Merge #960

960: bump version and update changelog r=MarinPostma a=LegendreM

* bump to 0.14.1
* update CHANGELOG.md file

Co-authored-by: many <maxime@meilisearch.com>

.dockerignore

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

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,38 @@
+---
+name: Bug report
+about: Create a report to help us improve
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+**Describe the bug**
+A clear and concise description of what the bug is.
+
+**To Reproduce**
+Steps to reproduce the behavior:
+1. Go to '...'
+2. Click on '....'
+3. Scroll down to '....'
+4. See error
+
+**Expected behavior**
+A clear and concise description of what you expected to happen.
+
+**Screenshots**
+If applicable, add screenshots to help explain your problem.
+
+**Desktop (please complete the following information):**
+ - OS: [e.g. iOS]
+ - Browser [e.g. chrome, safari]
+ - Version [e.g. 22]
+
+**Smartphone (please complete the following information):**
+ - Device: [e.g. iPhone6]
+ - OS: [e.g. iOS8.1]
+ - Browser [e.g. stock browser, safari]
+ - Version [e.g. 22]
+
+**Additional context**
+Add any other context about the problem here.

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,20 @@
+---
+name: Feature request
+about: Suggest an idea for this project
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+**Is your feature request related to a problem? Please describe.**
+A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]
+
+**Describe the solution you'd like**
+A clear and concise description of what you want to happen.
+
+**Describe alternatives you've considered**
+A clear and concise description of any alternative solutions or features you've considered.
+
+**Additional context**
+Add any other context or screenshots about the feature request here.

.github/ISSUE_TEMPLATE/tracking-issue.md

@@ -0,0 +1,40 @@
+---
+name: Tracking issue
+about: Template for a tracking issue
+title: ''
+labels: tracking-issue
+assignees: ''
+
+---
+
+# Summary
+
+One paragraph to explain the feature.
+
+# Motivations
+
+Why are we doing this? What use cases does it support? What is the expected outcome?
+
+# Explanation
+
+Explain the proposal like it was the final documentation of this proposal.
+
+- What is changing for end-users.
+- How it works.
+- What is breaking?
+- Examples.
+
+# Implementation
+
+Explain the technical specificities that will need to be known or done in order to implement this proposal.
+
+## Steps
+
+Describe each step to create the feature with it's associated issue/PR.
+
+# Related
+
+- [ ] Validated by the team (@people needed)
+- [ ] Test added
+- [ ] [Documentation](https://github.com/meilisearch/documentation/issues/#xxx) //Change xxx or remove the line
+- [ ] [SDK/Integrations](https://github.com/meilisearch/integration-guides/issues/#xxx) //Change xxx or remove the line

.github/is-latest-release.sh

@@ -0,0 +1,132 @@
+#!/bin/sh
+
+# Checks if the current tag should be the latest (in terms of semver and not of release date).
+# Ex: previous tag -> v0.10.1
+#     new tag -> v0.8.12
+#     The new tag should not be the latest
+#     So it returns "false", the CI should not run for the release v0.8.2
+
+# Used in GHA in publish-docker-latest.yml
+# Returns "true" or "false" (as a string) to be used in the `if` in GHA
+
+# GLOBAL
+GREP_SEMVER_REGEXP='v\([0-9]*\)[.]\([0-9]*\)[.]\([0-9]*\)$' # i.e. v[number].[number].[number]
+
+# FUNCTIONS
+
+# 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" < "_$SPECIAL_B" ]; then
+        return 0
+    fi
+
+    return 1
+}
+
+# Returns the tag of the latest stable release (in terms of semver and not of release date)
+get_latest() {
+    temp_file='temp_file' # temp_file needed because the grep would start before the download is over
+    curl -s 'https://api.github.com/repos/meilisearch/MeiliSearch/releases' > "$temp_file"
+    releases=$(cat "$temp_file" | \
+        grep -E "tag_name|draft|prerelease" \
+        | tr -d ',"' | cut -d ':' -f2 | tr -d ' ')
+        # Returns a list of [tag_name draft_boolean prerelease_boolean ...]
+        # Ex: v0.10.1 false false v0.9.1-rc.1 false true v0.9.0 false false...
+
+    i=0
+    latest=""
+    current_tag=""
+    for release_info in $releases; do
+        if [ $i -eq 0 ]; then # Cheking tag_name
+            if echo "$release_info" | grep -q "$GREP_SEMVER_REGEXP"; then # If it's not an alpha or beta release
+                current_tag=$release_info
+            else
+                current_tag=""
+            fi
+            i=1
+        elif [ $i -eq 1 ]; then # Checking draft boolean
+            if [ "$release_info" = "true" ]; then
+                current_tag=""
+            fi
+            i=2
+        elif [ $i -eq 2 ]; then # Checking prerelease boolean
+            if [ "$release_info" = "true" ]; then
+                current_tag=""
+            fi
+            i=0
+            if [ "$current_tag" != "" ]; then # If the current_tag is valid
+                if [ "$latest" = "" ]; then # If there is no latest yet
+                    latest="$current_tag"
+                else
+                    semverLT $current_tag $latest # Comparing latest and the current tag
+                    if [ $? -eq 1 ]; then
+                        latest="$current_tag"
+                    fi
+                fi
+            fi
+        fi
+    done
+
+    rm -f "$temp_file"
+    echo $latest
+}
+
+# MAIN
+current_tag="$(echo $GITHUB_REF | tr -d 'refs/tags/')"
+latest="$(get_latest)"
+
+if [ "$current_tag" != "$latest" ]; then
+    # The current release tag is not the latest
+    echo "false"
+else
+    # The current release tag is the latest
+    echo "true"
+fi

.github/workflows/README.md

@@ -0,0 +1,20 @@
+# GitHub Actions Workflow for MeiliSearch
+
+> **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 trigger `cargo test`.
+- On each tag, we build:
+    - the tagged Docker image and publish it to Docker Hub
+    - the binaries for MacOS, Ubuntu, and Windows
+    - the Debian package
+- On each stable release (`v*.*.*` tag):
+    - we build the `latest` Docker image and publish it to Docker Hub
+    - we publish the binary to Hombrew and Gemfury
+
+## 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 +1,86 @@
+on:
+  release:
+    types: [published]
+
+name: Publish binaries to release
+
+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.exe
+            asset_name: meilisearch-windows-amd64.exe
+
+    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:
+  release:
+    types: [released]
+
+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:
+      - name: Create PR to Homebrew
+        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 +1,22 @@
+---
+on:
+  release:
+    types: [released]
+
+name: Publish latest image to Docker Hub
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - name: Check if current release is latest
+        run: echo "##[set-output name=is_latest;]$(sh .github/is-latest-release.sh)"
+        id: release
+      - name: Publish to Registry
+        if: steps.release.outputs.is_latest == 'true'
+        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 +1,93 @@
+---
+on:
+  push:
+    branches:
+      - release-v*
+      - trying
+      - staging
+    tags:
+      - 'v[0-9]+.[0-9]+.[0-9]+' # this only concerns tags on stable
+
+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
+        components: clippy
+    - name: Run cargo test
+      uses: actions-rs/cargo@v1
+      with:
+        command: test
+        args: --locked --release
+    - name: Run cargo clippy
+      uses: actions-rs/cargo@v1
+      with:
+        command: clippy
+
+  build-image:
+    name: Test the build of Docker image
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v1
+    - run: docker build . --file Dockerfile -t meilisearch
+      name: Docker build
+
+  ## A push occurred on a release branch, a prerelease is created and assets are generated
+  prerelease:
+    name: create prerelease
+    needs: [check, build-image]
+    if: ${{ contains(github.ref, 'release-') && github.event_name == 'push' }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 0
+      - name: Get version number
+        id: version-number
+        run: echo "##[set-output name=number;]$(echo ${{ github.ref }} | sed 's/.*\(v.*\)/\1/')"
+      - name: Get commit count
+        id: commit-count
+        run: echo "##[set-output name=count;]$(git rev-list remotes/origin/master..remotes/origin/release-${{ steps.version-number.outputs.number }} --count)"
+      - name: Create Release
+        id: create_release
+        uses: actions/create-release@v1
+        env:
+          GITHUB_TOKEN: ${{ secrets.PUBLISH_TOKEN }} # Personal Access Token
+        with:
+          tag_name: ${{ steps.version-number.outputs.number }}rc${{ steps.commit-count.outputs.count }}
+          release_name: Pre-release ${{ steps.version-number.outputs.number }}-rc${{ steps.commit-count.outputs.count }}
+          prerelease: true
+
+  ## If a tag is pushed, a release is created for this tag, and assets will be generated
+  release:
+    name: create release
+    needs: [check, build-image]
+    if:  ${{ contains(github.ref, 'tags/v') }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v2
+      - name: Get version number
+        id: version-number
+        run: echo "##[set-output name=number;]$(echo ${{ github.ref }} | sed 's/.*\(v.*\)/\1/')" 
+      - name: Create Release
+        id: create_release
+        uses: actions/create-release@v1
+        env:
+          GITHUB_TOKEN: ${{ secrets.PUBLISH_TOKEN }} # PAT
+        with:
+          tag_name: ${{ steps.version-number.outputs.number }}
+          release_name: Meilisearch ${{ steps.version-number.outputs.number }}
+          prerelease: false

.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: 1.32.0
-    name: "meilidb on 1.32.0"
-    script: ./ci/meilidb.sh
-
-  # Test crates on nightly Rust.
-  - rust: nightly
-    name: "meilidb on nightly"
-    script: ./ci/meilidb.sh
-

CHANGELOG.md

@@ -0,0 +1,78 @@
+## v0.14.1
+
+  - Fix version mismatch in snapshot importation (#959)
+
+## v0.14.0
+
+  - Fix facet distribution case (#797)
+  - Snapshotting (#839)
+  - Fix bucket-sort unwrap bug (#915)
+
+## v0.13.0
+
+  - placeholder search (#771)
+  - Add database version mismatch check (#794)
+  - Displayed and searchable attributes wildcard (#846)
+  - Remove sys-info route (#810)
+  - Check database version mismatch (#794)
+  - Fix unique docid bug (#841)
+  - Error codes in updates (#792)
+  - Sentry disable argument (#813)
+  - Log analytics if enabled (#825)
+  - Fix default values displayed on web interface (#874)
+
+## v0.12.0
+
+  - Fix long documents not being indexed completely bug (#816)
+  - Fix distinct attribute returning id instead of name (#800)
+  - error code rename (#805)
+
+## v0.11.1
+
+  - Fix facet cache on document update (#789) 
+  - Improvements on settings consistency (#778)
+
+## v0.11.0
+
+  - Change the HTTP framework, moving from tide to actix-web (#601)
+  - Bump sentry version to 0.18.1 (#690)
+  - Enable max payload size override (#684)
+  - Disable sentry in debug (#681)
+  - Better terminal greeting (#680)
+  - Fix highlight misalignment (#679)
+  - Add support for facet count (#676)
+  - Add support for faceted search (#631)
+  - Add support for configuring the lmdb map size (#646, #647)
+  - Add exposed port for Dockerfile (#654)
+  - Add sentry probe (#664)
+  - Fix url trailing slash and double slash issues (#659)
+  - Fix accept all Content-Type by default (#653)
+  - Return the error message from Serde when a deserialization error is encountered (#661)
+  - Fix NormalizePath middleware to make the dashboard accessible (#695)
+  - Update sentry features to remove openssl (#702)
+  - Add SSL support (#669)
+  - Rename fieldsFrequency into fieldsDistribution in stats (#719)
+  - Add support for error code reporting (#703)
+  - Allow the dashboard to query private servers (#732)
+  - Add telemetry (#720)
+  - Add post route for search (#735)
+
+## 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)
+
+## v0.10.0
+
+  - 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)

CODE_OF_CONDUCT.md

@@ -0,0 +1,76 @@
+# Contributor Covenant Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as
+contributors and maintainers pledge to making participation in our project and
+our community a harassment-free experience for everyone, regardless of age, body
+size, disability, ethnicity, sex characteristics, gender identity and expression,
+level of experience, education, socio-economic status, nationality, personal
+appearance, race, religion, or sexual identity and orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment
+include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or
+ advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic
+ address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+ professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable
+behavior and are expected to take appropriate and fair corrective action in
+response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or
+reject comments, commits, code, wiki edits, issues, and other contributions
+that are not aligned to this Code of Conduct, or to ban temporarily or
+permanently any contributor for other behaviors that they deem inappropriate,
+threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies both within project spaces and in public spaces
+when an individual is representing the project or its community. Examples of
+representing a project or community include using an official project e-mail
+address, posting via an official social media account, or acting as an appointed
+representative at an online or offline event. Representation of a project may be
+further defined and clarified by project maintainers.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported by contacting the project team at bonjour@meilisearch.com. All
+complaints will be reviewed and investigated and will result in a response that
+is deemed necessary and appropriate to the circumstances. The project team is
+obligated to maintain confidentiality with regard to the reporter of an incident.
+Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good
+faith may face temporary or permanent repercussions as determined by other
+members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
+available at https://www.contributor-covenant.org/version/1/4/code-of-conduct.html
+
+[homepage]: https://www.contributor-covenant.org
+
+For answers to common questions about this code of conduct, see
+https://www.contributor-covenant.org/faq

Cargo.toml

@@ -1,55 +1,11 @@
-[package]
-edition = "2018"
-name = "meilidb"
-version = "0.3.1"
-authors = ["Kerollmops <renault.cle@gmail.com>"]
-
-[dependencies]
-arc-swap = "0.3"
-bincode = "1.0"
-byteorder = "1.2"
-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"] }
-lockfree = "0.5"
-log = "0.4"
-rayon = "1.0"
-sdset = "0.3"
-serde = "1.0"
-serde_derive = "1.0"
-serde_json = { version = "1.0", features = ["preserve_order"] }
-size_format = "1.0"
-slice-group-by = "0.2"
-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"
-
-[features]
-default = ["simd"]
-i128 = ["bincode/i128", "byteorder/i128"]
-portable = ["rocksdb/portable"]
-simd = ["rocksdb/sse"]
-nightly = ["hashbrown/nightly", "slice-group-by/nightly"]
-
-[dev-dependencies]
-csv = "1.0"
-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 +1,29 @@
+# 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 add -q --no-cache libgcc tini
+
+COPY    --from=compiler /meilisearch/target/release/meilisearch .
+
+ENV     MEILI_HTTP_ADDR 0.0.0.0:7700
+EXPOSE  7700/tcp
+
+ENTRYPOINT ["tini", "--"]
+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 +1,192 @@
-# MeiliDB
+<p align="center">
+  <img src="assets/logo.svg" alt="MeiliSearch" width="200" height="200" />
+</p>
 
-[![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)
+<h1 align="center">MeiliSearch</h1>
 
-A _full-text search database_ using a key-value store internally.
+<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/faq/">FAQ</a>
+</h4>
 
-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.
+<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>
+  <a href="https://app.bors.tech/repositories/26457"><img src="https://bors.tech/images/badge_small.svg" alt="Bors enabled"></a>
+</p>
 
-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 or you can take a look at the [typos and ranking rules](typos-ranking-rules.md) if you want to know the default rules used to sort the documents.
+<p align="center">⚡ Lightning Fast, Ultra Relevant, and Typo-Tolerant Search Engine 🔍</p>
 
-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!
+**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/).
 
