Merge pull request #13347 from vespa-engine/toregge/add-field-match-computer-shared-state

Factor out portions of field match computer to field match computer shared state.

6 files changed, 141 insertions, 61 deletions

diff --git a/searchlib/src/vespa/searchlib/features/fieldmatch/CMakeLists.txt b/searchlib/src/vespa/searchlib/features/fieldmatch/CMakeLists.txt
index 7786e3b45a1..a3273a4a39c 100644
--- a/searchlib/src/vespa/searchlib/features/fieldmatch/CMakeLists.txt
+++ b/searchlib/src/vespa/searchlib/features/fieldmatch/CMakeLists.txt
@@ -2,6 +2,7 @@
 vespa_add_library(searchlib_features_fieldmatch OBJECT
     SOURCES
     computer.cpp
+    computer_shared_state.cpp
     metrics.cpp
     params.cpp
     segmentstart.cpp
diff --git a/searchlib/src/vespa/searchlib/features/fieldmatch/computer.cpp b/searchlib/src/vespa/searchlib/features/fieldmatch/computer.cpp
index 43aee8167ee..7c3c0c5d638 100644
--- a/searchlib/src/vespa/searchlib/features/fieldmatch/computer.cpp
+++ b/searchlib/src/vespa/searchlib/features/fieldmatch/computer.cpp
@@ -1,8 +1,10 @@
 // Copyright 2017 Yahoo Holdings. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "computer.h"
+#include "computer_shared_state.h"
 #include <vespa/searchlib/features/utils.h>
 #include <vespa/searchlib/fef/phrase_splitter_query_env.h>
+#include <vespa/searchlib/fef/phrasesplitter.h>
 #include <vespa/searchlib/fef/properties.h>
 #include <vespa/vespalib/util/stringfmt.h>
 #include <vespa/vespalib/locale/c.h>
@@ -15,55 +17,24 @@ using namespace search::fef;
 
