Request;
+
+pub enum ACL {
+    Admin,
+    Private,
+    Public,
+}
+
+pub trait RequestExt {
+    fn is_allowed(&self, acl: ACL) -> SResult<()>;
+    fn url_param(&self, name: &str) -> SResult<String>;
+    fn index(&self) -> SResult<Index>;
+    fn document_id(&self) -> SResult<String>;
+}
+
+impl RequestExt for Request<Data> {
+    fn is_allowed(&self, acl: ACL) -> SResult<()> {
+        let user_api_key = self.header("X-Meili-API-Key");
+
+        if self.state().api_keys.master.is_none() {
+            return Ok(())
+        }
+
+        match acl {
+            ACL::Admin => {
+                if user_api_key == self.state().api_keys.master.as_deref() {
+                    return Ok(());
+                }
+            }
+            ACL::Private => {
+                if user_api_key == self.state().api_keys.master.as_deref() {
+                    return Ok(());
+                }
+                if user_api_key == self.state().api_keys.private.as_deref() {
+                    return Ok(());
+                }
+            }
+            ACL::Public => {
+                if user_api_key == self.state().api_keys.master.as_deref() {
+                    return Ok(());
+                }
+                if user_api_key == self.state().api_keys.private.as_deref() {
+                    return Ok(());
+                }
+                if user_api_key == self.state().api_keys.public.as_deref() {
+                    return Ok(());
+                }
+            }
+        }
+
+        Err(ResponseError::InvalidToken(
+            user_api_key.unwrap_or("Need a token").to_owned(),
+        ))
+    }
+
+    fn url_param(&self, name: &str) -> SResult<String> {
+        let param = self
+            .param::<String>(name)
+            .map_err(|e| ResponseError::bad_parameter(name, e))?;
+        Ok(param)
+    }
+
+    fn index(&self) -> SResult<Index> {
+        let index_uid = self.url_param("index")?;
+        let index = self
+            .state()
+            .db
+            .open_index(&index_uid)
+            .ok_or(ResponseError::index_not_found(index_uid))?;
+        Ok(index)
+    }
+
+    fn document_id(&self) -> SResult<String> {
+        let name = self
+            .param::<String>("document_id")
+            .map_err(|_| ResponseError::bad_parameter("documentId", "primaryKey"))?;
+
+        Ok(name)
+    }
+}

meilisearch-http/src/lib.rs

@@ -0,0 +1,10 @@
+#![allow(clippy::or_fun_call)]
+
+pub mod data;
+pub mod error;
+pub mod helpers;
+pub mod models;
+pub mod option;
+pub mod routes;
+
+pub use self::data::Data;