-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.
+<p align="center">
+  <img src="assets/movies-web-demo.gif" alt="Web interface gif" />
+</p>
 
-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.
+## ✨ Features
+* Search as-you-type experience (answers < 50 milliseconds)
+* Full-text search
+* Typo tolerant (understands typos and miss-spelling)
+* Faceted search and filters
+* Supports Kanji characters
+* Supports Synonym
+* Easy to install, deploy, and maintain
+* Whole documents are returned
+* Highly customizable
+* RESTful API
 
+## Getting started
 
+### Deploy the Server
 
-## Performances
-
-With a database composed of _100 353_ documents with _352_ attributes each and _3_ of them indexed.
-So more than _300 000_ fields indexed for _35 million_ stored we can handle more than _2.8k req/sec_ with an average response time of _9 ms_ on an Intel i7-7700 (8) @ 4.2GHz.
-
-Requests are made using [wrk](https://github.com/wg/wrk) and scripted to simulate real users queries.
-
-```
-Running 10s test @ http://localhost:2230
-  2 threads and 25 connections
-  Thread Stats   Avg      Stdev     Max   +/- Stdev
-    Latency     9.52ms    7.61ms  99.25ms   84.58%
-    Req/Sec     1.41k   119.11     1.78k    64.50%
-  28080 requests in 10.01s, 7.42MB read
-Requests/sec:   2806.46
-Transfer/sec:    759.17KB
-```
-
-### Notes
-
-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).
-
-## Usage and examples
-
-MeiliDB runs with an index like most search engines.
-So to test the library you can create one by indexing a simple csv file.
+#### Brew (Mac OS)
 
 ```bash
-cargo run --release --example create-database -- test.mdb misc/kaggle.csv --schema schema-example.toml
+brew update && brew install meilisearch
+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.
+#### Docker
 