 namespace search::features::fieldmatch {
 
-Computer::Computer(const vespalib::string &propertyNamespace, const PhraseSplitter &splitter,
-                   const FieldInfo &fieldInfo, const Params &params) :
-    _splitter(splitter),
-    _fieldId(fieldInfo.id()),
-    _params(params),
-    _useCachedHits(true),
-    _queryTerms(),
-    _queryTermFieldMatch(),
-    _totalTermWeight(0),
-    _totalTermSignificance(0.0f),
-    _fieldLength(FieldPositionsIterator::UNKNOWN_LENGTH),
-    _currentMetrics(this),
-    _finalMetrics(this),
-    _simpleMetrics(params),
-    _segments(),
-    _alternativeSegmentationsTried(0),
-    _cachedHits()
+Computer::Computer(const ComputerSharedState& shared_state, const PhraseSplitter& splitter)
+    : _shared_state(shared_state),
+      _splitter(splitter),
+      _fieldId(_shared_state.get_field_id()),
+      _params(_shared_state.get_params()),
+      _useCachedHits(_shared_state.get_use_cached_hits()),
+      _queryTerms(_shared_state.get_query_terms()),
+      _queryTermFieldMatch(_queryTerms.size()),
+      _totalTermWeight(_shared_state.get_total_term_weight()),
+      _totalTermSignificance(_shared_state.get_total_term_significance()),
+      _fieldLength(FieldPositionsIterator::UNKNOWN_LENGTH),
+      _currentMetrics(this),
+      _finalMetrics(this),
+      _simpleMetrics(_shared_state.get_simple_metrics()),
+      _segments(),
+      _alternativeSegmentationsTried(0),
+      _cachedHits(_queryTerms.size())
 {
-    // Store term data for all terms searching in this field
-    const auto& splitter_query_env = splitter.get_query_env();
-    _queryTermFieldMatch.reserve(splitter_query_env.getNumTerms());
-    _cachedHits.reserve(splitter_query_env.getNumTerms());
-    for (uint32_t i = 0; i < splitter_query_env.getNumTerms(); ++i) {
-        QueryTerm qt = QueryTermFactory::create(splitter_query_env, i, true);
-        _totalTermWeight += qt.termData()->getWeight().percent();
-        _totalTermSignificance += qt.significance();
-        _simpleMetrics.addQueryTerm(qt.termData()->getWeight().percent());
-        const ITermFieldData *field = qt.termData()->lookupField(_fieldId);
-        if (field != nullptr) {
-            qt.fieldHandle(field->getHandle());
-            _queryTerms.push_back(qt);
-            _simpleMetrics.addSearchedTerm(qt.termData()->getWeight().percent());
-            _queryTermFieldMatch.emplace_back(nullptr);
-            _cachedHits.emplace_back();
-        }
-    }
-
-    _totalTermWeight = atoi(splitter_query_env.getProperties().lookup(propertyNamespace, "totalTermWeight").
-                            get(vespalib::make_string("%d", _totalTermWeight)).c_str());
-    _totalTermSignificance = vespalib::locale::c::atof(splitter_query_env.getProperties().lookup(propertyNamespace, "totalTermSignificance").
-                                  get(vespalib::make_string("%f", _totalTermSignificance)).c_str());
-    if (splitter_query_env.getProperties().lookup(propertyNamespace, "totalTermWeight").found()) {
-        _simpleMetrics.setTotalWeightInQuery(_totalTermWeight);
-    }
-
-    // update current and final metrics after initialization
-    _currentMetrics = Metrics(this);
-    _finalMetrics = Metrics(this);
-
     // num query terms searching in this field + 1
     _segments.reserve(getNumQueryTerms() + 1);
     for (uint32_t i = 0; i < (getNumQueryTerms() + 1); ++i) {
diff --git a/searchlib/src/vespa/searchlib/features/fieldmatch/computer.h b/searchlib/src/vespa/searchlib/features/fieldmatch/computer.h
index b2644208b40..f684f42708a 100644
--- a/searchlib/src/vespa/searchlib/features/fieldmatch/computer.h
+++ b/searchlib/src/vespa/searchlib/features/fieldmatch/computer.h
@@ -5,15 +5,21 @@
 #include "params.h"
 #include "segmentstart.h"
 #include "simplemetrics.h"
-#include <vespa/searchlib/fef/iqueryenvironment.h>
-#include <vespa/searchlib/fef/fieldinfo.h>
-#include <vespa/searchlib/fef/matchdata.h>
-#include <vespa/searchlib/fef/phrasesplitter.h>
 #include <vespa/searchlib/features/queryterm.h>
 #include <vespa/searchlib/common/allocatedbitvector.h>
+#include <vespa/vespalib/util/arrayref.h>
+
+namespace search::fef {
+
+class PhraseSplitter;
+class TermFieldMatchData;
+
+}
 
 namespace search::features::fieldmatch {
 
+class ComputerSharedState;
+
 /**
  * <p>Calculates a set of metrics capturing information about the degree of agreement between a query and a field
  * string. This algorithm attempts to capture the property of text that very close tokens are usuall part of the same
@@ -57,13 +63,10 @@ public:
     /**
      * Constructs a new computer object.
      *
-     * @param propertyNamespace The namespace used in query properties.
+     * @param shared_state      The shared state for this computer
      * @param splitter          The environment that holds all query information.
-     * @param fieldInfo         The info object of the matched field.
-     * @param params            The parameter object for this computer.
      */
-    Computer(const vespalib::string &propertyNamespace, const fef::PhraseSplitter &splitter,
-             const fef::FieldInfo &fieldInfo, const Params &params);
+    Computer(const ComputerSharedState& shared_state, const fef::PhraseSplitter& splitter);
 
     /**
      * Resets this object according to the given document id
@@ -296,12 +299,13 @@ private:
     };
 