meilisearch-http/src/main.rs

@@ -0,0 +1,112 @@
+use std::{env, thread};
+
+use async_std::task;
+use log::info;
+use main_error::MainError;
+use structopt::StructOpt;
+use tide::middleware::{Cors, RequestLogger, Origin};
+use http::header::HeaderValue;
+
+use meilisearch_http::data::Data;
+use meilisearch_http::option::Opt;
+use meilisearch_http::routes;
+use meilisearch_http::routes::index::index_update_callback;
+
+mod analytics;
+
+#[cfg(target_os = "linux")]
+#[global_allocator]
+static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc;
+
+pub fn main() -> Result<(), MainError> {
+    let opt = Opt::from_args();
+
+    match opt.env.as_ref() {
+        "production" => {
+            if opt.master_key.is_none() {
+                return Err(
+                    "In production mode, the environment variable MEILI_MASTER_KEY is mandatory"
+                        .into(),
+                );
+            }
+            env_logger::init();
+        }
+        "development" => {
+            env_logger::from_env(env_logger::Env::default().default_filter_or("info")).init();
+        }
+        _ => unreachable!(),
+    }
+
+    if !opt.no_analytics {
+        thread::spawn(analytics::analytics_sender);
+    }
+
+    let data = Data::new(opt.clone());
+
+    let data_cloned = data.clone();
+    data.db.set_update_callback(Box::new(move |name, status| {
+        index_update_callback(name, &data_cloned, status);
+    }));
+
+    print_launch_resume(&opt, &data);
+
+    let mut app = tide::with_state(data);
+
+    app.middleware(Cors::new()
+        .allow_methods(HeaderValue::from_static("GET, POST, PUT, DELETE, OPTIONS"))
+        .allow_headers(HeaderValue::from_static("X-Meili-API-Key"))
+        .allow_origin(Origin::from("*")));
+    app.middleware(RequestLogger::new());
+
+    routes::load_routes(&mut app);
+
+    task::block_on(app.listen(opt.http_addr))?;
+    Ok(())
+}
+
+pub fn print_launch_resume(opt: &Opt, data: &Data) {
+    let ascii_name = r#"
+888b     d888          d8b 888 d8b  .d8888b.                                    888
+8888b   d8888          Y8P 888 Y8P d88P  Y88b                                   888
+88888b.d88888              888     Y88b.                                        888
+888Y88888P888  .d88b.  888 888 888  "Y888b.    .d88b.   8888b.  888d888 .d8888b 88888b.
+888 Y888P 888 d8P  Y8b 888 888 888     "Y88b. d8P  Y8b     "88b 888P"  d88P"    888 "88b
+888  Y8P  888 88888888 888 888 888       "888 88888888 .d888888 888    888      888  888
+888   "   888 Y8b.     888 888 888 Y88b  d88P Y8b.     888  888 888    Y88b.    888  888
+888       888  "Y8888  888 888 888  "Y8888P"   "Y8888  "Y888888 888     "Y8888P 888  888
+"#;
+
+    println!("{}", ascii_name);
+
+    info!("Database path: {:?}", opt.db_path);
+    info!("Start server on: {:?}", opt.http_addr);
+    info!("Environment: {:?}", opt.env);
+    info!("Commit SHA: {:?}", env!("VERGEN_SHA").to_string());
+    info!(
+        "Build date: {:?}",
+        env!("VERGEN_BUILD_TIMESTAMP").to_string()
+    );
+    info!(
+        "Package version: {:?}",
+        env!("CARGO_PKG_VERSION").to_string()
+    );
+
+    if let Some(master_key) = &data.api_keys.master {
+        info!("Master Key: {:?}", master_key);
+
+        if let Some(private_key) = &data.api_keys.private {
+            info!("Private Key: {:?}", private_key);
+        }
+
+        if let Some(public_key) = &data.api_keys.public {
+            info!("Public Key: {:?}", public_key);
+        }
+    } else {
+        info!("No master key found; The server will have no securities.\
+            If you need some protection in development mode, please export a key. export MEILI_MASTER_KEY=xxx");
+    }
+
+    info!("If you need extra information; Please refer to the documentation: http://docs.meilisearch.com");
+    info!("If you want to support us or help us; Please consult our Github repo: http://github.com/meilisearch/meilisearch");
+    info!("If you want to contact us; Please chat with us on http://meilisearch.com or by email to bonjour@meilisearch.com");
+}

meilisearch-http/src/models/mod.rs

@@ -0,0 +1 @@
+pub mod update_operation;

meilisearch-http/src/models/update_operation.rs

@@ -0,0 +1,33 @@
+use std::fmt;
+
+#[allow(dead_code)]
+#[derive(Debug)]
+pub enum UpdateOperation {
+    ClearAllDocuments,
+    DocumentsAddition,
+    DocumentsDeletion,
+    SynonymsUpdate,
+    SynonymsDeletion,
+    StopWordsAddition,
+    StopWordsDeletion,
+    Schema,
+    Config,
+}
+
+impl fmt::Display for UpdateOperation {
+    fn fmt(&self, f: &mut fmt::Formatter) -> std::fmt::Result {
+        use UpdateOperation::*;
+
+        match self {
+            ClearAllDocuments => write!(f, "ClearAllDocuments"),
+            DocumentsAddition => write!(f, "DocumentsAddition"),
+            DocumentsDeletion => write!(f, "DocumentsDeletion"),
+            SynonymsUpdate => write!(f, "SynonymsUpdate"),
+            SynonymsDeletion => write!(f, "SynonymsDelettion"),
+            StopWordsAddition => write!(f, "StopWordsAddition"),
+            StopWordsDeletion => write!(f, "StopWordsDeletion"),
+            Schema => write!(f, "Schema"),
+            Config => write!(f, "Config"),
+        }
+    }
+}