 ```bash
-cargo run --release --example query-database -- test.mdb -n 10 id title
+docker run -p 7700:7700 -v $(pwd)/data.ms:/data.ms getmeili/meilisearch
 ```
 
+#### Run on Digital Ocean
+
+[![DigitalOcean Marketplace](assets/do-btn-blue.svg)](https://marketplace.digitalocean.com/apps/meilisearch?action=deploy&refcode=7c67bd97e101)
+
+#### APT (Debian & Ubuntu)
+
+```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 (Linux & Mac OS)
+
+```bash
+curl -L https://install.meilisearch.com | sh
+./meilisearch
+```
+
+#### Compile and run it from sources
+
+If you have the latest stable Rust toolchain 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
+rustup override set stable
+rustup update stable
+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.
+
+| [See the gif above](#demo)
+
+## 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!
+
+## Telemetry
+
+MeiliSearch collects anonymous data regarding general usage.
+This helps us better understand developers usage of MeiliSearch features.<br/>
+To see what information we're retrieving, please see the complete list [on the dedicated issue](https://github.com/meilisearch/MeiliSearch/issues/720).<br/>
+We also use Sentry to make us crash and error reports. If you want to know more about what Sentry collects, please visit their [privacy policy website](https://sentry.io/privacy/).<br/>
+This program is optionnal, 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)
+* 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)
+* Join our [Slack community](https://slack.meilisearch.com/).
+* By opening an issue.
+
+MeiliSearch is developed by [Meili](https://www.meilisearch.com), a young company. To know more about us, you can [read our blog](https://blog.meilisearch.com). Any suggestion or feedback is highly appreciated. Thank you for your support!

assets/do-btn-blue.svg

@@ -0,0 +1,23 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<svg width="200px" height="42px" viewBox="0 0 200 42" version="1.1" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
+    <!-- Generator: Sketch 52.5 (67469) - http://www.bohemiancoding.com/sketch -->
+    <title>do-btn-blue</title>
+    <desc>Created with Sketch.</desc>
+    <g id="Page-1" stroke="none" stroke-width="1" fill="none" fill-rule="evenodd">
+        <g id="Partner-welcome-kit-Copy-3" transform="translate(-651.000000, -762.000000)">
+            <g id="do-btn-blue" transform="translate(651.000000, 763.000000)">
+                <rect id="Rectangle-Copy" fill="#0069FF" x="0" y="0" width="200" height="40" rx="6"></rect>
+                <path d="M45,0 L45,40" id="Line-2" stroke="#FFFFFF" stroke-linecap="square"></path>
+                <g id="DO_Logo_horizontal_blue-Copy" transform="translate(13.000000, 11.000000)" fill="#FFFFFF">
+                    <path d="M10.0098493,20 L10.0098493,16.1262429 C14.12457,16.1262429 17.2897398,12.0548452 15.7269372,7.74627862 C15.1334679,6.14538921 13.8674,4.86072487 12.2650328,4.28756693 C7.952489,2.72620566 3.87733294,5.88845634 3.87733294,9.99938223 C3.87733294,9.99938223 3.87733294,9.99938223 3.87733294,9.99938223 L0,9.99938223 C0,3.45747613 6.3303395,-1.64165309 13.1948014,0.492866119 C16.2017127,1.42177726 18.57559,3.81322933 19.5053586,6.79760341 C21.6418482,13.6754986 16.5577943,20 10.0098493,20 Z" id="XMLID_49_"></path>
+                    <polygon id="XMLID_47_" points="9.56521739 15.6521739 6.08695652 15.6521739 6.08695652 12.173913 6.08695652 12.173913 9.56521739 12.173913 9.56521739 12.173913"></polygon>
+                    <polygon id="XMLID_46_" points="6.08695652 19.1304348 3.47826087 19.1304348 3.47826087 19.1304348 3.47826087 16.5217391 6.08695652 16.5217391"></polygon>
+                    <polygon id="XMLID_45_" points="3.47826087 16.5217391 0.869565217 16.5217391 0.869565217 16.5217391 0.869565217 13.9130435 0.869565217 13.9130435 3.47826087 13.9130435 3.47826087 13.9130435"></polygon>
+                </g>
+                <text id="Create-a-Droplet-Copy" font-family="Sailec-Medium, Sailec" font-size="16" font-weight="400" fill="#FFFFFF">
+                    <tspan x="58" y="26">Create a Droplet</tspan>
+                </text>
+            </g>
+        </g>
+    </g>
+</svg>

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>

bors.toml

@@ -0,0 +1,3 @@
+status = ["Test on macos-latest", "Test on ubuntu-latest"]
+# 4 hours timeout
+timeout-sec = 14400

bump.sh

@@ -0,0 +1,38 @@
+#!/usr/bin/bash
+
+NEW_VERSION=$1
+
+if [ -z "$NEW_VERSION" ]
+then
+	echo "error: a version number must be provided"
+	exit 1
+fi
+
+# find current version
+CURRENT_VERSION=$(cat **/*.toml | grep meilisearch | grep version | sed 's/.*\([0-9]\+\.[0-9]\+\.[0-9]\+\).*/\1/' | sed "1q;d")
+
+# bump all version in .toml
+echo "bumping from version $CURRENT_VERSION to version $NEW_VERSION"
+while true
+do
+	read -r -p "Continue (y/n)?" choice
+	case "$choice" in
+		y|Y ) break;;
+		n|N ) echo "aborting bump" && exit 0;;
+		* ) echo "invalid choice";;
+	esac
+done
+# update all crate version
+sed -i "s/version = \"$CURRENT_VERSION\"/version = \"$NEW_VERSION\"/" **/*.toml
+
+printf "running cargo check: "
+
+CARGO_CHECK=$(cargo check 2>&1)
+
+if [ $? != "0" ]
+then
+	printf "\033[31;1m FAIL \033[0m\n"
+	printf "$CARGO_CHECK"
+	exit 1
+fi
+printf "\033[32;1m OK \033[0m\n"

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 +1,187 @@
+#!/bin/sh
+
+# COLORS
+RED='\033[31m'
+GREEN='\033[32m'
+DEFAULT='\033[0m'
+
+# GLOBALS
+BINARY_NAME='meilisearch'
+GREP_SEMVER_REGEXP='v\([0-9]*\)[.]\([0-9]*\)[.]\([0-9]*\)$' # i.e. v[number].[number].[number]
+
+# FUNCTIONS
+
+# 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" < "_$SPECIAL_B" ]; then
+        return 0
+    fi
+
+    return 1
+}
+
+# Returns the tag of the latest stable release (in terms of semver and not of release date)
+get_latest() {
+    temp_file='temp_file' # temp_file needed because the grep would start before the download is over
+    curl -s 'https://api.github.com/repos/meilisearch/MeiliSearch/releases' > "$temp_file"
+    releases=$(cat "$temp_file" | \
+        grep -E "tag_name|draft|prerelease" \
+        | tr -d ',"' | cut -d ':' -f2 | tr -d ' ')
+        # Returns a list of [tag_name draft_boolean prerelease_boolean ...]
+        # Ex: v0.10.1 false false v0.9.1-rc.1 false true v0.9.0 false false...
+
+    i=0
+    latest=""
+    current_tag=""
+    for release_info in $releases; do
+        if [ $i -eq 0 ]; then # Cheking tag_name
+            if echo "$release_info" | grep -q "$GREP_SEMVER_REGEXP"; then # If it's not an alpha or beta release
+                current_tag=$release_info
+            else
+                current_tag=""
+            fi
+            i=1
+        elif [ $i -eq 1 ]; then # Checking draft boolean
+            if [ "$release_info" = "true" ]; then
+                current_tag=""
+            fi
+            i=2
+        elif [ $i -eq 2 ]; then # Checking prerelease boolean
+            if [ "$release_info" = "true" ]; then
+                current_tag=""
+            fi
+            i=0
+            if [ "$current_tag" != "" ]; then # If the current_tag is valid
+                if [ "$latest" = "" ]; then # If there is no latest yet
+                    latest="$current_tag"
+                else
+                    semverLT $current_tag $latest # Comparing latest and the current tag
+                    if [ $? -eq 1 ]; then
+                        latest="$current_tag"
+                    fi
+                fi
+            fi
+        fi
+    done
+
+    rm -f "$temp_file"
+    echo $latest
+}
+
+# Gets the OS by setting the $os variable
+# Returns 0 in case of success, 1 otherwise.
+get_os() {
+    os_name=$(uname -s)
+    case "$os_name" in
+    'Darwin')
+        os='macos'
+        ;;
+    'Linux')
+        os='linux'
+        ;;
+    *)
+        return 1
+    esac
+    return 0
+}
+
+# Gets the architecture by setting the $archi variable
+# Returns 0 in case of success, 1 otherwise.
+get_archi() {
+    architecture=$(uname -m)
+    case "$architecture" in
+    'x86_64' | 'amd64')
+        archi='amd64'
+        ;;
+    'aarch64')
+        archi='armv8'
+        ;;
+    *)
+        return 1
+    esac
+    return 0
+}
+
+success_usage() {
+    printf "$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 or your architecture yet.'
+    echo ''
+    echo 'However, you can easily compile the binary from the source files.'
+    echo 'Follow the steps at the page ("Source" tab): https://docs.meilisearch.com/guides/advanced_guides/installation.html'
+}
+
+# MAIN
+latest="$(get_latest)"
+get_os
+if [ "$?" -eq 1 ]; then
+    failure_usage
+    exit 1
+fi
+get_archi
+if [ "$?" -eq 1 ]; then
+    failure_usage
+    exit 1
+fi
+echo "Downloading MeiliSearch binary $latest for $os, architecture $archi..."
+release_file="meilisearch-$os-$archi"
+link="https://github.com/meilisearch/MeiliSearch/releases/download/$latest/$release_file"
+curl -OL "$link"
+mv "$release_file" "$BINARY_NAME"
+chmod 744 "$BINARY_NAME"
+success_usage

examples/create-database.rs

@@ -1,137 +0,0 @@
-#[global_allocator]
-static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc;
-
-use std::collections::{HashMap, HashSet};
-use std::io::{self, BufRead, BufReader};
-use std::path::{Path, PathBuf};
-use std::time::Instant;
-use std::error::Error;
-use std::borrow::Cow;
-use std::fs::File;
-
-use serde_derive::{Serialize, Deserialize};
-use structopt::StructOpt;
-
-use meilidb::database::{Database, Schema};
-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)?;
-
-    database.create_index("default", &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 mut update = database.start_update("default")?;
-
-        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!("committing update...");
-        database.commit_update(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 start = Instant::now();
-    let result = 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 {:.2?} at: {:?}", start.elapsed(), opt.database_path);
-    Ok(())
-}

examples/query-database.rs

@@ -1,210 +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::time::Instant;
-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,
-
-    /// The number of characters before and after the first match
-    #[structopt(short = "C", long = "context", default_value = "35")]
-    pub char_context: 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]) -> Vec<usize> {
-    let mut byte_indexes = BTreeMap::new();
-
-    for match_ in matches {
-        let char_index = match_.char_index as usize;
-        let char_length = match_.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,
-    matches: impl IntoIterator<Item=Match>,
-    context: usize,
-) -> (String, Vec<Match>)
-{
-    let mut matches = matches.into_iter().peekable();
-
-    let char_index = matches.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 matches = matches
-        .take_while(|m| {
-            (m.char_index as usize) + (m.char_length as usize) <= start + (context * 2)
-        })
-        .map(|match_| {
-            Match { char_index: match_.char_index - start as u32, ..match_ }
-        })
-        .collect();
-
-    (text, matches)
-}
-
-fn main() -> Result<(), Box<Error>> {
-    let _ = env_logger::init();
-    let opt = Opt::from_args();
-
-    let start = Instant::now();
-    let database = Database::open(&opt.database_path)?;
-    println!("database prepared for you in {:.2?}", start.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("default")?;
-        let schema = view.schema();
-
-        let start = Instant::now();
-
-        let builder = view.query_builder().unwrap();
-        let documents = builder.query(query, 0..opt.number_results);
-
-        let number_of_documents = documents.len();
-        for mut doc in documents {
-
-            doc.matches.sort_unstable_by_key(|m| (m.char_index, m.char_index));
-
-            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 matches = doc.matches.iter()
-                                        .filter(|m| SchemaAttr::new(m.attribute) == attr)
-                                        .cloned();
-                        let (text, matches) = crop_text(&text, matches, opt.char_context);
-                        let areas = create_highlight_areas(&text, &matches);
-                        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);
-                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 {:.2?} =====", number_of_documents, start.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 +1,57 @@
+[package]
+name = "meilisearch-core"
+version = "0.14.1"
+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"
+cow-utils = "0.1.2"
+crossbeam-channel = "0.4.2"
+deunicode = "1.1.0"
+either = "1.5.3"
+env_logger = "0.7.1"
+fst = "0.4.4"
+hashbrown = { version = "0.7.1", features = ["serde"] }
+heed = "0.8.0"
+indexmap = { version = "1.3.2", features = ["serde-1"] }
+intervaltree = "0.2.5"
+itertools = "0.9.0"
+levenshtein_automata = { version = "0.2.0", features = ["fst_automaton"] }
+log = "0.4.8"
+meilisearch-error = { path = "../meilisearch-error", version = "0.14.1" }
+meilisearch-schema = { path = "../meilisearch-schema", version = "0.14.1" }
+meilisearch-tokenizer = { path = "../meilisearch-tokenizer", version = "0.14.1" }
+meilisearch-types = { path = "../meilisearch-types", version = "0.14.1" }
+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 = { version = "1.0.50", features = ["preserve_order"] }
+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"
+rustyline = { version = "6.0.0", default-features = false }
+structopt = "0.3.12"
+tempfile = "3.1.0"
+termcolor = "1.1.0"
+
+[target.'cfg(unix)'.dev-dependencies]
+jemallocator = "0.3.2"
+
+[[bench]]
+name = "search_benchmark"
+harness = false

meilisearch-core/benches/search_benchmark.rs

@@ -0,0 +1,108 @@
+#[cfg(test)]
+#[macro_use]
+extern crate assert_matches;
+
+use std::error::Error;
+use std::fs::File;
+use std::io::BufReader;
+use std::iter;
+use std::path::Path;
+use std::sync::mpsc;
+
+use meilisearch_core::{Database, DatabaseOptions};
+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, DatabaseOptions::default()).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));
+
+    db.main_write::<_, _, Box<dyn Error>>(|writer| {
+        index.main.put_schema(writer, &Schema::with_primary_key("id")).unwrap();
+        Ok(())
+    }).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.to_update().unwrap()
+    };
+
+    db.update_write::<_, _, Box<dyn Error>>(|writer| {
+        let _update_id = index.settings_update(writer, settings_update).unwrap();
+        Ok(())
+    }).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 update_id = db.update_write::<_, _, Box<dyn Error>>(|writer| {
+        let update_id = additions.finalize(writer).unwrap();
+        Ok(update_id)
+    }).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 +1,470 @@
+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, DatabaseOptions, 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.to_update().unwrap()
+    };
+
+    db.update_write(|w| index.settings_update(w, settings))?;
+
+    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 update_id = db.update_write(|w| additions.finalize(w))?;
+
+        println!("committing update...");
+        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().unwrap();
+
+    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 result = builder.query(ref_reader, Some(&query), 0..command.number_results)?;
+
+                let mut retrieve_duration = Duration::default();
+
+                let number_of_documents = result.documents.len();
+                for mut doc in result.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().unwrap();
+
+    Ok(())
+}
+
+fn main() -> Result<(), Box<dyn Error>> {
+    env_logger::init();
+
+    let opt = Command::from_args();
+    let database = Database::open_or_create(opt.path(), DatabaseOptions::default())?;
+
+    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 +1,661 @@
+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, error};
+use sdset::{Set, SetBuf, exponential_search, SetOperation, Counter, duo::OpBuilder};
+use slice_group_by::{GroupBy, GroupByMut};
+
+use meilisearch_types::DocIndex;
+
+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, Index, RankedMap, MainReader, Error};
+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;
+
+#[derive(Debug, Default)]
+pub struct SortResult {
+    pub documents: Vec<Document>,
+    pub nb_hits: usize,
+    pub exhaustive_nb_hit: bool,
+    pub facets: Option<HashMap<String, HashMap<String, usize>>>,
+    pub exhaustive_facets_count: Option<bool>,
+}
+
+#[allow(clippy::too_many_arguments)]
+pub fn bucket_sort<'c, FI>(
+    reader: &heed::RoTxn<MainT>,
+    query: &str,
+    range: Range<usize>,
+    facets_docids: Option<SetBuf<DocumentId>>,
+    facet_count_docids: Option<HashMap<String, HashMap<String, (&str, Cow<Set<DocumentId>>)>>>,
+    filter: Option<FI>,
+    criteria: Criteria<'c>,
+    searchable_attrs: Option<ReorderedAttrs>,
+    index: &Index,
+) -> MResult<SortResult>
+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,
+            facets_docids,
+            facet_count_docids,
+            filter,
+            distinct,
+            distinct_size,
+            criteria,
+            searchable_attrs,
+            index,
+        );
+    }
+
+    let mut result = SortResult::default();
+
+    let words_set = index.main.words_fst(reader)?;
+    let stop_words = index.main.stop_words_fst(reader)?;
+
+    let context = QTContext {
+        words_set,
+        stop_words,
+        synonyms: index.synonyms,
+        postings_lists: index.postings_lists,
+        prefix_postings_lists: index.prefix_postings_lists_cache,
+    };
+
+    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 { mut docids, queries } = traverse_query_tree(reader, &context, &operation)?;
+    debug!("found {} documents", docids.len());
+    debug!("number of postings {:?}", queries.len());
+
+    if let Some(facets_docids) = facets_docids {
+        let intersection = sdset::duo::OpBuilder::new(docids.as_ref(), facets_docids.as_set())
+            .intersection()
+            .into_set_buf();
+        docids = Cow::Owned(intersection);
+    }
+
+    if let Some(f) = facet_count_docids {
+        // hardcoded value, until approximation optimization
+        result.exhaustive_facets_count = Some(true);
+        result.facets = Some(facet_count(f, &docids));
+    }
+
+    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: index.documents_fields_counts,
+            };
+
+            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 = index.main.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());
+
+    result.documents = documents;
+    result.nb_hits = docids.len();
+
+    Ok(result)
+}
+
+#[allow(clippy::too_many_arguments)]
+pub fn bucket_sort_with_distinct<'c, FI, FD>(
+    reader: &heed::RoTxn<MainT>,
+    query: &str,
+    range: Range<usize>,
+    facets_docids: Option<SetBuf<DocumentId>>,
+    facet_count_docids: Option<HashMap<String, HashMap<String, (&str, Cow<Set<DocumentId>>)>>>,
+    filter: Option<FI>,
+    distinct: FD,
+    distinct_size: usize,
+    criteria: Criteria<'c>,
+    searchable_attrs: Option<ReorderedAttrs>,
+    index: &Index,
+) -> MResult<SortResult>
+where
+    FI: Fn(DocumentId) -> bool,
+    FD: Fn(DocumentId) -> Option<u64>,
+{
+    let mut result = SortResult::default();
+
+    let words_set = index.main.words_fst(reader)?;
+    let stop_words = index.main.stop_words_fst(reader)?;
+
+    let context = QTContext {
+        words_set,
+        stop_words,
+        synonyms: index.synonyms,
+        postings_lists: index.postings_lists,
+        prefix_postings_lists: index.prefix_postings_lists_cache,
+    };
+
+    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 { mut docids, queries } = traverse_query_tree(reader, &context, &operation)?;
+    debug!("found {} documents", docids.len());
+    debug!("number of postings {:?}", queries.len());
+
+    if let Some(facets_docids) = facets_docids {
+        let intersection = OpBuilder::new(docids.as_ref(), facets_docids.as_set())
+            .intersection()
+            .into_set_buf();
+        docids = Cow::Owned(intersection);
+    }
+
+    if let Some(f) = facet_count_docids {
+        // hardcoded value, until approximation optimization
+        result.exhaustive_facets_count = Some(true);
+        result.facets = Some(facet_count(f, &docids));
+    }
+
+    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: index.documents_fields_counts,
+            };
+
+            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 = index.main.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_or_else(|| {
