[WIP] Fix phrase search containing stop words

Fixes #661 and meilisearch/meilisearch#2905

milli/src/search/criteria/attribute.rs

@@ -579,6 +579,7 @@ fn flatten_query_tree(query_tree: &Operation) -> FlattenedQueryTree {
             Phrase(words) => {
                 let queries = words
                     .iter()
+                    .filter_map(|w| w.as_ref())
                     .map(|word| vec![Query { prefix: false, kind: QueryKind::exact(word.clone()) }])
                     .collect();
                 vec![queries]

milli/src/search/criteria/exactness.rs

@@ -298,7 +298,7 @@ fn attribute_start_with_docids(
                 pos += 1;
             }
             Phrase(phrase) => {
-                for word in phrase {
+                for word in phrase.iter().filter_map(|w| w.as_ref()) {
                     let wc = ctx.word_position_docids(word, pos)?;
                     if let Some(word_candidates) = wc {
                         attribute_candidates_array.push(word_candidates);
@@ -323,7 +323,7 @@ fn intersection_of(mut rbs: Vec<&RoaringBitmap>) -> RoaringBitmap {
 
 #[derive(Debug, Clone)]
 pub enum ExactQueryPart {
-    Phrase(Vec<String>),
+    Phrase(Vec<Option<String>>),
     Synonyms(Vec<String>),
 }
 

milli/src/search/criteria/mod.rs

@@ -418,15 +418,21 @@ pub fn resolve_query_tree(
     resolve_operation(ctx, query_tree, wdcache)
 }
 
-pub fn resolve_phrase(ctx: &dyn Context, phrase: &[String]) -> Result<RoaringBitmap> {
+pub fn resolve_phrase(ctx: &dyn Context, phrase: &[Option<String>]) -> Result<RoaringBitmap> {
     let mut candidates = RoaringBitmap::new();
     let mut first_iter = true;
     let winsize = phrase.len().min(3);
     for win in phrase.windows(winsize) {
         // Get all the documents with the matching distance for each word pairs.
         let mut bitmaps = Vec::with_capacity(winsize.pow(2));
-        for (offset, s1) in win.iter().enumerate() {
-            for (dist, s2) in win.iter().skip(offset + 1).enumerate() {
+        for (offset, s1) in win.iter().filter_map(|w| w.as_ref()).enumerate() {
+            for (dist, s2) in win.iter().skip(offset + 1).enumerate().filter_map(|(index, word)| {
+                if let Some(word) = word {
+                    Some((index, word))
+                } else {
+                    None
+                }
+            }) {
                 if dist == 0 {
                     match ctx.word_pair_proximity_docids(s1, s2, 1)? {
                         Some(m) => bitmaps.push(m),

milli/src/search/criteria/proximity.rs

@@ -188,9 +188,13 @@ fn resolve_candidates<'t>(
                 if proximity == 0 {
                     let most_left = words
                         .first()
+                        .map(|o| o.as_ref())
+                        .flatten()
                         .map(|w| Query { prefix: false, kind: QueryKind::exact(w.clone()) });
                     let most_right = words
                         .last()
+                        .map(|o| o.as_ref())
+                        .flatten()
                         .map(|w| Query { prefix: false, kind: QueryKind::exact(w.clone()) });
 
                     match (most_left, most_right) {
@@ -473,7 +477,7 @@ fn resolve_plane_sweep_candidates(
             }
             Phrase(words) => {
                 let mut groups_positions = Vec::with_capacity(words.len());
-                for word in words {
+                for word in words.iter().filter_map(|w| w.as_ref()) {
                     let positions = match words_positions.get(word) {
                         Some(positions) => positions.iter().map(|p| (p, 0, p)).collect(),
                         None => return Ok(vec![]),

milli/src/search/criteria/typo.rs

@@ -2,6 +2,7 @@ use std::borrow::Cow;
 use std::collections::HashMap;
 use std::mem::take;
 
+use itertools::Itertools;
 use log::debug;
 use roaring::RoaringBitmap;
 
@@ -259,8 +260,7 @@ fn resolve_candidates<'t>(
             Phrase(words) => {
                 let mut candidates = RoaringBitmap::new();
                 let mut first_loop = true;
-                for slice in words.windows(2) {
-                    let (left, right) = (&slice[0], &slice[1]);
+                for (left, right) in words.iter().filter_map(|w| w.as_ref()).tuple_windows() {
                     match ctx.word_pair_proximity_docids(left, right, 1)? {
                         Some(pair_docids) => {
                             if pair_docids.is_empty() {