Merge pull request #13408 from vespa-engine/toregge/record-need-for-normal-term-field-match-data-for-phrase-splitter-input

Record need for normal term field match data for phrase splitter input.

3 files changed, 16 insertions, 10 deletions

diff --git a/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.cpp b/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.cpp
index 593b8fb29ce..eb2915c4e90 100644
--- a/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.cpp
+++ b/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.cpp
@@ -5,7 +5,7 @@
 namespace search::fef {
 
 void
-PhraseSplitterQueryEnv::considerTerm(uint32_t termIdx, const ITermData &term, std::vector<PhraseTerm> &phraseTerms, uint32_t fieldId)
+PhraseSplitterQueryEnv::considerTerm(uint32_t termIdx, const ITermData &term, uint32_t fieldId)
 {
     typedef search::fef::ITermFieldRangeAdapter FRA;
 
@@ -19,7 +19,7 @@ PhraseSplitterQueryEnv::considerTerm(uint32_t termIdx, const ITermData &term, st
                 prototype.setPhraseLength(1);
                 prototype.setUniqueId(term.getUniqueId());
                 prototype.addField(fieldId);
-                phraseTerms.push_back(PhraseTerm(term, _terms.size(), h));
+                _phrase_terms.push_back(PhraseTerm(term, _terms.size(), h));
                 for (uint32_t i = 0; i < term.getPhraseLength(); ++i) {
                     _terms.push_back(prototype);
                     _termIdxMap.push_back(TermIdx(_terms.size() - 1, true));
@@ -37,15 +37,14 @@ PhraseSplitterQueryEnv::PhraseSplitterQueryEnv(const IQueryEnvironment & queryEn
       _termIdxMap(),
       _maxHandle(0),
       _skipHandles(0),
-      _field_id(fieldId)
+      _field_id(fieldId),
+      _phrase_terms()
 {
     TermFieldHandle numHandles = 0; // how many handles existed in underlying data
-    std::vector<PhraseTerm> phraseTerms; // data about original phrase terms
-
     for (uint32_t i = 0; i < queryEnv.getNumTerms(); ++i) {
         const ITermData *td = queryEnv.getTerm(i);
         assert(td != nullptr);
-        considerTerm(i, *td, phraseTerms, fieldId);
+        considerTerm(i, *td, fieldId);
         numHandles += td->numFields();
     }
 
@@ -57,8 +56,8 @@ PhraseSplitterQueryEnv::PhraseSplitterQueryEnv(const IQueryEnvironment & queryEn
         ++term_handle;
     }
 
-    for (uint32_t i = 0; i < phraseTerms.size(); ++i) {
-        const PhraseTerm &pterm = phraseTerms[i];
+    for (uint32_t i = 0; i < _phrase_terms.size(); ++i) {
+        const PhraseTerm &pterm = _phrase_terms[i];
 
         for (uint32_t j = 0; j < pterm.term.getPhraseLength(); ++j) {
             const ITermData &splitp_td = _terms[pterm.idx + j];
diff --git a/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.h b/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.h
index 9bdac380f1a..b2a3d416f5a 100644
--- a/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.h
+++ b/searchlib/src/vespa/searchlib/fef/phrase_splitter_query_env.h
@@ -25,13 +25,13 @@ private:
         bool     splitted; // whether this term has been splitted or not
         TermIdx(uint32_t i, bool s) : idx(i), splitted(s) {}
     };
+public:
     struct PhraseTerm {
         const ITermData & term; // for original phrase
         uint32_t idx; // index into vector of our TermData objects
         TermFieldHandle orig_handle;
         PhraseTerm(const ITermData & t, uint32_t i, uint32_t h) : term(t), idx(i), orig_handle(h) {}
     };
-public:
     struct HowToCopy {
         TermFieldHandle orig_handle;
         TermFieldHandle split_handle;
@@ -45,8 +45,9 @@ private:
     TermFieldHandle                 _maxHandle;   // the largest among original term field handles
     TermFieldHandle                 _skipHandles;   // how many handles to skip
     uint32_t                        _field_id;
+    std::vector<PhraseTerm>         _phrase_terms; // data about original phrase terms
 
-    void considerTerm(uint32_t termIdx, const ITermData &term, std::vector<PhraseTerm> &phraseTerms, uint32_t fieldId);
+    void considerTerm(uint32_t termIdx, const ITermData &term, uint32_t fieldId);
 
 public:
     /**
@@ -82,6 +83,7 @@ public:
     uint32_t get_num_phrase_split_terms() const { return _terms.size(); }
     uint32_t get_field_id() const { return _field_id; }
     const std::vector<HowToCopy>& get_copy_info() const { return _copyInfo; }
+    const std::vector<PhraseTerm>& get_phrase_terms() const { return _phrase_terms; }
 };
 
 
diff --git a/searchlib/src/vespa/searchlib/fef/phrasesplitter.cpp b/searchlib/src/vespa/searchlib/fef/phrasesplitter.cpp
index b80a9c9e085..b07498e8608 100644
--- a/searchlib/src/vespa/searchlib/fef/phrasesplitter.cpp
+++ b/searchlib/src/vespa/searchlib/fef/phrasesplitter.cpp
@@ -15,6 +15,11 @@ PhraseSplitter::PhraseSplitter(const PhraseSplitterQueryEnv& phrase_splitter_que
     for (auto & term_match : _termMatches) {
         term_match.setFieldId(field_id);
     }
+    auto &phrase_terms = _phrase_splitter_query_env.get_phrase_terms();
+    for (const auto &phrase_term : phrase_terms) {
+        // Record that we need normal term field match data
+        (void) phrase_term.term.lookupField(field_id)->getHandle(MatchDataDetails::Normal);
+    }
 }
 
 PhraseSplitter::~PhraseSplitter() = default;