+                error!("error during filtering: expected value for document id {}", &raw_document.id.0);
+                Default::default()
+            }),
+            None => true,
+        };
+
+        if filter_accepted {
+            let key = key_cache.remove(&raw_document.id).unwrap_or_else(|| {
+                error!("error during distinct: expected value for document id {}", &raw_document.id.0);
+                Default::default()
+            });
+            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;
+                }
+            }
+        }
+    }
+    result.documents = documents;
+    result.nb_hits = docids.len();
+
+    Ok(result)
+}
+
+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,
+        }
+    }
+}
+
+/// sorts documents ids according to user defined ranking rules.
+pub fn placeholder_document_sort(
+    document_ids: &mut [DocumentId],
+    index: &store::Index,
+    reader: &MainReader,
+    ranked_map: &RankedMap
+) -> MResult<()> {
+    use crate::settings::RankingRule;
+    use std::cmp::Ordering;
+
+    enum SortOrder {
+        Asc,
+        Desc,
+    }
+
+    if let Some(ranking_rules) = index.main.ranking_rules(reader)? {
+        let schema = index.main.schema(reader)?
+            .ok_or(Error::SchemaMissing)?;
+
+        // Select custom rules from ranking rules, and map them to custom rules
+        // containing a field_id
+        let ranking_rules = ranking_rules.iter().filter_map(|r|
+            match r {
+                RankingRule::Asc(name) => schema.id(name).map(|f| (f, SortOrder::Asc)),
+                RankingRule::Desc(name) => schema.id(name).map(|f| (f, SortOrder::Desc)),
+                _ => None,
+            }).collect::<Vec<_>>();
+
+        document_ids.sort_unstable_by(|a, b| {
+            for (field_id, order) in &ranking_rules {
+                let a_value = ranked_map.get(*a, *field_id);
+                let b_value = ranked_map.get(*b, *field_id);
+                let (a, b) = match order {
+                    SortOrder::Asc => (a_value, b_value),
+                    SortOrder::Desc => (b_value, a_value),
+                };
+                match a.cmp(&b) {
+                    Ordering::Equal => continue,
+                    ordering => return ordering,
+                }
+            }
+            Ordering::Equal
+        });
+    }
+    Ok(())
+}
+
+/// For each entry in facet_docids, calculates the number of documents in the intersection with candidate_docids.
+pub fn facet_count(
+    facet_docids: HashMap<String, HashMap<String, (&str, Cow<Set<DocumentId>>)>>,
+    candidate_docids: &Set<DocumentId>,
+) -> HashMap<String, HashMap<String, usize>> {
+    let mut facets_counts = HashMap::with_capacity(facet_docids.len());
+    for (key, doc_map) in facet_docids {
+        let mut count_map = HashMap::with_capacity(doc_map.len());
+        for (_, (value, docids)) in doc_map {
+            let mut counter = Counter::new();
+            let op = OpBuilder::new(docids.as_ref(), candidate_docids).intersection();
+            SetOperation::<DocumentId>::extend_collection(op, &mut counter);
+            count_map.insert(value.to_string(), counter.0);
+        }
+        facets_counts.insert(key, count_map);
+    }
+    facets_counts
+}

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 +1,292 @@
+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)
+    }
+
+    #[allow(clippy::should_implement_trait)]
+    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

@@ -1,11 +1,9 @@
 use std::cmp::Ordering;
 use std::error::Error;
 use std::fmt;
-
-use crate::database::schema::{Schema, SchemaAttr};
-use crate::rank::criterion::Criterion;
-use crate::database::RankedMap;
-use crate::rank::RawDocument;
+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.
@@ -24,17 +22,17 @@ use crate::rank::RawDocument;
 ///
 /// ```ignore
 /// use serde_derive::Deserialize;
-/// use meilidb::rank::criterion::*;
+/// use meilisearch::rank::criterion::*;
 ///
 /// let custom_ranking = SortByAttr::lower_is_better(&ranked_map, &schema, "published_at")?;
 ///
 /// let builder = CriteriaBuilder::with_capacity(8)
-///        .add(SumOfTypos)
-///        .add(NumberOfWords)
-///        .add(WordsProximity)
-///        .add(SumOfWordsAttribute)
-///        .add(SumOfWordsPosition)
-///        .add(Exact)
+///        .add(Typo)
+///        .add(Words)
+///        .add(Proximity)
+///        .add(Attribute)
+///        .add(WordsPosition)
+///        .add(Exactness)
 ///        .add(custom_ranking)
 ///        .add(DocumentId);
 ///
@@ -43,7 +41,7 @@ use crate::rank::RawDocument;
 /// ```
 pub struct SortByAttr<'a> {
     ranked_map: &'a RankedMap,
-    attr: SchemaAttr,
+    field_id: FieldId,
     reversed: bool,
 }
 
@@ -52,8 +50,7 @@ impl<'a> SortByAttr<'a> {
         ranked_map: &'a RankedMap,
         schema: &Schema,
         attr_name: &str,
-    ) -> Result<SortByAttr<'a>, SortByAttrError>
-    {
+    ) -> Result<SortByAttr<'a>, SortByAttrError> {
         SortByAttr::new(ranked_map, schema, attr_name, false)
     }
 
@@ -61,8 +58,7 @@ impl<'a> SortByAttr<'a> {
         ranked_map: &'a RankedMap,
         schema: &Schema,
         attr_name: &str,
-    ) -> Result<SortByAttr<'a>, SortByAttrError>
-    {
+    ) -> Result<SortByAttr<'a>, SortByAttrError> {
         SortByAttr::new(ranked_map, schema, attr_name, true)
     }
 
@@ -71,34 +67,45 @@ impl<'a> SortByAttr<'a> {
         schema: &Schema,
         attr_name: &str,
         reversed: bool,
-    ) -> Result<SortByAttr<'a>, SortByAttrError>
-    {
-        let attr = match schema.attribute(attr_name) {
-            Some(attr) => attr,
+    ) -> Result<SortByAttr<'a>, SortByAttrError> {
+        let field_id = match schema.id(attr_name) {
+            Some(field_id) => field_id,
             None => return Err(SortByAttrError::AttributeNotFound),
         };
 
-        if schema.props(attr).is_ranked() {
+        if !schema.is_ranked(field_id) {
             return Err(SortByAttrError::AttributeNotRegisteredForRanking);
         }
 
-        Ok(SortByAttr { ranked_map, attr, reversed })
+        Ok(SortByAttr {
+            ranked_map,
+            field_id,
+            reversed,
+        })
     }
 }
 
-impl<'a> Criterion for SortByAttr<'a> {
-    fn evaluate(&self, lhs: &RawDocument, rhs: &RawDocument) -> Ordering {
-        let lhs = self.ranked_map.get(&(lhs.id, self.attr));
-        let rhs = self.ranked_map.get(&(rhs.id, self.attr));
+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,
+                if self.reversed {
+                    order.reverse()
+                } else {
+                    order
+                }
+            }
+            (None, Some(_)) => Ordering::Greater,
+            (Some(_), None) => Ordering::Less,
+            (None, None) => Ordering::Equal,
         }
     }
 }
@@ -119,4 +126,4 @@ impl fmt::Display for SortByAttrError {
     }
 }
 
-impl Error for SortByAttrError { }
+impl Error for SortByAttrError {}

meilisearch-core/src/criterion/typo.rs

@@ -0,0 +1,56 @@
+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]
+        #[allow(clippy::approx_constant)]
+        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 +1,224 @@
+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;
+
+use meilisearch_error::{ErrorCode, Code};
+
+pub type MResult<T> = Result<T, Error>;
+
+#[derive(Debug)]
+pub enum Error {
+    Bincode(bincode::Error),
+    Deserializer(DeserializerError),
+    FacetError(FacetError),
+    FilterParseError(PestError<Rule>),
+    Fst(fst::Error),
+    Heed(heed::Error),
+    IndexAlreadyExists,
+    Io(io::Error),
+    MaxFieldsLimitExceeded,
+    MissingDocumentId,
+    MissingPrimaryKey,
+    Schema(meilisearch_schema::Error),
+    SchemaMissing,
+    SerdeJson(SerdeJsonError),
+    Serializer(SerializerError),
+    VersionMismatch(String),
+    WordIndexMissing,
+}
+
+impl ErrorCode for Error {
+    fn error_code(&self) -> Code {
+        use Error::*;
+
+        match self {
+            FacetError(_) => Code::Facet,
+            FilterParseError(_) => Code::Filter,
+            IndexAlreadyExists => Code::IndexAlreadyExists,
+            MissingPrimaryKey => Code::MissingPrimaryKey,
+            MissingDocumentId => Code::MissingDocumentId,
+            MaxFieldsLimitExceeded => Code::MaxFieldsLimitExceeded,
+            Schema(s) =>  s.error_code(),
+            WordIndexMissing
+            | SchemaMissing => Code::InvalidState,
+            Heed(_)
+            | Fst(_)
+            | SerdeJson(_)
+            | Bincode(_)
+            | Serializer(_)
+            | Deserializer(_)
+            | VersionMismatch(_)
+            | Io(_) => Code::Internal,
+        }
+    }
+}
+
+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<FacetError> for Error {
+    fn from(error: FacetError) -> Error {
+        Error::FacetError(error)
+    }
+}
+
+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::Heed(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 {
+        match error {
+            SerializerError::DocumentIdNotFound => Error::MissingDocumentId,
+            e => Error::Serializer(e),
+        }
+    }
+}
+
+impl From<DeserializerError> for Error {
+    fn from(error: DeserializerError) -> Error {
+        Error::Deserializer(error)
+    }
+}
+
+impl fmt::Display for Error {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        use self::Error::*;
+        match self {
+            Bincode(e) => write!(f, "bincode error; {}", e),
+            Deserializer(e) => write!(f, "deserializer error; {}", e),
+            FacetError(e) => write!(f, "error processing facet filter: {}", e),
+            FilterParseError(e) => write!(f, "error parsing filter; {}", e),
+            Fst(e) => write!(f, "fst error; {}", e),
+            Heed(e) => write!(f, "heed error; {}", e),
+            IndexAlreadyExists => write!(f, "index already exists"),
+            Io(e) => write!(f, "{}", e),
+            MaxFieldsLimitExceeded => write!(f, "maximum number of fields in a document exceeded"),
+            MissingDocumentId => write!(f, "document id is missing"),
+            MissingPrimaryKey => write!(f, "schema cannot be built without a primary key"),
+            Schema(e) => write!(f, "schema error; {}", e),
+            SchemaMissing => write!(f, "this index does not have a schema"),
+            SerdeJson(e) => write!(f, "serde json error; {}", e),
+            Serializer(e) => write!(f, "serializer error; {}", e),
+            VersionMismatch(version) => write!(f, "Cannot open database, expected MeiliSearch engine version: {}, current engine version: {}.{}.{}",
+                version,
+                env!("CARGO_PKG_VERSION_MAJOR"),
+                env!("CARGO_PKG_VERSION_MINOR"),
+                env!("CARGO_PKG_VERSION_PATCH")),
+            WordIndexMissing => write!(f, "this index does not have a word index"),
+        }
+    }
+}
+
+impl error::Error for Error {}
+
+struct FilterParseError(PestError<Rule>);
+
+impl fmt::Display for FilterParseError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        use crate::pest_error::LineColLocation::*;
+
+        let (line, column) = match self.0.line_col {
+            Span((line, _), (column, _)) => (line, column),
+            Pos((line, column)) => (line, column),
+        };
+        write!(f, "parsing error on line {} at column {}: {}", line, column, self.0.variant.message())
+    }
+}
+
+#[derive(Debug)]
+pub enum FacetError {
+    EmptyArray,
+    ParsingError(String),
+    UnexpectedToken { expected: &'static [&'static str], found: String },
+    InvalidFormat(String),
+    AttributeNotFound(String),
+    AttributeNotSet { expected: Vec<String>, found: String },
+    InvalidDocumentAttribute(String),
+    NoAttributesForFaceting,
+}
+
+impl FacetError {
+    pub fn unexpected_token(expected: &'static [&'static str], found: impl ToString) -> FacetError {
+        FacetError::UnexpectedToken{ expected, found: found.to_string() }
+    }
+
+    pub fn attribute_not_set(expected: Vec<String>, found: impl ToString) -> FacetError {
+        FacetError::AttributeNotSet{ expected, found: found.to_string() }
+    }
+}
+
+impl fmt::Display for FacetError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        use FacetError::*;
+
+        match self {
+            EmptyArray => write!(f, "empty array in facet filter is unspecified behavior"),
+            ParsingError(msg) => write!(f, "parsing error: {}", msg),
+            UnexpectedToken { expected, found } => write!(f, "unexpected token {}, expected {}", found, expected.join("or")),
+            InvalidFormat(found) => write!(f, "invalid facet: {}, facets should be \"facetName:facetValue\"", found),
+            AttributeNotFound(attr) => write!(f, "unknown {:?} attribute", attr),
+            AttributeNotSet { found, expected } => write!(f, "`{}` is not set as a faceted attribute. available facet attributes: {}", found, expected.join(", ")),
+            InvalidDocumentAttribute(attr) => write!(f, "invalid document attribute {}, accepted types: String and [String]", attr),
+            NoAttributesForFaceting => write!(f, "impossible to perform faceted search, no attributes for faceting are set"),
+        }
+    }
+}

meilisearch-core/src/facets.rs

@@ -0,0 +1,357 @@
+use std::borrow::Cow;
+use std::collections::HashMap;
+use std::hash::Hash;
+use std::ops::Deref;
+
+use cow_utils::CowUtils;
+use either::Either;
+use heed::types::{Str, OwnedType};
+use indexmap::IndexMap;
+use serde_json::Value;
+
+use meilisearch_schema::{FieldId, Schema};
+use meilisearch_types::DocumentId;
+
+use crate::database::MainT;
+use crate::error::{FacetError, MResult};
+use crate::store::BEU16;
+
+/// Data structure used to represent a boolean expression in the form of nested arrays.
+/// Values in the outer array are and-ed together, values in the inner arrays are or-ed together.
+#[derive(Debug, PartialEq)]
+pub struct FacetFilter(Vec<Either<Vec<FacetKey>, FacetKey>>);
+
+impl Deref for FacetFilter {
+    type Target = Vec<Either<Vec<FacetKey>, FacetKey>>;
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+impl FacetFilter {
+    pub fn from_str(
+        s: &str,
+        schema: &Schema,
+        attributes_for_faceting: &[FieldId],
+    ) -> MResult<FacetFilter> {
+        if attributes_for_faceting.is_empty() {
+            return Err(FacetError::NoAttributesForFaceting.into());
+        }
+        let parsed = serde_json::from_str::<Value>(s).map_err(|e| FacetError::ParsingError(e.to_string()))?;
+        let mut filter = Vec::new();
+        match parsed {
+            Value::Array(and_exprs) => {
+                if and_exprs.is_empty() {
+                    return Err(FacetError::EmptyArray.into());
+                }
+                for expr in and_exprs {
+                    match expr {
+                        Value::String(s) => {
+                            let key = FacetKey::from_str( &s, schema, attributes_for_faceting)?;
+                            filter.push(Either::Right(key));
+                        }
+                        Value::Array(or_exprs) => {
+                            if or_exprs.is_empty() {
+                                return Err(FacetError::EmptyArray.into());
+                            }
+                            let mut inner = Vec::new();
+                            for expr in or_exprs {
+                                match expr {
+                                    Value::String(s) => {
+                                        let key = FacetKey::from_str( &s, schema, attributes_for_faceting)?;
+                                        inner.push(key);
+                                    }
+                                    bad_value => return Err(FacetError::unexpected_token(&["String"], bad_value).into()),
+                                }
+                            }
+                            filter.push(Either::Left(inner));
+                        }
+                        bad_value => return Err(FacetError::unexpected_token(&["Array", "String"], bad_value).into()),
+                    }
+                }
+                Ok(Self(filter))
+            }
+            bad_value => Err(FacetError::unexpected_token(&["Array"], bad_value).into()),
+        }
+    }
+}
+
+#[derive(Debug, Eq, PartialEq, Hash)]
+#[repr(C)]
+pub struct FacetKey(FieldId, String);
+
+impl FacetKey {
+    pub fn new(field_id: FieldId, value: String) -> Self {
+        let value = match value.cow_to_lowercase() {
+                Cow::Borrowed(_) => value,
+                Cow::Owned(s) => s,
+        };
+        Self(field_id, value)
+    }
+
+    pub fn key(&self) -> FieldId {
+        self.0
+    }
+
+    pub fn value(&self) -> &str {
+        &self.1
+    }
+
+    // TODO improve parser
+    fn from_str(
+        s: &str,
+        schema: &Schema,
+        attributes_for_faceting: &[FieldId],
+    ) -> Result<Self, FacetError> {
+        let mut split = s.splitn(2, ':');
+        let key = split
+            .next()
+            .ok_or_else(|| FacetError::InvalidFormat(s.to_string()))?
+            .trim();
+        let field_id = schema
+            .id(key)
+            .ok_or_else(|| FacetError::AttributeNotFound(key.to_string()))?;
+
+        if !attributes_for_faceting.contains(&field_id) {
+            return Err(FacetError::attribute_not_set(
+                    attributes_for_faceting
+                    .iter()
+                    .filter_map(|&id| schema.name(id))
+                    .map(str::to_string)
+                    .collect::<Vec<_>>(),
+                    key))
+        }
+        let value = split
+            .next()
+            .ok_or_else(|| FacetError::InvalidFormat(s.to_string()))?
+            .trim();
+        // unquoting the string if need be:
+        let mut indices = value.char_indices();
+        let value =  match (indices.next(), indices.last()) {
+            (Some((s, '\'')), Some((e, '\''))) |
+            (Some((s, '\"')), Some((e, '\"'))) => value[s + 1..e].to_string(),
+            _ => value.to_string(),
+        };
+        Ok(Self::new(field_id, value))
+    }
+}
+
+impl<'a> heed::BytesEncode<'a> for FacetKey {
+    type EItem = FacetKey;
+
+    fn bytes_encode(item: &'a Self::EItem) -> Option<Cow<'a, [u8]>> {
+        let mut buffer = Vec::with_capacity(2 + item.1.len());
+        let id = BEU16::new(item.key().into());
+        let id_bytes = OwnedType::bytes_encode(&id)?;
+        let value_bytes = Str::bytes_encode(item.value())?;
+        buffer.extend_from_slice(id_bytes.as_ref());
+        buffer.extend_from_slice(value_bytes.as_ref());
+        Some(Cow::Owned(buffer))
+    }
+}
+
+impl<'a> heed::BytesDecode<'a> for FacetKey {
+    type DItem = FacetKey;
+
+    fn bytes_decode(bytes: &'a [u8]) -> Option<Self::DItem> {
+        let (id_bytes, value_bytes) = bytes.split_at(2);
+        let id = OwnedType::<BEU16>::bytes_decode(id_bytes)?;
+        let id = id.get().into();
+        let string = Str::bytes_decode(&value_bytes)?;
+        Some(FacetKey(id, string.to_string()))
+    }
+}
+
+pub fn add_to_facet_map(
+    facet_map: &mut HashMap<FacetKey, (String, Vec<DocumentId>)>,
+    field_id: FieldId,
+    value: Value,
+    document_id: DocumentId,
+) -> Result<(), FacetError> {
+    let value = match value {
+        Value::String(s) => s,
+        // ignore null
+        Value::Null => return Ok(()),
+        value => return Err(FacetError::InvalidDocumentAttribute(value.to_string())),
+    };
+    let key = FacetKey::new(field_id, value.clone());
+    facet_map.entry(key).or_insert_with(|| (value, Vec::new())).1.push(document_id);
+    Ok(())
+}
+
+pub fn facet_map_from_docids(
+    rtxn: &heed::RoTxn<MainT>,
+    index: &crate::Index,
+    document_ids: &[DocumentId],
+    attributes_for_facetting: &[FieldId],
+) -> MResult<HashMap<FacetKey, (String, Vec<DocumentId>)>> {
+    // A hashmap that ascociate a facet key to a pair containing the original facet attribute
+    // string with it's case preserved, and a list of document ids for that facet attribute.
+    let mut facet_map: HashMap<FacetKey, (String, Vec<DocumentId>)> = HashMap::new();
+    for document_id in document_ids {
+        for result in index
+            .documents_fields
+            .document_fields(rtxn, *document_id)?
+        {
+            let (field_id, bytes) = result?;
+            if attributes_for_facetting.contains(&field_id) {
+                match serde_json::from_slice(bytes)? {
+                    Value::Array(values) => {
+                        for v in values {
+                            add_to_facet_map(&mut facet_map, field_id, v, *document_id)?;
