From 22b19c0d9316a93af04e274a015fade56943cd72 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9ment=20Renault?= <renault.cle@gmail.com>
Date: Sat, 7 Dec 2019 13:32:43 +0100
Subject: [PATCH] Fix the processed distance algorithm

---
 meilisearch-core/src/bucket_sort.rs | 24 ++++++++++++++++--------
 meilisearch-core/src/criterion2.rs  | 14 +++++++++++---
 2 files changed, 27 insertions(+), 11 deletions(-)

diff --git a/meilisearch-core/src/bucket_sort.rs b/meilisearch-core/src/bucket_sort.rs
index 7477ff383..303e94e50 100644
--- a/meilisearch-core/src/bucket_sort.rs
+++ b/meilisearch-core/src/bucket_sort.rs
@@ -1,4 +1,5 @@
 use std::ops::Deref;
+use std::fmt;
 use std::borrow::Cow;
 use std::cmp::Ordering;
 use std::collections::HashSet;
@@ -145,7 +146,6 @@ pub struct RawDocument<'a, 'tag> {
     pub raw_matches: &'a mut [BareMatch<'tag>],
     pub processed_matches: Vec<SimpleMatch>,
     /// The list of minimum `distance` found
-    /// where the `query_index` is the index
     pub processed_distances: Vec<Option<u8>>,
 }
 
@@ -157,6 +157,17 @@ pub struct BareMatch<'tag> {
     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()
+    }
+}
+
 // TODO remove that
 #[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
 pub struct SimpleMatch {
@@ -238,14 +249,11 @@ fn fetch_matches<'txn, 'tag>(
     for (query_index, automaton) in automatons.iter().enumerate() {
         let before_dfa = Instant::now();
         let dfa = automaton.dfa();
-        let QueryWordAutomaton { query, is_exact, is_prefix, .. } = automaton;
+        let QueryWordAutomaton { query, is_exact, is_prefix, phrase_query } = automaton;
         dfa_time += before_dfa.elapsed();
 
         let mut number_of_words = 0;
-
-        let before_fst_search = Instant::now();
         let mut stream = words.search(&dfa).into_stream();
-        debug!("fst search took {:.02?}", before_fst_search.elapsed());
 
         // while let Some(input) = stream.next() {
         loop {
@@ -272,7 +280,7 @@ fn fetch_matches<'txn, 'tag>(
 
                     let posting_list_index = arena.add(postings_list_view.range(offset, group.len()));
                     let document_id = group[0].document_id;
-                    let stuffed = BareMatch {
+                    let bare_match = BareMatch {
                         document_id,
                         query_index: query_index as u16,
                         distance,
@@ -280,7 +288,7 @@ fn fetch_matches<'txn, 'tag>(
                         postings_list: posting_list_index,
                     };
 
-                    total_postings_lists.push(stuffed);
+                    total_postings_lists.push(bare_match);
                     offset += group.len();
                 }
             }
@@ -434,7 +442,7 @@ fn construct_automatons2(
                 }
             }
 
-            if n == 1 {
+            if false && n == 1 {
                 if let Some((left, right)) = split_best_frequency(reader, &normalized, postings_lists_store)? {
                     let mut left_automaton = QueryWordAutomaton::exact(left);
                     left_automaton.phrase_query = Some((0, 2));
diff --git a/meilisearch-core/src/criterion2.rs b/meilisearch-core/src/criterion2.rs
index dd7769261..4c40b9969 100644
--- a/meilisearch-core/src/criterion2.rs
+++ b/meilisearch-core/src/criterion2.rs
@@ -46,14 +46,22 @@ pub trait Criterion {
 fn prepare_query_distances(
     documents: &mut [RawDocument],
     query_enhancer: &QueryEnhancer,
+    automatons: &[QueryWordAutomaton],
 ) {
     for document in documents {
         if !document.processed_distances.is_empty() { continue }
 
         let mut processed = Vec::new();
         for m in document.raw_matches.iter() {
+            // FIXME we really need to take splitted words into account
+            //       those must be seen at the same level as the non-splitteds
+            // if automatons[m.query_index as usize].phrase_query.is_some() {
+            //     continue
+            // }
+
             let range = query_enhancer.replacement(m.query_index as u32);
-            processed.resize(range.end as usize, None);
+            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;
@@ -81,7 +89,7 @@ impl Criterion for Typo {
         query_enhancer: &QueryEnhancer,
         automatons: &[QueryWordAutomaton],
     ) {
-        prepare_query_distances(documents, query_enhancer);
+        prepare_query_distances(documents, query_enhancer, automatons);
     }
 
     fn evaluate(
@@ -139,7 +147,7 @@ impl Criterion for Words {
         query_enhancer: &QueryEnhancer,
         automatons: &[QueryWordAutomaton],
     ) {
-        prepare_query_distances(documents, query_enhancer);
+        prepare_query_distances(documents, query_enhancer, automatons);
     }
 
     fn evaluate(