     // per query
+    const ComputerSharedState&                 _shared_state;
     const search::fef::PhraseSplitter        & _splitter;
     uint32_t                                   _fieldId;
-    Params                                     _params;
+    const Params                               _params;
     bool                                       _useCachedHits;
 
-    QueryTermVector                            _queryTerms;
+    const vespalib::ConstArrayRef<QueryTerm>   _queryTerms;
     TermFieldMatchDataVector                   _queryTermFieldMatch;
     uint32_t                                   _totalTermWeight;
     feature_t                                  _totalTermSignificance;
diff --git a/searchlib/src/vespa/searchlib/features/fieldmatch/computer_shared_state.cpp b/searchlib/src/vespa/searchlib/features/fieldmatch/computer_shared_state.cpp
new file mode 100644
index 00000000000..c6093f52619
--- /dev/null
+++ b/searchlib/src/vespa/searchlib/features/fieldmatch/computer_shared_state.cpp
@@ -0,0 +1,49 @@
+// Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#include "computer_shared_state.h"
+#include <vespa/searchlib/features/utils.h>
+#include <vespa/searchlib/fef/phrase_splitter_query_env.h>
+#include <vespa/searchlib/fef/properties.h>
+#include <vespa/vespalib/util/stringfmt.h>
+#include <vespa/vespalib/locale/c.h>
+#include <set>
+
+using namespace search::fef;
+
+namespace search::features::fieldmatch {
+
+ComputerSharedState::ComputerSharedState(const vespalib::string& propertyNamespace, const PhraseSplitterQueryEnv& splitter_query_env,
+                   const FieldInfo& fieldInfo, const Params& params)
+    : _splitter_query_env(splitter_query_env),
+      _field_id(fieldInfo.id()),
+      _params(params),
+      _use_cached_hits(true),
+      _query_terms(),
+      _total_term_weight(0),
+      _total_term_significance(0.0f),
+      _simple_metrics(_params)
+{
+    // Store term data for all terms searching in this field
+    for (uint32_t i = 0; i < splitter_query_env.getNumTerms(); ++i) {
+        QueryTerm qt = QueryTermFactory::create(splitter_query_env, i, true);
+        _total_term_weight += qt.termData()->getWeight().percent();
+        _total_term_significance += qt.significance();
+        _simple_metrics.addQueryTerm(qt.termData()->getWeight().percent());
+        const ITermFieldData *field = qt.termData()->lookupField(_field_id);
+        if (field != nullptr) {
+            qt.fieldHandle(field->getHandle());
+            _query_terms.push_back(qt);
+            _simple_metrics.addSearchedTerm(qt.termData()->getWeight().percent());
+        }
+    }
+
+    _total_term_weight = atoi(splitter_query_env.getProperties().lookup(propertyNamespace, "totalTermWeight").
+                              get(vespalib::make_string("%d", _total_term_weight)).c_str());
+    _total_term_significance = vespalib::locale::c::atof(splitter_query_env.getProperties().lookup(propertyNamespace, "totalTermSignificance").
+                                                         get(vespalib::make_string("%f", _total_term_significance)).c_str());
+    if (splitter_query_env.getProperties().lookup(propertyNamespace, "totalTermWeight").found()) {
+        _simple_metrics.setTotalWeightInQuery(_total_term_weight);
+    }
+}
+
+}
diff --git a/searchlib/src/vespa/searchlib/features/fieldmatch/computer_shared_state.h b/searchlib/src/vespa/searchlib/features/fieldmatch/computer_shared_state.h
new file mode 100644
index 00000000000..2d0eaaf2737
--- /dev/null
+++ b/searchlib/src/vespa/searchlib/features/fieldmatch/computer_shared_state.h
@@ -0,0 +1,52 @@
+// Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+#pragma once