+                        }
+                    }
+                    v => add_to_facet_map(&mut facet_map, field_id, v, *document_id)?,
+                };
+            }
+        }
+    }
+    Ok(facet_map)
+}
+
+pub fn facet_map_from_docs(
+    schema: &Schema,
+    documents: &HashMap<DocumentId, IndexMap<String, Value>>,
+    attributes_for_facetting: &[FieldId],
+) -> MResult<HashMap<FacetKey, (String, Vec<DocumentId>)>> {
+    let mut facet_map = HashMap::new();
+    let attributes_for_facetting = attributes_for_facetting
+        .iter()
+        .filter_map(|&id| schema.name(id).map(|name| (id, name)))
+        .collect::<Vec<_>>();
+
+    for (id, document) in documents {
+        for (field_id, name) in &attributes_for_facetting {
+            if let Some(value) = document.get(*name) {
+                match value {
+                    Value::Array(values) => {
+                        for v in values {
+                            add_to_facet_map(&mut facet_map, *field_id, v.clone(), *id)?;
+                        }
+                    }
+                    v => add_to_facet_map(&mut facet_map, *field_id, v.clone(), *id)?,
+                }
+            }
+        }
+    }
+    Ok(facet_map)
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use meilisearch_schema::Schema;
+
+    #[test]
+    fn test_facet_key() {
+        let mut schema = Schema::new();
+        let id = schema.insert_and_index("hello").unwrap();
+        let facet_list = [schema.id("hello").unwrap()];
+        assert_eq!(
+            FacetKey::from_str("hello:12", &schema, &facet_list).unwrap(),
+            FacetKey::new(id, "12".to_string())
+        );
+        assert_eq!(
+            FacetKey::from_str("hello:\"foo bar\"", &schema, &facet_list).unwrap(),
+            FacetKey::new(id, "foo bar".to_string())
+        );
+        assert_eq!(
+            FacetKey::from_str("hello:'foo bar'", &schema, &facet_list).unwrap(),
+            FacetKey::new(id, "foo bar".to_string())
+        );
+        // weird case
+        assert_eq!(
+            FacetKey::from_str("hello:blabla:machin", &schema, &facet_list).unwrap(),
+            FacetKey::new(id, "blabla:machin".to_string())
+        );
+
+        assert_eq!(
+            FacetKey::from_str("hello:\"\"", &schema, &facet_list).unwrap(),
+            FacetKey::new(id, "".to_string())
+        );
+
+        assert_eq!(
+            FacetKey::from_str("hello:'", &schema, &facet_list).unwrap(),
+            FacetKey::new(id, "'".to_string())
+        );
+        assert_eq!(
+            FacetKey::from_str("hello:''", &schema, &facet_list).unwrap(),
+            FacetKey::new(id, "".to_string())
+        );
+        assert!(FacetKey::from_str("hello", &schema, &facet_list).is_err());
+        assert!(FacetKey::from_str("toto:12", &schema, &facet_list).is_err());
+    }
+
+    #[test]
+    fn test_parse_facet_array() {
+        use either::Either::{Left, Right};
+        let mut schema = Schema::new();
+        let _id = schema.insert_and_index("hello").unwrap();
+        let facet_list = [schema.id("hello").unwrap()];
+        assert_eq!(
+            FacetFilter::from_str("[[\"hello:12\"]]", &schema, &facet_list).unwrap(),
+            FacetFilter(vec![Left(vec![FacetKey(FieldId(0), "12".to_string())])])
+        );
+        assert_eq!(
+            FacetFilter::from_str("[\"hello:12\"]", &schema, &facet_list).unwrap(),
+            FacetFilter(vec![Right(FacetKey(FieldId(0), "12".to_string()))])
+        );
+        assert_eq!(
+            FacetFilter::from_str("[\"hello:12\", \"hello:13\"]", &schema, &facet_list).unwrap(),
+            FacetFilter(vec![
+                Right(FacetKey(FieldId(0), "12".to_string())),
+                Right(FacetKey(FieldId(0), "13".to_string()))
+            ])
+        );
+        assert_eq!(
+            FacetFilter::from_str("[[\"hello:12\", \"hello:13\"]]", &schema, &facet_list).unwrap(),
+            FacetFilter(vec![Left(vec![
+                FacetKey(FieldId(0), "12".to_string()),
+                FacetKey(FieldId(0), "13".to_string())
+            ])])
+        );
+        assert_eq!(
+            FacetFilter::from_str(
+                "[[\"hello:12\", \"hello:13\"], \"hello:14\"]",
+                &schema,
+                &facet_list
+            )
+            .unwrap(),
+            FacetFilter(vec![
+                Left(vec![
+                    FacetKey(FieldId(0), "12".to_string()),
+                    FacetKey(FieldId(0), "13".to_string())
+                ]),
+                Right(FacetKey(FieldId(0), "14".to_string()))
+            ])
+        );
+
+        // invalid array depths
+        assert!(FacetFilter::from_str(
+            "[[[\"hello:12\", \"hello:13\"], \"hello:14\"]]",
+            &schema,
+            &facet_list
+        )
+        .is_err());
+        assert!(FacetFilter::from_str(
+            "[[[\"hello:12\", \"hello:13\"]], \"hello:14\"]]",
+            &schema,
+            &facet_list
+        )
+        .is_err());
+        assert!(FacetFilter::from_str("\"hello:14\"", &schema, &facet_list).is_err());
+
+        // unexisting key
+        assert!(FacetFilter::from_str("[\"foo:12\"]", &schema, &facet_list).is_err());
+
+        // invalid facet key
+        assert!(FacetFilter::from_str("[\"foo=12\"]", &schema, &facet_list).is_err());
+        assert!(FacetFilter::from_str("[\"foo12\"]", &schema, &facet_list).is_err());
+        assert!(FacetFilter::from_str("[\"\"]", &schema, &facet_list).is_err());
+
+        // empty array error
+        assert!(FacetFilter::from_str("[]", &schema, &facet_list).is_err());
+        assert!(FacetFilter::from_str("[\"hello:12\", []]", &schema, &facet_list).is_err());
+    }
+}

meilisearch-core/src/filters/condition.rs

@@ -0,0 +1,276 @@
+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 {
+        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!(),
+        }
+    }
+
+    pub fn as_str(&self) -> &str {
+        self.string
+    }
+
+    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_else(|| 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)?;
+        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 +1,202 @@
+#![allow(clippy::type_complexity)]
+
+#[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;
+pub mod criterion;
+pub mod facets;
+pub mod raw_indexer;
+pub mod serde;
+pub mod settings;
+pub mod store;
+pub mod update;
+
+pub use self::database::{BoxUpdateFn, Database, DatabaseOptions, MainT, UpdateT, MainWriter, MainReader, UpdateWriter, UpdateReader};
+pub use self::error::{Error, HeedError, FstError, MResult, pest_error, FacetError};
+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 compact_arena::SmallArena;
+use log::{error, trace};
+use std::borrow::Cow;
+use std::collections::HashMap;
+use std::convert::TryFrom;
+
+use crate::bucket_sort::PostingsListView;
+use crate::levenshtein::prefix_damerau_levenshtein;
+use crate::query_tree::{QueryId, QueryKind};
+use crate::reordered_attrs::ReorderedAttrs;
+
+type FstSetCow<'a> = fst::Set<Cow<'a, [u8]>>;
+type FstMapCow<'a> = fst::Map<Cow<'a, [u8]>>;
+
+#[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>(), 12);
+    }
+}

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)]
+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, FstSetCow};
+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<'a> {
+    pub words_set: FstSetCow<'a>,
+    pub stop_words: FstSetCow<'a>,
+    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_fst(reader, words.as_bytes())?;
+
+    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 +1,416 @@
+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() }
+    }
+
+    #[allow(clippy::len_zero)]
+    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].to_vec();
+            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 +1,312 @@
+use std::borrow::Cow;
+use std::collections::{BTreeMap, HashMap};
+use std::convert::TryFrom;
+
+use deunicode::deunicode_with_tofu;
+use meilisearch_schema::IndexedPos;
+use meilisearch_tokenizer::{is_cjk, SeqTokenizer, Token, Tokenizer};
+use sdset::SetBuf;
+
+use crate::{DocIndex, DocumentId};
+use crate::FstSetCow;
+
+const WORD_LENGTH_LIMIT: usize = 80;
+
+type Word = Vec<u8>; // TODO make it be a SmallVec
+
+pub struct RawIndexer<A> {
+    word_limit: usize, // the maximum number of indexed words
+    stop_words: fst::Set<A>,
+    words_doc_indexes: BTreeMap<Word, Vec<DocIndex>>,
+    docs_words: HashMap<DocumentId, Vec<Word>>,
+}
+
+pub struct Indexed<'a> {
+    pub words_doc_indexes: BTreeMap<Word, SetBuf<DocIndex>>,
+    pub docs_words: HashMap<DocumentId, FstSetCow<'a>>,
+}
+
+impl<A> RawIndexer<A> {
+    pub fn new(stop_words: fst::Set<A>) -> RawIndexer<A> {
+        RawIndexer::with_word_limit(stop_words, 1000)
+    }
+
+    pub fn with_word_limit(stop_words: fst::Set<A>, limit: usize) -> RawIndexer<A> {
+        RawIndexer {
+            word_limit: limit,
+            stop_words,
+            words_doc_indexes: BTreeMap::new(),
+            docs_words: HashMap::new(),
+        }
+    }
+}
+
+impl<A: AsRef<[u8]>> RawIndexer<A> {
+    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<'s, I>(&mut self, id: DocumentId, indexed_pos: IndexedPos, iter: I)
+    where
+        I: IntoIterator<Item = &'s 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<'static> {
+        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();
+                let fst = fst::Set::from_iter(words).unwrap().map_data(Cow::Owned).unwrap();
+                (id, fst)
+            })
+            .collect();
+
+        Indexed {
+            words_doc_indexes,
+            docs_words,
+        }
+    }
+}
+
+fn index_token<A>(
+    token: Token,
+    id: DocumentId,
+    indexed_pos: IndexedPos,
+    word_limit: usize,
+    stop_words: &fst::Set<A>,
+    words_doc_indexes: &mut BTreeMap<Word, Vec<DocIndex>>,
+    docs_words: &mut HashMap<DocumentId, Vec<Word>>,
+) -> bool
+where A: AsRef<[u8]>,
+{
+    if token.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());
+    }
+
+    #[test]
+    // test sample from 807
+    fn very_long_text() {
+        let mut indexer = RawIndexer::new(fst::Set::default());
+        let indexed_pos = IndexedPos(0);
+        let docid = DocumentId(0);
+        let text = " The locations block is the most powerful, and potentially most involved, section of the .platform.app.yaml file. It allows you to control how the application container responds to incoming requests at a very fine-grained level. Common patterns also vary between language containers due to the way PHP-FPM handles incoming requests.\nEach entry of the locations block is an absolute URI path (with leading /) and its value includes the configuration directives for how the web server should handle matching requests. That is, if your domain is example.com then '/' means &ldquo;requests for example.com/&rdquo;, while '/admin' means &ldquo;requests for example.com/admin&rdquo;. If multiple blocks could match an incoming request then the most-specific will apply.\nweb:locations:&#39;/&#39;:# Rules for all requests that don&#39;t otherwise match....&#39;/sites/default/files&#39;:# Rules for any requests that begin with /sites/default/files....The simplest possible locations configuration is one that simply passes all requests on to your application unconditionally:\nweb:locations:&#39;/&#39;:passthru:trueThat is, all requests to /* should be forwarded to the process started by web.commands.start above. Note that for PHP containers the passthru key must specify what PHP file the request should be forwarded to, and must also specify a docroot under which the file lives. For example:\nweb:locations:&#39;/&#39;:root:&#39;web&#39;passthru:&#39;/app.php&#39;This block will serve requests to / from the web directory in the application, and if a file doesn&rsquo;t exist on disk then the request will be forwarded to the /app.php script.\nA full list of the possible subkeys for locations is below.\n  root: The folder from which to serve static assets for this location relative to the application root. The application root is the directory in which the .platform.app.yaml file is located. Typical values for this property include public or web. Setting it to '' is not recommended, and its behavior may vary depending on the type of application. Absolute paths are not supported.\n  passthru: Whether to forward disallowed and missing resources from this location to the application and can be true, false or an absolute URI path (with leading /). The default value is false. For non-PHP applications it will generally be just true or false. In a PHP application this will typically be the front controller such as /index.php or /app.php. This entry works similar to mod_rewrite under Apache. Note: If the value of passthru does not begin with the same value as the location key it is under, the passthru may evaluate to another entry. That may be useful when you want different cache settings for different paths, for instance, but want missing files in all of them to map back to the same front controller. See the example block below.\n  index: The files to consider when serving a request for a directory: an array of file names or null. (typically ['index.html']). Note that in order for this to work, access to the static files named must be allowed by the allow or rules keys for this location.\n  expires: How long to allow static assets from this location to be cached (this enables the Cache-Control and Expires headers) and can be a time or -1 for no caching (default). Times can be suffixed with &ldquo;ms&rdquo; (milliseconds), &ldquo;s&rdquo; (seconds), &ldquo;m&rdquo; (minutes), &ldquo;h&rdquo; (hours), &ldquo;d&rdquo; (days), &ldquo;w&rdquo; (weeks), &ldquo;M&rdquo; (months, 30d) or &ldquo;y&rdquo; (years, 365d).\n  scripts: Whether to allow loading scripts in that location (true or false). This directive is only meaningful on PHP.\n  allow: Whether to allow serving files which don&rsquo;t match a rule (true or false, default: true).\n  headers: Any additional headers to apply to static assets. This section is a mapping of header names to header values. Responses from the application aren&rsquo;t affected, to avoid overlap with the application&rsquo;s own ability to include custom headers in the response.\n  rules: Specific overrides for a specific location. The key is a PCRE (regular expression) that is matched against the full request path.\n  request_buffering: Most application servers do not support chunked requests (e.g. fpm, uwsgi), so Platform.sh enables request_buffering by default to handle them. That default configuration would look like this if it was present in .platform.app.yaml:\nweb:locations:&#39;/&#39;:passthru:truerequest_buffering:enabled:truemax_request_size:250mIf the application server can already efficiently handle chunked requests, the request_buffering subkey can be modified to disable it entirely (enabled: false). Additionally, applications that frequently deal with uploads greater than 250MB in size can update the max_request_size key to the application&rsquo;s needs. Note that modifications to request_buffering will need to be specified at each location where it is desired.\n ";
+        indexer.index_text(docid, indexed_pos, text);
+        let Indexed {
+            words_doc_indexes, ..
+        } = indexer.build();
+        assert!(words_doc_indexes.get(&"buffering".to_owned().into_bytes()).is_some());
+    }
+
+    #[test]
+    fn words_over_index_1000_not_indexed() {
+        let mut indexer = RawIndexer::new(fst::Set::default());
+        let indexed_pos = IndexedPos(0);
+        let docid = DocumentId(0);
+        let mut text = String::with_capacity(5000);
+        for _ in 0..1000 {
+            text.push_str("less ");
+        }
+        text.push_str("more");
+        indexer.index_text(docid, indexed_pos, &text);
+        let Indexed {
+            words_doc_indexes, ..
+        } = indexer.build();
+        assert!(words_doc_indexes.get(&"less".to_owned().into_bytes()).is_some());
+        assert!(words_doc_indexes.get(&"more".to_owned().into_bytes()).is_none());
+    }
+}

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/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/mod.rs

@@ -0,0 +1,92 @@
+mod deserializer;
+
+pub use self::deserializer::{Deserializer, DeserializerError};
+
+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,
+    InvalidDocumentIdFormat,
+    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("Primary key is missing.")
+            }
+            SerializerError::InvalidDocumentIdFormat => {
+                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/settings.rs

@@ -0,0 +1,183 @@
+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(|| {
+    regex::Regex::new(r"(asc|desc)\(([a-zA-Z0-9-_]*)\)").unwrap()
+});
+
+#[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 attributes_for_faceting: Option<Option<Vec<String>>>,
+}
+
+// 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 to_update(&self) -> Result<SettingsUpdate, RankingRuleConversionError> {
+        let settings = self.clone();
+
+        let ranking_rules = match settings.ranking_rules {
+            Some(Some(rules)) => UpdateState::Update(RankingRule::try_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(),
+            attributes_for_faceting: settings.attributes_for_faceting.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 try_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 attributes_for_faceting: UpdateState<Vec<String>>,
+}
+
+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,
+            attributes_for_faceting: UpdateState::Nothing,
+        }
+    }
+}

meilisearch-core/src/store/cow_set.rs

@@ -0,0 +1,32 @@
+use std::borrow::Cow;
+
+use heed::{types::CowSlice, BytesEncode, BytesDecode};
+use sdset::{Set, SetBuf};
+use zerocopy::{AsBytes, FromBytes};
+
+pub struct CowSet<T>(std::marker::PhantomData<T>);
+
+impl<'a, T: 'a> BytesEncode<'a> for CowSet<T>
+where
+    T: AsBytes,
+{
+    type EItem = Set<T>;
+
+    fn bytes_encode(item: &'a Self::EItem) -> Option<Cow<[u8]>> {
+        CowSlice::bytes_encode(item.as_slice())
+    }
+}
+
+impl<'a, T: 'a> BytesDecode<'a> for CowSet<T>
+where
+    T: FromBytes + Copy,
+{
+    type DItem = Cow<'a, Set<T>>;
+
+    fn bytes_decode(bytes: &'a [u8]) -> Option<Self::DItem> {
+        match CowSlice::<T>::bytes_decode(bytes)? {
+            Cow::Owned(vec) => Some(Cow::Owned(SetBuf::new_unchecked(vec))),
+            Cow::Borrowed(slice) => Some(Cow::Borrowed(Set::new_unchecked(slice))),
+        }
+    }
+}

meilisearch-core/src/store/docs_words.rs

@@ -0,0 +1,43 @@
+use std::borrow::Cow;
+
+use heed::Result as ZResult;
+use heed::types::{ByteSlice, OwnedType};
+
+use crate::database::MainT;
+use crate::{DocumentId, FstSetCow};
+use super::BEU32;
+
+#[derive(Copy, Clone)]
+pub struct DocsWords {
+    pub(crate) docs_words: heed::Database<OwnedType<BEU32>, ByteSlice>,
+}
+
+impl DocsWords {
+    pub fn put_doc_words(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        document_id: DocumentId,
+        words: &FstSetCow,
+    ) -> ZResult<()> {
+        let document_id = BEU32::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 = BEU32::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<FstSetCow> {
+        let document_id = BEU32::new(document_id.0);
+        match self.docs_words.get(reader, &document_id)? {
+            Some(bytes) => Ok(fst::Set::new(bytes).unwrap().map_data(Cow::Borrowed).unwrap()),
+            None => Ok(fst::Set::default().map_data(Cow::Owned).unwrap()),
+        }
+    }
+}

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 +1,143 @@
+use super::DocumentFieldIndexedKey;
+use crate::database::MainT;
+use crate::DocumentId;
+use heed::types::OwnedType;
+use heed::Result as ZResult;
+use meilisearch_schema::IndexedPos;
+use crate::MResult;
+
+#[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>) -> MResult<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 = MResult<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.into())),
+            }
+        }
+        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/documents_ids.rs

@@ -0,0 +1,75 @@
+use std::borrow::Cow;
+
+use heed::{BytesDecode, BytesEncode};
+use sdset::Set;
+
+use crate::DocumentId;
+use super::cow_set::CowSet;
+
+pub struct DocumentsIds;
+
+impl BytesEncode<'_> for DocumentsIds {
+    type EItem = Set<DocumentId>;
+
+    fn bytes_encode(item: &Self::EItem) -> Option<Cow<[u8]>> {
+        CowSet::bytes_encode(item)
+    }
+}
+
+impl<'a> BytesDecode<'a> for DocumentsIds {
+    type DItem = Cow<'a, Set<DocumentId>>;
+
+    fn bytes_decode(bytes: &'a [u8]) -> Option<Self::DItem> {
+        CowSet::bytes_decode(bytes)
+    }
+}
+
+pub struct DiscoverIds<'a> {
+    ids_iter: std::slice::Iter<'a, DocumentId>,
+    left_id: Option<u32>,
+    right_id: Option<u32>,
+    available_range: std::ops::Range<u32>,
+}
+
+impl DiscoverIds<'_> {
+    pub fn new(ids: &Set<DocumentId>) -> DiscoverIds {
+        let mut ids_iter = ids.iter();
+        let right_id = ids_iter.next().map(|id| id.0);
+        let available_range = 0..right_id.unwrap_or(u32::max_value());
+        DiscoverIds { ids_iter, left_id: None, right_id, available_range }
+    }
+}
+
+impl Iterator for DiscoverIds<'_> {
+    type Item = DocumentId;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        loop {
+            match self.available_range.next() {
+                // The available range gives us a new id, we return it.
+                Some(id) => return Some(DocumentId(id)),
+                // The available range is exhausted, we need to find the next one.
+                None if self.available_range.end == u32::max_value() => return None,
+                None => loop {
+                    self.left_id = self.right_id.take();
+                    self.right_id = self.ids_iter.next().map(|id| id.0);
+                    match (self.left_id, self.right_id) {
+                        // We found a gap in the used ids, we can yield all ids
+                        // until the end of the gap
+                        (Some(l), Some(r)) => if l.saturating_add(1) != r {
+                            self.available_range = (l + 1)..r;
+                            break;
+                        },
+                        // The last used id has been reached, we can use all ids
+                        // until u32 MAX
+                        (Some(l), None) => {
+                            self.available_range = l.saturating_add(1)..u32::max_value();
+                            break;
+                        },
+                        _ => (),
+                    }
+                },
+            }
+        }
+    }
+}

meilisearch-core/src/store/facets.rs

@@ -0,0 +1,97 @@
+use std::borrow::Cow;
+use std::collections::HashMap;
+use std::mem;
+
+use heed::{RwTxn, RoTxn, RoRange, types::Str, BytesEncode, BytesDecode};
+use sdset::{SetBuf, Set, SetOperation};
+
+use meilisearch_types::DocumentId;
+use meilisearch_schema::FieldId;
+
+use crate::MResult;
+use crate::database::MainT;
+use crate::facets::FacetKey;
+use super::cow_set::CowSet;
+
+/// contains facet info
+#[derive(Clone, Copy)]
+pub struct Facets {
+    pub(crate) facets: heed::Database<FacetKey, FacetData>,
+}
+
+pub struct FacetData;
+
+impl<'a> BytesEncode<'a> for FacetData {
+    type EItem = (&'a str, &'a Set<DocumentId>);
+
+    fn bytes_encode(item: &'a Self::EItem) -> Option<Cow<'a, [u8]>> {
+        // get size of the first item
+        let first_size =  item.0.as_bytes().len();
+        let size = mem::size_of::<u64>()
+            + first_size
+            + item.1.len() * mem::size_of::<DocumentId>();
+        let mut buffer = Vec::with_capacity(size);
+        // encode the length of the first item
+        buffer.extend_from_slice(&first_size.to_be_bytes());
+        buffer.extend_from_slice(Str::bytes_encode(&item.0)?.as_ref());
+        let second_slice = CowSet::bytes_encode(&item.1)?;
+        buffer.extend_from_slice(second_slice.as_ref());
+        Some(Cow::Owned(buffer))
+    }
+}
+
+impl<'a> BytesDecode<'a> for FacetData {
+    type DItem = (&'a str, Cow<'a, Set<DocumentId>>);
+
+    fn bytes_decode(bytes: &'a [u8]) -> Option<Self::DItem> {
+        const LEN: usize = mem::size_of::<u64>();
+        let mut size_buf = [0; LEN];
+        size_buf.copy_from_slice(bytes.get(0..LEN)?);
+        // decode size of the first item from the bytes
+        let first_size = usize::from_be_bytes(size_buf);
+        // decode first and second items
+        let first_item = Str::bytes_decode(bytes.get(LEN..(LEN + first_size))?)?;
+        let second_item = CowSet::bytes_decode(bytes.get((LEN + first_size)..)?)?;
+        Some((first_item, second_item))
+    }
+}
+
+impl Facets {
+    // we use sdset::SetBuf to ensure the docids are sorted.
+    pub fn put_facet_document_ids(&self, writer: &mut RwTxn<MainT>, facet_key: FacetKey, doc_ids: &Set<DocumentId>, facet_value: &str) -> MResult<()> {
+        Ok(self.facets.put(writer, &facet_key, &(facet_value, doc_ids))?)
+    }
+
+    pub fn field_document_ids<'txn>(&self, reader: &'txn RoTxn<MainT>, field_id: FieldId) -> MResult<RoRange<'txn, FacetKey, FacetData>> {
+        Ok(self.facets.prefix_iter(reader, &FacetKey::new(field_id, String::new()))?)
+    }
+
+    pub fn facet_document_ids<'txn>(&self, reader: &'txn RoTxn<MainT>, facet_key: &FacetKey) -> MResult<Option<(&'txn str,Cow<'txn, Set<DocumentId>>)>> {
+        Ok(self.facets.get(reader, &facet_key)?)
+    }
+
+    /// updates the facets  store, revmoving the documents from the facets provided in the
+    /// `facet_map` argument
+    pub fn remove(&self, writer: &mut RwTxn<MainT>, facet_map: HashMap<FacetKey, (String, Vec<DocumentId>)>) -> MResult<()> {
+        for (key, (name, document_ids)) in facet_map {
+            if let Some((_, old)) = self.facets.get(writer, &key)? {
+                let to_remove = SetBuf::from_dirty(document_ids);
+                let new = sdset::duo::OpBuilder::new(old.as_ref(), to_remove.as_set()).difference().into_set_buf();
+                self.facets.put(writer, &key, &(&name, new.as_set()))?;
+            }
+        }
+        Ok(())
+    }
+
+    pub fn add(&self, writer: &mut RwTxn<MainT>, facet_map: HashMap<FacetKey, (String, Vec<DocumentId>)>) -> MResult<()> {
+        for (key, (facet_name, document_ids)) in facet_map {
+            let set = SetBuf::from_dirty(document_ids);
+            self.put_facet_document_ids(writer, key, set.as_set(), &facet_name)?;
+        }
+        Ok(())
+    }
+
+    pub fn clear(self, writer: &mut heed::RwTxn<MainT>) -> MResult<()> {
+        Ok(self.facets.clear(writer)?)
+    }
+}

meilisearch-core/src/store/main.rs

@@ -0,0 +1,320 @@
+use std::borrow::Cow;
+use std::collections::HashMap;
+
+use chrono::{DateTime, Utc};
+use heed::types::{ByteSlice, OwnedType, SerdeBincode, Str, CowSlice};
+use meilisearch_schema::{FieldId, Schema};
+use meilisearch_types::DocumentId;
+use sdset::Set;
+
+use crate::database::MainT;
+use crate::{RankedMap, MResult};
+use crate::settings::RankingRule;
+use crate::{FstSetCow, FstMapCow};
+use super::{CowSet, DocumentsIds};
+
+const ATTRIBUTES_FOR_FACETING_KEY: &str = "attributes-for-faceting";
+const CREATED_AT_KEY: &str = "created-at";
+const CUSTOMS_KEY: &str = "customs";
+const DISTINCT_ATTRIBUTE_KEY: &str = "distinct-attribute";
+const EXTERNAL_DOCIDS_KEY: &str = "external-docids";
+const FIELDS_DISTRIBUTION_KEY: &str = "fields-distribution";
+const INTERNAL_DOCIDS_KEY: &str = "internal-docids";
+const NAME_KEY: &str = "name";
+const NUMBER_OF_DOCUMENTS_KEY: &str = "number-of-documents";
+const RANKED_MAP_KEY: &str = "ranked-map";
+const RANKING_RULES_KEY: &str = "ranking-rules";
+const SCHEMA_KEY: &str = "schema";
+const SORTED_DOCUMENT_IDS_CACHE_KEY: &str = "sorted-document-ids-cache";
+const STOP_WORDS_KEY: &str = "stop-words";
+const SYNONYMS_KEY: &str = "synonyms";
+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>) -> MResult<()> {
+        Ok(self.main.clear(writer)?)
+    }
+
+    pub fn put_name(self, writer: &mut heed::RwTxn<MainT>, name: &str) -> MResult<()> {
+        Ok(self.main.put::<_, Str, Str>(writer, NAME_KEY, name)?)
+    }
+
+    pub fn name(self, reader: &heed::RoTxn<MainT>) -> MResult<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>) -> MResult<()> {
+        Ok(self.main.put::<_, Str, SerdeDatetime>(writer, CREATED_AT_KEY, &Utc::now())?)
+    }
+
+    pub fn created_at(self, reader: &heed::RoTxn<MainT>) -> MResult<Option<DateTime<Utc>>> {
+        Ok(self.main.get::<_, Str, SerdeDatetime>(reader, CREATED_AT_KEY)?)
+    }
+
+    pub fn put_updated_at(self, writer: &mut heed::RwTxn<MainT>) -> MResult<()> {
+        Ok(self.main.put::<_, Str, SerdeDatetime>(writer, UPDATED_AT_KEY, &Utc::now())?)
+    }
+
+    pub fn updated_at(self, reader: &heed::RoTxn<MainT>) -> MResult<Option<DateTime<Utc>>> {
+        Ok(self.main.get::<_, Str, SerdeDatetime>(reader, UPDATED_AT_KEY)?)
+    }
+
+    pub fn put_internal_docids(self, writer: &mut heed::RwTxn<MainT>, ids: &sdset::Set<DocumentId>) -> MResult<()> {
+        Ok(self.main.put::<_, Str, DocumentsIds>(writer, INTERNAL_DOCIDS_KEY, ids)?)
+    }
+
+    pub fn internal_docids<'txn>(self, reader: &'txn heed::RoTxn<MainT>) -> MResult<Cow<'txn, sdset::Set<DocumentId>>> {
+        match self.main.get::<_, Str, DocumentsIds>(reader, INTERNAL_DOCIDS_KEY)? {
+            Some(ids) => Ok(ids),
+            None => Ok(Cow::default()),
+        }
+    }
+
+    pub fn merge_internal_docids(self, writer: &mut heed::RwTxn<MainT>, new_ids: &sdset::Set<DocumentId>) -> MResult<()> {
+        use sdset::SetOperation;
+
+        // We do an union of the old and new internal ids.
+        let internal_docids = self.internal_docids(writer)?;
+        let internal_docids = sdset::duo::Union::new(&internal_docids, new_ids).into_set_buf();
+        Ok(self.put_internal_docids(writer, &internal_docids)?)
+    }
+
+    pub fn remove_internal_docids(self, writer: &mut heed::RwTxn<MainT>, ids: &sdset::Set<DocumentId>) -> MResult<()> {
+        use sdset::SetOperation;
+
+        // We do a difference of the old and new internal ids.
+        let internal_docids = self.internal_docids(writer)?;
+        let internal_docids = sdset::duo::Difference::new(&internal_docids, ids).into_set_buf();
+        Ok(self.put_internal_docids(writer, &internal_docids)?)
+    }
+
+    pub fn put_external_docids<A>(self, writer: &mut heed::RwTxn<MainT>, ids: &fst::Map<A>) -> MResult<()>
+    where A: AsRef<[u8]>,
+    {
+        Ok(self.main.put::<_, Str, ByteSlice>(writer, EXTERNAL_DOCIDS_KEY, ids.as_fst().as_bytes())?)
+    }
+
+    pub fn merge_external_docids<A>(self, writer: &mut heed::RwTxn<MainT>, new_docids: &fst::Map<A>) -> MResult<()>
+    where A: AsRef<[u8]>,
+    {
+        use fst::{Streamer, IntoStreamer};
+
+        // Do an union of the old and the new set of external docids.
+        let external_docids = self.external_docids(writer)?;
+        let mut op = external_docids.op().add(new_docids.into_stream()).r#union();
+        let mut build = fst::MapBuilder::memory();
+        while let Some((docid, values)) = op.next() {
+            build.insert(docid, values[0].value).unwrap();
+        }
+        drop(op);
+
+        let external_docids = build.into_map();
+        Ok(self.put_external_docids(writer, &external_docids)?)
+    }
+
+    pub fn remove_external_docids<A>(self, writer: &mut heed::RwTxn<MainT>, ids: &fst::Map<A>) -> MResult<()>
+    where A: AsRef<[u8]>,
+    {
+        use fst::{Streamer, IntoStreamer};
+
+        // Do an union of the old and the new set of external docids.
+        let external_docids = self.external_docids(writer)?;
+        let mut op = external_docids.op().add(ids.into_stream()).difference();
+        let mut build = fst::MapBuilder::memory();
+        while let Some((docid, values)) = op.next() {
+            build.insert(docid, values[0].value).unwrap();
+        }
+        drop(op);
+
+        let external_docids = build.into_map();
+        self.put_external_docids(writer, &external_docids)
+    }
+
+    pub fn external_docids(self, reader: &heed::RoTxn<MainT>) -> MResult<FstMapCow> {
+        match self.main.get::<_, Str, ByteSlice>(reader, EXTERNAL_DOCIDS_KEY)? {
+            Some(bytes) => Ok(fst::Map::new(bytes).unwrap().map_data(Cow::Borrowed).unwrap()),
+            None => Ok(fst::Map::default().map_data(Cow::Owned).unwrap()),
+        }
+    }
+
+    pub fn external_to_internal_docid(self, reader: &heed::RoTxn<MainT>, external_docid: &str) -> MResult<Option<DocumentId>> {
+        let external_ids = self.external_docids(reader)?;
+        Ok(external_ids.get(external_docid).map(|id| DocumentId(id as u32)))
+    }
+
+    pub fn words_fst(self, reader: &heed::RoTxn<MainT>) -> MResult<FstSetCow> {
+        match self.main.get::<_, Str, ByteSlice>(reader, WORDS_KEY)? {
+            Some(bytes) => Ok(fst::Set::new(bytes).unwrap().map_data(Cow::Borrowed).unwrap()),
+            None => Ok(fst::Set::default().map_data(Cow::Owned).unwrap()),
+        }
+    }
+
+    pub fn put_words_fst<A: AsRef<[u8]>>(self, writer: &mut heed::RwTxn<MainT>, fst: &fst::Set<A>) -> MResult<()> {
+        Ok(self.main.put::<_, Str, ByteSlice>(writer, WORDS_KEY, fst.as_fst().as_bytes())?)
+    }
+
+    pub fn put_sorted_document_ids_cache(self, writer: &mut heed::RwTxn<MainT>, documents_ids: &[DocumentId]) -> MResult<()> {
+        Ok(self.main.put::<_, Str, CowSlice<DocumentId>>(writer, SORTED_DOCUMENT_IDS_CACHE_KEY, documents_ids)?)
+    }
+
+    pub fn sorted_document_ids_cache(self, reader: &heed::RoTxn<MainT>) -> MResult<Option<Cow<[DocumentId]>>> {
+        Ok(self.main.get::<_, Str, CowSlice<DocumentId>>(reader, SORTED_DOCUMENT_IDS_CACHE_KEY)?)
+    }
+
+    pub fn put_schema(self, writer: &mut heed::RwTxn<MainT>, schema: &Schema) -> MResult<()> {
+        Ok(self.main.put::<_, Str, SerdeBincode<Schema>>(writer, SCHEMA_KEY, schema)?)
+    }
+
+    pub fn schema(self, reader: &heed::RoTxn<MainT>) -> MResult<Option<Schema>> {
+        Ok(self.main.get::<_, Str, SerdeBincode<Schema>>(reader, SCHEMA_KEY)?)
+    }
+
+    pub fn delete_schema(self, writer: &mut heed::RwTxn<MainT>) -> MResult<bool> {
+        Ok(self.main.delete::<_, Str>(writer, SCHEMA_KEY)?)
+    }
+
+    pub fn put_ranked_map(self, writer: &mut heed::RwTxn<MainT>, ranked_map: &RankedMap) -> MResult<()> {
+        Ok(self.main.put::<_, Str, SerdeBincode<RankedMap>>(writer, RANKED_MAP_KEY, &ranked_map)?)
+    }
+
+    pub fn ranked_map(self, reader: &heed::RoTxn<MainT>) -> MResult<Option<RankedMap>> {
+        Ok(self.main.get::<_, Str, SerdeBincode<RankedMap>>(reader, RANKED_MAP_KEY)?)
+    }
+
+    pub fn put_synonyms_fst<A: AsRef<[u8]>>(self, writer: &mut heed::RwTxn<MainT>, fst: &fst::Set<A>) -> MResult<()> {
+        let bytes = fst.as_fst().as_bytes();
+        Ok(self.main.put::<_, Str, ByteSlice>(writer, SYNONYMS_KEY, bytes)?)
+    }
+
+    pub(crate) fn synonyms_fst(self, reader: &heed::RoTxn<MainT>) -> MResult<FstSetCow> {
+        match self.main.get::<_, Str, ByteSlice>(reader, SYNONYMS_KEY)? {
+            Some(bytes) => Ok(fst::Set::new(bytes).unwrap().map_data(Cow::Borrowed).unwrap()),
+            None => Ok(fst::Set::default().map_data(Cow::Owned).unwrap()),
+        }
+    }
+
+    pub fn synonyms(self, reader: &heed::RoTxn<MainT>) -> MResult<Vec<String>> {
+        let synonyms = self
+            .synonyms_fst(&reader)?
+            .stream()
+            .into_strs()?;
+        Ok(synonyms)
+    }
+
+    pub fn put_stop_words_fst<A: AsRef<[u8]>>(self, writer: &mut heed::RwTxn<MainT>, fst: &fst::Set<A>) -> MResult<()> {
+        let bytes = fst.as_fst().as_bytes();
+        Ok(self.main.put::<_, Str, ByteSlice>(writer, STOP_WORDS_KEY, bytes)?)
+    }
+
+    pub(crate) fn stop_words_fst(self, reader: &heed::RoTxn<MainT>) -> MResult<FstSetCow> {
+        match self.main.get::<_, Str, ByteSlice>(reader, STOP_WORDS_KEY)? {
+            Some(bytes) => Ok(fst::Set::new(bytes).unwrap().map_data(Cow::Borrowed).unwrap()),
+            None => Ok(fst::Set::default().map_data(Cow::Owned).unwrap()),
+        }
+    }
+
+    pub fn stop_words(self, reader: &heed::RoTxn<MainT>) -> MResult<Vec<String>> {
+        let stop_word_list = self
+            .stop_words_fst(reader)?
+            .stream()
+            .into_strs()?;
+        Ok(stop_word_list)
+    }
+
+    pub fn put_number_of_documents<F>(self, writer: &mut heed::RwTxn<MainT>, f: F) -> MResult<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>) -> MResult<u64> {
+        match self
+            .main
+            .get::<_, Str, OwnedType<u64>>(reader, NUMBER_OF_DOCUMENTS_KEY)? {
+            Some(value) => Ok(value),
+            None => Ok(0),
+        }
+    }
+
+    pub fn put_fields_distribution(
+        self,
+        writer: &mut heed::RwTxn<MainT>,
+        fields_frequency: &FreqsMap,
+    ) -> MResult<()> {
+        Ok(self.main.put::<_, Str, SerdeFreqsMap>(writer, FIELDS_DISTRIBUTION_KEY, fields_frequency)?)
+    }
+
+    pub fn fields_distribution(&self, reader: &heed::RoTxn<MainT>) -> MResult<Option<FreqsMap>> {
+        match self
+            .main
+            .get::<_, Str, SerdeFreqsMap>(reader, FIELDS_DISTRIBUTION_KEY)?
+        {
+            Some(freqs) => Ok(Some(freqs)),
+            None => Ok(None),
+        }
+    }
+
+    pub fn attributes_for_faceting<'txn>(&self, reader: &'txn heed::RoTxn<MainT>) -> MResult<Option<Cow<'txn, Set<FieldId>>>> {
+        Ok(self.main.get::<_, Str, CowSet<FieldId>>(reader, ATTRIBUTES_FOR_FACETING_KEY)?)
+    }
+
+    pub fn put_attributes_for_faceting(self, writer: &mut heed::RwTxn<MainT>, attributes: &Set<FieldId>) -> MResult<()> {
+        Ok(self.main.put::<_, Str, CowSet<FieldId>>(writer, ATTRIBUTES_FOR_FACETING_KEY, attributes)?)
+    }
+
+    pub fn delete_attributes_for_faceting(self, writer: &mut heed::RwTxn<MainT>) -> MResult<bool> {
+        Ok(self.main.delete::<_, Str>(writer, ATTRIBUTES_FOR_FACETING_KEY)?)
+    }
+
+    pub fn ranking_rules(&self, reader: &heed::RoTxn<MainT>) -> MResult<Option<Vec<RankingRule>>> {
+        Ok(self.main.get::<_, Str, SerdeBincode<Vec<RankingRule>>>(reader, RANKING_RULES_KEY)?)
+    }
+
+    pub fn put_ranking_rules(self, writer: &mut heed::RwTxn<MainT>, value: &[RankingRule]) -> MResult<()> {
+        Ok(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>) -> MResult<bool> {
+        Ok(self.main.delete::<_, Str>(writer, RANKING_RULES_KEY)?)
+    }
+
+    pub fn distinct_attribute(&self, reader: &heed::RoTxn<MainT>) -> MResult<Option<FieldId>> {
+        match self.main.get::<_, Str, OwnedType<u16>>(reader, DISTINCT_ATTRIBUTE_KEY)? {
+            Some(value) => Ok(Some(FieldId(value.to_owned()))),
+            None => Ok(None),
+        }
+    }
+
+    pub fn put_distinct_attribute(self, writer: &mut heed::RwTxn<MainT>, value: FieldId) -> MResult<()> {
+        Ok(self.main.put::<_, Str, OwnedType<u16>>(writer, DISTINCT_ATTRIBUTE_KEY, &value.0)?)
+    }
+
+    pub fn delete_distinct_attribute(self, writer: &mut heed::RwTxn<MainT>) -> MResult<bool> {
+        Ok(self.main.delete::<_, Str>(writer, DISTINCT_ATTRIBUTE_KEY)?)
+    }
+
+    pub fn put_customs(self, writer: &mut heed::RwTxn<MainT>, customs: &[u8]) -> MResult<()> {
+        Ok(self.main.put::<_, Str, ByteSlice>(writer, CUSTOMS_KEY, customs)?)
+    }
+
+    pub fn customs<'txn>(self, reader: &'txn heed::RoTxn<MainT>) -> MResult<Option<&'txn [u8]>> {
+        Ok(self.main.get::<_, Str, ByteSlice>(reader, CUSTOMS_KEY)?)
+    }
+}

meilisearch-core/src/store/mod.rs

@@ -0,0 +1,522 @@
+mod cow_set;
+mod docs_words;
+mod documents_ids;
+mod documents_fields;
+mod documents_fields_counts;
+mod facets;
+mod main;
+mod postings_lists;
+mod prefix_documents_cache;
+mod prefix_postings_lists_cache;
+mod synonyms;
+mod updates;
+mod updates_results;
+
+pub use self::cow_set::CowSet;
+pub use self::docs_words::DocsWords;
+pub use self::documents_fields::{DocumentFieldsIter, DocumentsFields};
+pub use self::documents_fields_counts::{DocumentFieldsCountsIter, DocumentsFieldsCounts, DocumentsIdsIter};
+pub use self::documents_ids::{DocumentsIds, DiscoverIds};
+pub use self::facets::Facets;
+pub use self::main::Main;
+pub use self::postings_lists::PostingsLists;
+pub use self::prefix_documents_cache::PrefixDocumentsCache;
+pub use self::prefix_postings_lists_cache::PrefixPostingsListsCache;
+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::{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 BEU32 = zerocopy::U32<byteorder::BigEndian>;
+type BEU64 = zerocopy::U64<byteorder::BigEndian>;
+pub type BEU16 = zerocopy::U16<byteorder::BigEndian>;
+
+#[derive(Debug, Copy, Clone, AsBytes, FromBytes)]
+#[repr(C)]
+pub struct DocumentFieldIndexedKey {
+    docid: BEU32,
+    indexed_pos: BEU16,
+}
+
+impl DocumentFieldIndexedKey {
+    fn new(docid: DocumentId, indexed_pos: IndexedPos) -> DocumentFieldIndexedKey {
+        DocumentFieldIndexedKey {
+            docid: BEU32::new(docid.0),
+            indexed_pos: BEU16::new(indexed_pos.0),
+        }
+    }
+}
+
+#[derive(Debug, Copy, Clone, AsBytes, FromBytes)]
+#[repr(C)]
+pub struct DocumentFieldStoredKey {
+    docid: BEU32,
+    field_id: BEU16,
+}
+
+impl DocumentFieldStoredKey {
+    fn new(docid: DocumentId, field_id: FieldId) -> DocumentFieldStoredKey {