+
+#include "params.h"
+#include "simplemetrics.h"
+#include <vespa/searchlib/features/queryterm.h>
+
+namespace search::fef { class PhraseSplitterQueryEnv; }
+
+namespace search::features::fieldmatch {
+
+/**
+ * Shared state for field match computer.
+ */
+class ComputerSharedState {
+public:
+    /**
+     * Constructs a new computer shared state object.
+     *
+     * @param propertyNamespace The namespace used in query properties.
+     * @param splitter_query_env The environment that holds all query information.
+     * @param fieldInfo         The info object of the matched field.
+     * @param params            The parameter object for this computer.
+     */
+    ComputerSharedState(const vespalib::string& propertyNamespace, const fef::PhraseSplitterQueryEnv& splitter_query_env,
+                        const fef::FieldInfo& fieldInfo, const Params& params);
+
+    uint32_t get_field_id() const { return _field_id; }
+    const Params& get_params() const { return _params; }
+    bool get_use_cached_hits() const { return _use_cached_hits; }
+    const QueryTermVector& get_query_terms() const { return _query_terms; }
+    uint32_t get_total_term_weight() const { return _total_term_weight; }
+    feature_t get_total_term_significance() const { return _total_term_significance; }
+    const SimpleMetrics& get_simple_metrics() const { return _simple_metrics; }
+
+private:
+
+    // per query
+    const search::fef::PhraseSplitterQueryEnv& _splitter_query_env;
+    uint32_t                                   _field_id;
+    Params                                     _params;
+    bool                                       _use_cached_hits;
+
+    QueryTermVector                            _query_terms;
+    uint32_t                                   _total_term_weight;
+    feature_t                                  _total_term_significance;
+
+    // portions per docid (not used here), portions per query
+    SimpleMetrics                              _simple_metrics;  // The metrics used to compute simple features.
+};
+
+}
diff --git a/searchlib/src/vespa/searchlib/features/fieldmatchfeature.cpp b/searchlib/src/vespa/searchlib/features/fieldmatchfeature.cpp
index 62b5f1d8165..9e4aa0d96ab 100644
--- a/searchlib/src/vespa/searchlib/features/fieldmatchfeature.cpp
+++ b/searchlib/src/vespa/searchlib/features/fieldmatchfeature.cpp
@@ -3,9 +3,11 @@
 #include "fieldmatchfeature.h"
 #include "utils.h"
 #include <vespa/searchlib/features/fieldmatch/computer.h>
+#include <vespa/searchlib/features/fieldmatch/computer_shared_state.h>
 #include <vespa/searchlib/fef/featurenamebuilder.h>
 #include <vespa/searchlib/fef/indexproperties.h>
 #include <vespa/searchlib/fef/phrase_splitter_query_env.h>
+#include <vespa/searchlib/fef/phrasesplitter.h>
 #include <vespa/searchlib/fef/properties.h>
 #include <vespa/vespalib/util/stringfmt.h>
 #include <vespa/vespalib/locale/c.h>
@@ -24,6 +26,7 @@ private:
     PhraseSplitterQueryEnv _splitter_env;
     fef::PhraseSplitter    _splitter;
     const fef::FieldInfo & _field;
+    fieldmatch::ComputerSharedState _cmp_shared_state;
     fieldmatch::Computer   _cmp;
 
     void handle_bind_match_data(const fef::MatchData &md) override;
@@ -42,8 +45,8 @@ FieldMatchExecutor::FieldMatchExecutor(const IQueryEnvironment & queryEnv,
     _splitter_env(queryEnv, field.id()),
     _splitter(_splitter_env),
     _field(field),
-    _cmp(vespalib::make_string("fieldMatch(%s)", _field.name().c_str()),
-         _splitter, field, params)
+    _cmp_shared_state(vespalib::make_string("fieldMatch(%s)", _field.name().c_str()), _splitter_env, field, params),
+    _cmp(_cmp_shared_state, _splitter)
 {
     // empty
 }