+        DocumentFieldStoredKey {
+            docid: BEU32::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() as u64;
+        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)
+}
+
+fn facets_name(name: &str) -> String {
+    format!("store-{}-facets", name)
+}
+
+#[derive(Clone)]
+pub struct Index {
+    pub main: Main,
+    pub postings_lists: PostingsLists,
+    pub documents_fields: DocumentsFields,
+    pub documents_fields_counts: DocumentsFieldsCounts,
+    pub facets: Facets,
+    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>) -> MResult<u64> {
+        let _ = self.updates_notifier.send(UpdateEvent::NewUpdate);
+        Ok(update::push_customs_update(writer, self.updates, self.updates_results, customs)?)
+    }
+
+    pub fn settings_update(&self, writer: &mut heed::RwTxn<UpdateT>, update: SettingsUpdate) -> MResult<u64> {
+        let _ = self.updates_notifier.send(UpdateEvent::NewUpdate);
+        Ok(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)
+    }
+
+    pub fn query_builder_with_criteria<'c, 'f, 'd, 'i>(
+        &'i self,
+        criteria: Criteria<'c>,
+    ) -> QueryBuilder<'c, 'f, 'd, 'i> {
+        QueryBuilder::with_criteria(self, 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);
+    let facets_name = facets_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 facets = env.create_database(Some(&facets_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 },
+        facets: Facets { facets },
+
+        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 facets_name = facets_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 facets = match env.open_database(Some(&facets_name))? {
+        Some(facets) => facets,
+        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 },
+        facets: Facets { facets },
+        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, BEU32};
+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: BEU32,
+}
+
+impl PrefixKey {
+    pub fn new(prefix: [u8; 4], index: u64, docid: u32) -> PrefixKey {
+        PrefixKey {
+            prefix,
+            index: BEU64::new(index),
+            docid: BEU32::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(), u32::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 +1,44 @@
+use std::borrow::Cow;
+
+use heed::Result as ZResult;
+use heed::types::ByteSlice;
+
+use crate::database::MainT;
+use crate::{FstSetCow, MResult};
+
+#[derive(Copy, Clone)]
+pub struct Synonyms {
+    pub(crate) synonyms: heed::Database<ByteSlice, ByteSlice>,
+}
+
+impl Synonyms {
+    pub fn put_synonyms<A>(self, writer: &mut heed::RwTxn<MainT>, word: &[u8], synonyms: &fst::Set<A>) -> ZResult<()>
+    where A: AsRef<[u8]>,
+    {
+        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(crate) fn synonyms_fst<'txn>(self, reader: &'txn heed::RoTxn<MainT>, word: &[u8]) -> ZResult<FstSetCow<'txn>> {
+        match self.synonyms.get(reader, word)? {
+            Some(bytes) => Ok(fst::Set::new(bytes).unwrap().map_data(Cow::Borrowed).unwrap()),
+            None => Ok(fst::Set::default().map_data(Cow::Owned).unwrap()),
+        }
+    }
+
+    pub fn synonyms(self, reader: &heed::RoTxn<MainT>, word: &[u8]) -> MResult<Vec<String>> {
+        let synonyms = self
+            .synonyms_fst(&reader, word)?
+            .stream()
+            .into_strs()?;
+        Ok(synonyms)
+    }
+}

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 +1,34 @@
+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_external_docids(writer, &fst::Map::default())?;
+    index.main.put_internal_docids(writer, &sdset::SetBuf::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 +1,26 @@
+
+use crate::database::{MainT, UpdateT};
+use crate::{store, MResult};
+use crate::update::{next_update_id, Update};
+
+pub fn apply_customs_update(
+    writer: &mut heed::RwTxn<MainT>,
+    main_store: store::Main,
+    customs: &[u8],
+) -> MResult<()> {
+    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>,
+) -> MResult<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 +1,426 @@
+use std::borrow::Cow;
+use std::collections::{HashMap, BTreeMap};
+
+use fst::{set::OpBuilder, SetBuilder};
+use indexmap::IndexMap;
+use meilisearch_schema::{Schema, FieldId};
+use meilisearch_types::DocumentId;
+use sdset::{duo::Union, SetOperation};
+use serde::Deserialize;
+use serde_json::Value;
+
+use crate::database::{MainT, UpdateT};
+use crate::database::{UpdateEvent, UpdateEventsEmitter};
+use crate::facets;
+use crate::raw_indexer::RawIndexer;
+use crate::serde::Deserializer;
+use crate::store::{self, DocumentsFields, DocumentsFieldsCounts, DiscoverIds};
+use crate::update::helpers::{index_value, value_to_number, extract_document_id};
+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)
+}
+
+#[allow(clippy::too_many_arguments)]
+fn index_document<A>(
+    writer: &mut heed::RwTxn<MainT>,
+    documents_fields: DocumentsFields,
+    documents_fields_counts: DocumentsFieldsCounts,
+    ranked_map: &mut RankedMap,
+    indexer: &mut RawIndexer<A>,
+    schema: &Schema,
+    field_id: FieldId,
+    document_id: DocumentId,
+    value: &Value,
+) -> MResult<()>
+where A: AsRef<[u8]>,
+{
+    let serialized = serde_json::to_vec(value)?;
+    documents_fields.put_document_field(writer, document_id, field_id, &serialized)?;
+
+    if let Some(indexed_pos) = schema.is_indexed(field_id) {
+        let number_of_words = index_value(indexer, document_id, *indexed_pos, value);
+        if let Some(number_of_words) = number_of_words {
+            documents_fields_counts.put_document_field_count(
+                writer,
+                document_id,
+                *indexed_pos,
+                number_of_words as u16,
+            )?;
+        }
+    }
+
+    if schema.is_ranked(field_id) {
+        let number = value_to_number(value).unwrap_or_default();
+        ranked_map.insert(document_id, field_id, number);
+    }
+
+    Ok(())
+}
+
+pub fn apply_addition<'a, 'b>(
+    writer: &'a mut heed::RwTxn<'b, MainT>,
+    index: &store::Index,
+    new_documents: Vec<IndexMap<String, Value>>,
+    partial: bool
+) -> MResult<()>
+{
+    let mut schema = match index.main.schema(writer)? {
+        Some(schema) => schema,
+        None => return Err(Error::SchemaMissing),
+    };
+
+    // Retrieve the documents ids related structures
+    let external_docids = index.main.external_docids(writer)?;
+    let internal_docids = index.main.internal_docids(writer)?;
+    let mut available_ids = DiscoverIds::new(&internal_docids);
+
+    let primary_key = schema.primary_key().ok_or(Error::MissingPrimaryKey)?;
+
+    // 1. store documents ids for future deletion
+    let mut documents_additions = HashMap::new();
+    let mut new_external_docids = BTreeMap::new();
+    let mut new_internal_docids = Vec::with_capacity(new_documents.len());
+
+    for mut document in new_documents {
+        let external_docids_get = |docid: &str| {
+            match (external_docids.get(docid), new_external_docids.get(docid)) {
+                (_, Some(&id))
+                | (Some(id), _) => Some(id as u32),
+                (None, None) => None,
+            }
+        };
+
+        let (internal_docid, external_docid) =
+            extract_document_id(
+                &primary_key,
+                &document,
+                &external_docids_get,
+                &mut available_ids,
+            )?;
+
+        new_external_docids.insert(external_docid, internal_docid.0 as u64);
+        new_internal_docids.push(internal_docid);
+
+        if partial {
+            let mut deserializer = Deserializer {
+                document_id: internal_docid,
+                reader: writer,
+                documents_fields: index.documents_fields,
+                schema: &schema,
+                fields: None,
+            };
+
+            let old_document = Option::<HashMap<String, Value>>::deserialize(&mut deserializer)?;
+            if let Some(old_document) = old_document {
+                for (key, value) in old_document {
+                    document.entry(key).or_insert(value);
+                }
+            }
+        }
+        documents_additions.insert(internal_docid, document);
+    }
+
+    // 2. remove the documents postings lists
+    let number_of_inserted_documents = documents_additions.len();
+    let documents_ids = new_external_docids.iter().map(|(id, _)| id.clone()).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 = index.main.stop_words_fst(writer)?.map_data(Cow::into_owned)?;
+
+
+    let mut indexer = RawIndexer::new(stop_words);
+
+    // For each document in this update
+    for (document_id, document) in &documents_additions {
+        // For each key-value pair in the document.
+        for (attribute, value) in document {
+            let field_id = schema.insert_and_index(&attribute)?;
+            index_document(
+                writer,
+                index.documents_fields,
+                index.documents_fields_counts,
+                &mut ranked_map,
+                &mut indexer,
+                &schema,
+                field_id,
+                *document_id,
+                &value,
+            )?;
+        }
+    }
+
+    write_documents_addition_index(
+        writer,
+        index,
+        &ranked_map,
+        number_of_inserted_documents,
+        indexer,
+    )?;
+
+    index.main.put_schema(writer, &schema)?;
+
+    let new_external_docids = fst::Map::from_iter(new_external_docids.iter().map(|(ext, id)| (ext, *id as u64)))?;
+    let new_internal_docids = sdset::SetBuf::from_dirty(new_internal_docids);
+    index.main.merge_external_docids(writer, &new_external_docids)?;
+    index.main.merge_internal_docids(writer, &new_internal_docids)?;
+
+    // recompute all facet attributes after document update.
+    if let Some(attributes_for_facetting) = index.main.attributes_for_faceting(writer)? {
+        let docids = index.main.internal_docids(writer)?;
+        let facet_map = facets::facet_map_from_docids(writer, index, &docids, attributes_for_facetting.as_ref())?;
+        index.facets.add(writer, facet_map)?;
+    }
+
+    // update is finished; update sorted document id cache with new state
+    let mut document_ids = index.main.internal_docids(writer)?.to_vec();
+    super::cache_document_ids_sorted(writer, &ranked_map, index, &mut document_ids)?;
+
+    Ok(())
+}
+
+pub fn apply_documents_partial_addition<'a, 'b>(
+    writer: &'a mut heed::RwTxn<'b, MainT>,
+    index: &store::Index,
+    new_documents: Vec<IndexMap<String, Value>>,
+) -> MResult<()> {
+    apply_addition(writer, index, new_documents, true)
+}
+
+pub fn apply_documents_addition<'a, 'b>(
+    writer: &'a mut heed::RwTxn<'b, MainT>,
+    index: &store::Index,
+    new_documents: Vec<IndexMap<String, Value>>,
+) -> MResult<()> {
+    apply_addition(writer, index, new_documents, false)
+}
+
+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.facets.clear(writer)?;
+    index.postings_lists.clear(writer)?;
+    index.docs_words.clear(writer)?;
+
+    let stop_words = index.main
+        .stop_words_fst(writer)?
+        .map_data(Cow::into_owned)
+        .unwrap();
+
+    let number_of_inserted_documents = documents_ids_to_reindex.len();
+    let mut indexer = RawIndexer::new(stop_words);
+    let mut ram_store = HashMap::new();
+
+    if let Some(ref attributes_for_facetting) = index.main.attributes_for_faceting(writer)? {
+        let facet_map = facets::facet_map_from_docids(writer, &index, &documents_ids_to_reindex, &attributes_for_facetting)?;
+        index.facets.add(writer, facet_map)?;
+    }
+    // ^-- https://github.com/meilisearch/MeiliSearch/pull/631#issuecomment-626624470 --v
+    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: Value = serde_json::from_slice(bytes)?;
+            ram_store.insert((document_id, field_id), value);
+        }
+
+        // For each key-value pair in the document.
+        for ((document_id, field_id), value) in ram_store.drain() {
+            index_document(
+                writer,
+                index.documents_fields,
+                index.documents_fields_counts,
+                &mut ranked_map,
+                &mut indexer,
+                &schema,
+                field_id,
+                *document_id,
+                &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)?;
+
+    // recompute all facet attributes after document update.
+    if let Some(attributes_for_facetting) = index.main.attributes_for_faceting(writer)? {
+        let docids = index.main.internal_docids(writer)?;
+        let facet_map = facets::facet_map_from_docids(writer, index, &docids, attributes_for_facetting.as_ref())?;
+        index.facets.add(writer, facet_map)?;
+    }
+
+    // update is finished; update sorted document id cache with new state
+    let mut document_ids = index.main.internal_docids(writer)?.to_vec();
+    super::cache_document_ids_sorted(writer, &ranked_map, index, &mut document_ids)?;
+
+    Ok(())
+}
+
+pub fn write_documents_addition_index<A>(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    ranked_map: &RankedMap,
+    number_of_inserted_documents: usize,
+    indexer: RawIndexer<A>,
+) -> MResult<()>
+where A: AsRef<[u8]>,
+{
+    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_set();
+
+    let words_fst = index.main.words_fst(writer)?;
+    let words = if !words_fst.is_empty() {
+        let op = OpBuilder::new()
+            .add(words_fst.stream())
+            .add(delta_words.stream())
+            .r#union();
+
+        let mut words_builder = SetBuilder::memory();
+        words_builder.extend_stream(op).unwrap();
+        words_builder.into_set()
+    } else {
+        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, &words, index)?;
+
+    Ok(())
+}

meilisearch-core/src/update/documents_deletion.rs

@@ -0,0 +1,207 @@
+use std::collections::{BTreeSet, HashMap, HashSet};
+
+use fst::{SetBuilder, Streamer};
+use sdset::{duo::DifferenceByKey, SetBuf, SetOperation};
+
+use crate::database::{MainT, UpdateT};
+use crate::database::{UpdateEvent, UpdateEventsEmitter};
+use crate::facets;
+use crate::store;
+use crate::update::{next_update_id, compute_short_prefixes, Update};
+use crate::{DocumentId, Error, MResult, RankedMap, MainWriter, Index};
+
+pub struct DocumentsDeletion {
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    updates_notifier: UpdateEventsEmitter,
+    external_docids: Vec<String>,
+}
+
+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,
+            external_docids: Vec::new(),
+        }
+    }
+
+    pub fn delete_document_by_external_docid(&mut self, document_id: String) {
+        self.external_docids.push(document_id);
+    }
+
+    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.external_docids,
+        )?;
+        Ok(update_id)
+    }
+}
+
+impl Extend<String> for DocumentsDeletion {
+    fn extend<T: IntoIterator<Item=String>>(&mut self, iter: T) {
+        self.external_docids.extend(iter)
+    }
+}
+
+pub fn push_documents_deletion(
+    writer: &mut heed::RwTxn<UpdateT>,
+    updates_store: store::Updates,
+    updates_results_store: store::UpdatesResults,
+    external_docids: Vec<String>,
+) -> MResult<u64> {
+    let last_update_id = next_update_id(writer, updates_store, updates_results_store)?;
+
+    let update = Update::documents_deletion(external_docids);
+    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,
+    external_docids: Vec<String>,
+) -> MResult<()>
+{
+    let (external_docids, internal_docids) = {
+        let new_external_docids = SetBuf::from_dirty(external_docids);
+        let mut internal_docids = Vec::new();
+
+        let old_external_docids = index.main.external_docids(writer)?;
+        for external_docid in new_external_docids.as_slice() {
+            if let Some(id) = old_external_docids.get(external_docid) {
+                internal_docids.push(DocumentId(id as u32));
+            }
+        }
+
+        let new_external_docids = fst::Map::from_iter(new_external_docids.into_iter().map(|k| (k, 0))).unwrap();
+        (new_external_docids, SetBuf::from_dirty(internal_docids))
+    };
+
+    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(),
+    };
+
+    // facet filters deletion
+    if let Some(attributes_for_facetting) = index.main.attributes_for_faceting(writer)? {
+        let facet_map = facets::facet_map_from_docids(writer, &index, &internal_docids, &attributes_for_facetting)?;
+        index.facets.remove(writer, facet_map)?;
+    }
+
+    // collect the ranked attributes according to the schema
+    let ranked_fields = schema.ranked();
+
+    let mut words_document_ids = HashMap::new();
+    for id in internal_docids.iter().cloned() {
+        // remove all the ranked attributes from the ranked_map
+        for ranked_attr in ranked_fields {
+            ranked_map.remove(id, *ranked_attr);
+        }
+
+        let words = index.docs_words.doc_words(writer, id)?;
+        if !words.is_empty() {
+            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 = {
+        let words_set = index.main.words_fst(writer)?;
+        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_set()
+    };
+
+    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)?;
+
+    // We apply the changes to the user and internal ids
+    index.main.remove_external_docids(writer, &external_docids)?;
+    index.main.remove_internal_docids(writer, &internal_docids)?;
+
+    compute_short_prefixes(writer, &words, index)?;
+
+    // update is finished; update sorted document id cache with new state
+    document_cache_remove_deleted(writer, index, &ranked_map, &deleted_documents)?;
+
+    Ok(())
+}
+
+/// rebuilds the document id cache by either removing deleted documents from the existing cache,
+/// and generating a new one from docs in store
+fn document_cache_remove_deleted(writer: &mut MainWriter, index: &Index, ranked_map: &RankedMap, documents_to_delete: &HashSet<DocumentId>) -> MResult<()> {
+    let new_cache = match index.main.sorted_document_ids_cache(writer)? {
+        // only keep documents that are not in the list of deleted documents. Order is preserved,
+        // no need to resort
+        Some(old_cache) => {
+            old_cache.iter().filter(|docid| !documents_to_delete.contains(docid)).cloned().collect::<Vec<_>>()
+        }
+        // couldn't find cached documents, try building a new cache from documents in store
+        None => {
+            let mut document_ids = index.main.internal_docids(writer)?.to_vec();
+            super::cache_document_ids_sorted(writer, ranked_map, index, &mut document_ids)?;
+            document_ids
+        }
+    };
+    index.main.put_sorted_document_ids_cache(writer, &new_cache)?;
+    Ok(())
+}

meilisearch-core/src/update/helpers.rs

@@ -0,0 +1,143 @@
+use std::fmt::Write as _;
+
+use indexmap::IndexMap;
+use meilisearch_schema::IndexedPos;
+use meilisearch_types::DocumentId;
+use ordered_float::OrderedFloat;
+use serde_json::Value;
+
+use crate::Number;
+use crate::raw_indexer::RawIndexer;
+use crate::serde::SerializerError;
+use crate::store::DiscoverIds;
+
+/// Returns the number of words indexed or `None` if the type is unindexable.
+pub fn index_value<A>(
+    indexer: &mut RawIndexer<A>,
+    document_id: DocumentId,
+    indexed_pos: IndexedPos,
+    value: &Value,
+) -> Option<usize>
+where A: AsRef<[u8]>,
+{
+    match value {
+        Value::Null => None,
+        Value::Bool(boolean) => {
+            let text = boolean.to_string();
+            let number_of_words = indexer.index_text(document_id, indexed_pos, &text);
+            Some(number_of_words)
+        },
+        Value::Number(number) => {
+            let text = number.to_string();
+            Some(indexer.index_text(document_id, indexed_pos, &text))
+        },
+        Value::String(string) => {
+            Some(indexer.index_text(document_id, indexed_pos, &string))
+        },
+        Value::Array(_) => {
+            let text = value_to_string(value);
+            Some(indexer.index_text(document_id, indexed_pos, &text))
+        },
+        Value::Object(_) => {
+            let text = value_to_string(value);
+            Some(indexer.index_text(document_id, indexed_pos, &text))
+        },
+    }
+}
+
+/// Transforms the JSON Value type into a String.
+pub fn value_to_string(value: &Value) -> String {
+    fn internal_value_to_string(string: &mut String, value: &Value) {
+        match value {
+            Value::Null => (),
+            Value::Bool(boolean) => { let _ = write!(string, "{}", &boolean); },
+            Value::Number(number) => { let _ = write!(string, "{}", &number); },
+            Value::String(text) => string.push_str(&text),
+            Value::Array(array) => {
+                for value in array {
+                    internal_value_to_string(string, value);
+                    let _ = string.write_str(". ");
+                }
+            },
+            Value::Object(object) => {
+                for (key, value) in object {
+                    string.push_str(key);
+                    let _ = string.write_str(". ");
+                    internal_value_to_string(string, value);
+                    let _ = string.write_str(". ");
+                }
+            },
+        }
+    }
+
+    let mut string = String::new();
+    internal_value_to_string(&mut string, value);
+    string
+}
+
+/// Transforms the JSON Value type into a Number.
+pub fn value_to_number(value: &Value) -> Option<Number> {
+    use std::str::FromStr;
+
+    match value {
+        Value::Null => None,
+        Value::Bool(boolean) => Some(Number::Unsigned(*boolean as u64)),
+        Value::Number(number) => {
+            match (number.as_i64(), number.as_u64(), number.as_f64()) {
+                (Some(n), _, _) => Some(Number::Signed(n)),
+                (_, Some(n), _) => Some(Number::Unsigned(n)),
+                (_, _, Some(n)) => Some(Number::Float(OrderedFloat(n))),
+                (None, None, None) => None,
+            }
+        },
+        Value::String(string) => Number::from_str(string).ok(),
+        Value::Array(_array) => None,
+        Value::Object(_object) => None,
+    }
+}
+
+/// Validates a string representation to be a correct document id and returns
+/// the corresponding id or generate a new one, this is the way we produce documents ids.
+pub fn discover_document_id<F>(
+    docid: &str,
+    external_docids_get: F,
+    available_docids: &mut DiscoverIds<'_>,
+) -> Result<DocumentId, SerializerError>
+where
+    F: FnOnce(&str) -> Option<u32>
+{
+    if docid.chars().all(|x| x.is_ascii_alphanumeric() || x == '-' || x == '_') {
+        match external_docids_get(docid) {
+            Some(id) => Ok(DocumentId(id)),
+            None => {
+                let internal_id = available_docids.next().expect("no more ids available");
+                Ok(internal_id)
+            },
+        }
+    } else {
+        Err(SerializerError::InvalidDocumentIdFormat)
+    }
+}
+
+/// Extracts and validates the document id of a document.
+pub fn extract_document_id<F>(
+    primary_key: &str,
+    document: &IndexMap<String, Value>,
+    external_docids_get: F,
+    available_docids: &mut DiscoverIds<'_>,
+) -> Result<(DocumentId, String), SerializerError>
+where
+    F: FnOnce(&str) -> Option<u32>
+{
+    match document.get(primary_key) {
+        Some(value) => {
+            let docid = match value {
+                Value::Number(number) => number.to_string(),
+                Value::String(string) => string.clone(),
+                _ => return Err(SerializerError::InvalidDocumentIdFormat),
+            };
+            discover_document_id(&docid, external_docids_get, available_docids).map(|id| (id, docid))
+        }
+        None => Err(SerializerError::DocumentIdNotFound),
+    }
+}

meilisearch-core/src/update/mod.rs

@@ -0,0 +1,384 @@
+mod clear_all;
+mod customs_update;
+mod documents_addition;
+mod documents_deletion;
+mod settings_update;
+mod helpers;
+
+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::helpers::{index_value, value_to_string, value_to_number, discover_document_id, extract_document_id};
+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 serde_json::Value;
+
+use meilisearch_error::ErrorCode;
+use meilisearch_types::DocumentId;
+
+use crate::{store, MResult, RankedMap};
+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(documents: Vec<IndexMap<String, Value>>) -> Update {
+        Update {
+            data: UpdateData::DocumentsAddition(documents),
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn documents_partial(documents: Vec<IndexMap<String, Value>>) -> Update {
+        Update {
+            data: UpdateData::DocumentsPartial(documents),
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn documents_deletion(data: Vec<String>) -> Update {
+        Update {
+            data: UpdateData::DocumentsDeletion(data),
+            enqueued_at: Utc::now(),
+        }
+    }
+
+    fn settings(data: SettingsUpdate) -> Update {
+        Update {
+            data: UpdateData::Settings(Box::new(data)),
+            enqueued_at: Utc::now(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub enum UpdateData {
+    ClearAll,
+    Customs(Vec<u8>),
+    DocumentsAddition(Vec<IndexMap<String, Value>>),
+    DocumentsPartial(Vec<IndexMap<String, Value>>),
+    DocumentsDeletion(Vec<String>),
+    Settings(Box<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: Box<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>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub error_type: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub error_code: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub error_link: 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.as_ref().map_err(|e| e.to_string()).err(),
+        error_code: result.as_ref().map_err(|e| e.error_name()).err(),
+        error_type: result.as_ref().map_err(|e| e.error_type()).err(),
+        error_link: result.as_ref().map_err(|e| e.error_url()).err(),
+        duration: duration.as_secs_f64(),
+        enqueued_at,
+        processed_at: Utc::now(),
+    };
+
+    Ok(status)
+}
+
+fn compute_short_prefixes<A>(
+    writer: &mut heed::RwTxn<MainT>,
+    words_fst: &fst::Set<A>,
+    index: &store::Index,
+) -> MResult<()>
+where A: AsRef<[u8]>,
+{
+    // 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(())
+}
+
+fn cache_document_ids_sorted(
+    writer: &mut heed::RwTxn<MainT>,
+    ranked_map: &RankedMap,
+    index: &store::Index,
+    document_ids: &mut [DocumentId],
+) -> MResult<()> {
+    crate::bucket_sort::placeholder_document_sort(document_ids, index, writer, ranked_map)?;
+    index.main.put_sorted_document_ids_cache(writer, &document_ids)
+}

meilisearch-core/src/update/settings_update.rs

@@ -0,0 +1,313 @@
+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)?;
+            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) => {
+            let field_id = schema.insert(&v)?;
+            index.main.put_distinct_attribute(writer, field_id)?;
+        },
+        UpdateState::Clear => {
+            index.main.delete_distinct_attribute(writer)?;
+        },
+        UpdateState::Nothing => (),
+    }
+
+    match settings.searchable_attributes.clone() {
+        UpdateState::Update(v) => {
+            if v.iter().any(|e| e == "*") || v.is_empty() {
+                schema.set_all_fields_as_indexed();
+            } else {
+                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) => {
+            if v.contains("*") || v.is_empty() {
+                schema.set_all_fields_as_displayed();
+            } else {
+                schema.update_displayed(v)?
+            }
+        },
+        UpdateState::Clear => {
+            schema.set_all_fields_as_displayed();
+        },
+        UpdateState::Nothing => (),
+    }
+
+    match settings.attributes_for_faceting {
+        UpdateState::Update(attrs) => {
+            apply_attributes_for_faceting_update(writer, index, &mut schema, &attrs)?;
+            must_reindex = true;
+        },
+        UpdateState::Clear => {
+            index.main.delete_attributes_for_faceting(writer)?;
+            index.facets.clear(writer)?;
+        },
+        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(())
+}
+
+fn apply_attributes_for_faceting_update(
+    writer: &mut heed::RwTxn<MainT>,
+    index: &store::Index,
+    schema: &mut Schema,
+    attributes: &[String]
+    ) -> MResult<()> {
+    let mut attribute_ids = Vec::new();
+    for name in attributes {
+        attribute_ids.push(schema.insert(name)?);
+    }
+    let attributes_for_faceting = SetBuf::from_dirty(attribute_ids);
+    index.main.put_attributes_for_faceting(writer, &attributes_for_faceting)?;
+    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)?
+        .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)?;
+    }
+
+    let words_fst = index.main.words_fst(writer)?;
+    if !words_fst.is_empty() {
+        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_set();
+
+        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_set();
+
+    // we also need to remove all the stop words from the main fst
+    let words_fst = main_store.words_fst(writer)?;
+    if !words_fst.is_empty() {
+        let op = OpBuilder::new()
+            .add(&words_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_set();
+
+        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)?;
+
+    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_set();
+
+    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_set();
+
+    // now we delete all of these stop words from the main store
+    let stop_words_fst = index.main.stop_words_fst(writer)?;
+
+    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_set();
+
+    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)?;
+            alternatives_builder.into_set()
+        };
+
+        synonyms_store.put_synonyms(writer, word.as_bytes(), &alternatives)?;
+    }
+
+    let synonyms_set = synonyms_builder.into_set();
+
+    main_store.put_synonyms_fst(writer, &synonyms_set)?;
+
+    Ok(())
+}

meilisearch-error/Cargo.toml

@@ -0,0 +1,8 @@
+[package]
+name = "meilisearch-error"
+version = "0.14.1"
+authors = ["marin <postma.marin@protonmail.com>"]
+edition = "2018"
+
+[dependencies]
+actix-http = "1.0.1"

meilisearch-error/src/lib.rs

@@ -0,0 +1,180 @@
+use std::fmt;
+
+use actix_http::http::StatusCode;
+
+pub trait ErrorCode: std::error::Error {
+    fn error_code(&self) -> Code;
+
+    /// returns the HTTP status code ascociated with the error
+    fn http_status(&self) -> StatusCode {
+        self.error_code().http()
+    }
+
+    /// returns the doc url ascociated with the error
+    fn error_url(&self) -> String {
+        self.error_code().url()
+    }
+
+    /// returns error name, used as error code
+    fn error_name(&self) -> String {
+        self.error_code().name()
+    }
+
+    /// return the error type
+    fn error_type(&self) -> String {
+        self.error_code().type_()
+    }
+}
+
+#[allow(clippy::enum_variant_names)]
+enum ErrorType {
+    InternalError,
+    InvalidRequestError,
+    AuthenticationError,
+}
+
+impl fmt::Display for ErrorType {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        use ErrorType::*;
+
+        match self {
+            InternalError => write!(f, "internal_error"),
+            InvalidRequestError => write!(f, "invalid_request_error"),
+            AuthenticationError => write!(f, "authentication_error"),
+        }
+    }
+}
+
+pub enum Code {
+    // index related error
+    CreateIndex,
+    IndexAlreadyExists,
+    IndexNotFound,
+    InvalidIndexUid,
+    OpenIndex,
+
+    // invalid state error
+    InvalidState,
+    MissingPrimaryKey,
+    PrimaryKeyAlreadyPresent,
+
+    MaxFieldsLimitExceeded,
+    MissingDocumentId,
+
+    Facet,
+    Filter,
+
+    BadParameter,
+    BadRequest,
+    DocumentNotFound,
+    Internal,
+    InvalidToken,
+    Maintenance,
+    MissingAuthorizationHeader,
+    NotFound,
+    PayloadTooLarge,
+    RetrieveDocument,
+    SearchDocuments,
+    UnsupportedMediaType,
+}
+
+impl Code {
+
+    /// ascociate a `Code` variant to the actual ErrCode
+    fn err_code(&self) -> ErrCode {
+        use Code::*;
+
+        match self {
+            // index related errors
+            // create index is thrown on internal error while creating an index.
+            CreateIndex => ErrCode::internal("index_creation_failed", StatusCode::BAD_REQUEST),
+            IndexAlreadyExists => ErrCode::invalid("index_already_exists", StatusCode::BAD_REQUEST),
+            // thrown when requesting an unexisting index
+            IndexNotFound => ErrCode::invalid("index_not_found", StatusCode::NOT_FOUND),
+            InvalidIndexUid => ErrCode::invalid("invalid_index_uid", StatusCode::BAD_REQUEST),
+            OpenIndex => ErrCode::internal("index_not_accessible", StatusCode::INTERNAL_SERVER_ERROR),
+
+            // invalid state error
+            InvalidState => ErrCode::internal("invalid_state", StatusCode::INTERNAL_SERVER_ERROR),
+            // thrown when no primary key has been set
+            MissingPrimaryKey => ErrCode::invalid("missing_primary_key", StatusCode::BAD_REQUEST),
+            // error thrown when trying to set an already existing primary key
+            PrimaryKeyAlreadyPresent => ErrCode::invalid("primary_key_already_present", StatusCode::BAD_REQUEST),
+
+            // invalid document
+            MaxFieldsLimitExceeded => ErrCode::invalid("max_fields_limit_exceeded", StatusCode::BAD_REQUEST),
+            MissingDocumentId => ErrCode::invalid("missing_document_id", StatusCode::BAD_REQUEST),
+
+            // error related to facets
+            Facet => ErrCode::invalid("invalid_facet", StatusCode::BAD_REQUEST),
+            // error related to filters
+            Filter => ErrCode::invalid("invalid_filter", StatusCode::BAD_REQUEST),
+
+            BadParameter => ErrCode::invalid("bad_parameter", StatusCode::BAD_REQUEST),
+            BadRequest => ErrCode::invalid("bad_request", StatusCode::BAD_REQUEST),
+            DocumentNotFound => ErrCode::invalid("document_not_found", StatusCode::NOT_FOUND),
+            Internal => ErrCode::internal("internal", StatusCode::INTERNAL_SERVER_ERROR),
+            InvalidToken => ErrCode::authentication("invalid_token", StatusCode::FORBIDDEN),
+            Maintenance =>  ErrCode::internal("maintenance", StatusCode::SERVICE_UNAVAILABLE),
+            MissingAuthorizationHeader => ErrCode::authentication("missing_authorization_header", StatusCode::UNAUTHORIZED),
+            NotFound => ErrCode::invalid("not_found", StatusCode::NOT_FOUND),
+            PayloadTooLarge => ErrCode::invalid("payload_too_large", StatusCode::PAYLOAD_TOO_LARGE),
+            RetrieveDocument => ErrCode::internal("unretrievable_document", StatusCode::BAD_REQUEST),
+            SearchDocuments => ErrCode::internal("search_error", StatusCode::BAD_REQUEST),
+            UnsupportedMediaType => ErrCode::invalid("unsupported_media_type", StatusCode::UNSUPPORTED_MEDIA_TYPE),
+        }
+    }
+
+    /// return the HTTP status code ascociated with the `Code`
+    fn http(&self) -> StatusCode {
+        self.err_code().status_code
+    }
+
+    /// return error name, used as error code
+    fn name(&self) -> String {
+        self.err_code().error_name.to_string()
+    }
+
+    /// return the error type
+    fn type_(&self) -> String {
+        self.err_code().error_type.to_string()
+    }
+
+    /// return the doc url ascociated with the error
+    fn url(&self) -> String {
+        format!("https://docs.meilisearch.com/errors#{}", self.name())
+    }
+}
+
+/// Internal structure providing a convenient way to create error codes
+struct ErrCode {
+    status_code: StatusCode,
+    error_type: ErrorType,
+    error_name: &'static str,
+}
+
+impl ErrCode {
+    fn authentication(error_name: &'static str, status_code: StatusCode) -> ErrCode {
+        ErrCode {
+            status_code,
+            error_name,
+            error_type: ErrorType::AuthenticationError,
+        }
+    }
+
+    fn internal(error_name: &'static str, status_code: StatusCode) -> ErrCode {
+        ErrCode {
+            status_code,
+            error_name,
+            error_type: ErrorType::InternalError,
+        }
+    }
+
+    fn invalid(error_name: &'static str, status_code: StatusCode) -> ErrCode {
+        ErrCode {
+            status_code,
+            error_name,
+            error_type: ErrorType::InvalidRequestError,
+        }
+    }
+}

meilisearch-http/Cargo.toml

@@ -0,0 +1,86 @@
+[package]
+name = "meilisearch-http"
+description = "MeiliSearch HTTP server"
+version = "0.14.1"
+license = "MIT"
+authors = [
+    "Quentin de Quelen <quentin@dequelen.me>",
+    "Clément Renault <clement@meilisearch.com>",
+]
+edition = "2018"
+
+[[bin]]
+name = "meilisearch"
+path = "src/main.rs"
+
+[features]
+default = ["sentry"]
+
+[dependencies]
+actix-cors = "0.2.0"
+actix-http = "1"
+actix-rt = "1"
+actix-service = "1.0.5"
+actix-web = { version = "2.0.0", features = ["rustls"] }
+actix-web-macros = "0.1.0"
+bytes = "0.5.4"
+chrono = { version = "0.4.11", features = ["serde"] }
+crossbeam-channel = "0.4.2"
+env_logger = "0.7.1"
+flate2 = "1.0.16"
+futures = "0.3.4"
+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 = "0.14.1" }
+meilisearch-error = { path = "../meilisearch-error", version = "0.14.1" }
+meilisearch-schema = { path = "../meilisearch-schema", version = "0.14.1" }
+meilisearch-tokenizer = {path = "../meilisearch-tokenizer", version = "0.14.1"}
+mime = "0.3.16"
+rand = "0.7.3"
+regex = "1.3.6"
+rustls = "0.16.0"
+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"
+slice-group-by = "0.2.6"
+structopt = "0.3.12"
+tar = "0.4.29"
+tempfile = "3.1.0"
+tokio = { version = "0.2.18", features = ["macros"] }
+ureq = { version = "0.12.0", features = ["tls"], default-features = false }
+walkdir = "2.3.1"
+whoami = "0.8.1"
+
+[dependencies.sentry]
+version = "0.18.1"
+default-features = false
+features = [
+    "with_client_implementation",
+    "with_panic",
+    "with_failure",
+    "with_device_info",
+    "with_rust_info",
+    "with_reqwest_transport",
+    "with_rustls",
+    "with_env_logger"
+]
+optional = true
+
+[dev-dependencies]
+serde_url_params = "0.2.0"
+tempdir = "0.3.7"
+tokio = { version = "0.2.18", features = ["macros", "time"] }
+
+[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 +1,308 @@
+<!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 class="field">
+            <!-- API Key -->
+            <div class="field">
+              <div class="control">
+                <input id="apiKey" class="input is-small" type="password" placeholder="API key (optional)">
+                <div class="help">At least a private API key is required for the dashboard to access the indexes list.</div>
+              </div>
+            </div>
+          </div>
+
+        </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">0</p>
+                </div>
+              </div>
+              <div class="level-item has-text-centered">
+                <div>
+                  <p class="heading">Time Spent</p>
+                  <p id="time" class="title">N/A</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, apiKey) {
+        var xmlHttp = new XMLHttpRequest();
+        xmlHttp.open("GET", theUrl, false); // false for synchronous request
+        if (apiKey) {
+          xmlHttp.setRequestHeader("x-Meili-API-Key", apiKey);
+        }
+        xmlHttp.send(null);
+        return xmlHttp.responseText;
+    }
+
+    function refreshIndexList() {
+        // TODO we must not block here
+        let result = JSON.parse(httpGet(`${baseUrl}/indexes`, localStorage.getItem('apiKey')));
+
+        if (!Array.isArray(result)) { return }
+
+        let select = document.getElementById("index");
+        select.innerHTML = '';
+
+        for (index of result) {
+            const option = document.createElement('option');
+            option.value = index.uid;
+            option.innerHTML = index.name;
+            select.appendChild(option);
+        }
+    }
+
+    let lastRequest = undefined;
+
+    function triggerSearch() {
+        var e = document.getElementById("index");
+        if (e.selectedIndex == -1) { return }
+        var index = e.options[e.selectedIndex].value;
+
+        let theUrl = `${baseUrl}/indexes/${index}/search?q=${search.value}&attributesToHighlight=*`;
+
+        if (lastRequest) { lastRequest.abort() }
+        lastRequest = new XMLHttpRequest();
+
+        lastRequest.open("GET", theUrl, true);
+
+        if (localStorage.getItem('apiKey')) {
+          lastRequest.setRequestHeader("x-Meili-API-Key", localStorage.getItem('apiKey'));
+        }
+
+        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.nbHits;
+                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);
+    }
+
+    if (!apiKey.value) {
+        apiKey.value = localStorage.getItem('apiKey');
+    }
+
+    apiKey.addEventListener('input', function(e) {
+        localStorage.setItem('apiKey', apiKey.value);
+        refreshIndexList();
+    }, false);
+
+    let baseUrl = window.location.origin;
+    refreshIndexList();
+
+    search.oninput = triggerSearch;
+    
+    let select = document.getElementById("index");
+    select.onchange = triggerSearch;
+
+    triggerSearch();
+  </script>
+</html>