Merge pull request #29157 from vespa-engine/toregge/add-attribute-tokens-dfw

Add attribute tokens dfw.

18 files changed, 412 insertions, 17 deletions

diff --git a/searchlib/src/vespa/searchcommon/attribute/iattributevector.h b/searchlib/src/vespa/searchcommon/attribute/iattributevector.h
index d613bf61a16..b44b6a4baf2 100644
--- a/searchlib/src/vespa/searchcommon/attribute/iattributevector.h
+++ b/searchlib/src/vespa/searchcommon/attribute/iattributevector.h
@@ -419,6 +419,13 @@ public:
      */
     virtual bool isImported() const = 0;
 
+    /*
+     * Returns whether the match setting for the attribute is uncased.
+     * This is only relevant for string attributes (i.e. when isStringType()
+     * returns true). The default for string attributes is uncased matching.
+     */
+    virtual bool has_uncased_matching() const noexcept { return true; }
+
     /**
      * Will serialize the values for the documentid in ascending order. The serialized form can be used by memcmp and
      * sortorder will be preserved.
diff --git a/searchlib/src/vespa/searchlib/attribute/stringbase.cpp b/searchlib/src/vespa/searchlib/attribute/stringbase.cpp
index ac8178f8afa..0e261059777 100644
--- a/searchlib/src/vespa/searchlib/attribute/stringbase.cpp
+++ b/searchlib/src/vespa/searchlib/attribute/stringbase.cpp
@@ -283,6 +283,12 @@ StringAttribute::get_match_is_cased() const noexcept {
     return getConfig().get_match() == attribute::Config::Match::CASED;
 }
 
+bool
+StringAttribute::has_uncased_matching() const noexcept
+{
+    return !get_match_is_cased();
+}
+
 template bool AttributeVector::clearDoc(StringAttribute::ChangeVector& changes, DocId doc);
 template bool AttributeVector::update(StringAttribute::ChangeVector& changes, DocId doc, const StringChangeData& v);
 template bool AttributeVector::append(StringAttribute::ChangeVector& changes, DocId doc, const StringChangeData& v, int32_t w, bool doCount);
diff --git a/searchlib/src/vespa/searchlib/attribute/stringbase.h b/searchlib/src/vespa/searchlib/attribute/stringbase.h
index 1440b945428..e65b181f37d 100644
--- a/searchlib/src/vespa/searchlib/attribute/stringbase.h
+++ b/searchlib/src/vespa/searchlib/attribute/stringbase.h
@@ -72,6 +72,7 @@ protected:
     vespalib::MemoryUsage getChangeVectorMemoryUsage() const override;
 
     bool get_match_is_cased() const noexcept;
+    bool has_uncased_matching() const noexcept override;
     long onSerializeForAscendingSort(DocId doc, void * serTo, long available, const common::BlobConverter * bc) const override;
     long onSerializeForDescendingSort(DocId doc, void * serTo, long available, const common::BlobConverter * bc) const override;
 private:
diff --git a/searchsummary/CMakeLists.txt b/searchsummary/CMakeLists.txt
index f771a8e4494..a5d65721aeb 100644
--- a/searchsummary/CMakeLists.txt
+++ b/searchsummary/CMakeLists.txt
@@ -19,6 +19,7 @@ vespa_define_module(
     src/tests/docsummary/annotation_converter
     src/tests/docsummary/attribute_combiner
     src/tests/docsummary/attributedfw
+    src/tests/docsummary/attribute_tokens_dfw
     src/tests/docsummary/document_id_dfw
     src/tests/docsummary/tokens_converter
     src/tests/docsummary/matched_elements_filter
diff --git a/searchsummary/src/tests/docsummary/attribute_tokens_dfw/CMakeLists.txt b/searchsummary/src/tests/docsummary/attribute_tokens_dfw/CMakeLists.txt
new file mode 100644
index 00000000000..adcb18585d0
--- /dev/null
+++ b/searchsummary/src/tests/docsummary/attribute_tokens_dfw/CMakeLists.txt
@@ -0,0 +1,10 @@
+# Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+vespa_add_executable(searchsummary_attribute_tokens_dfw_test_app TEST
+    SOURCES
+    attribute_tokens_dfw_test.cpp
+    DEPENDS
+    searchsummary
+    searchsummary_test
+    GTest::GTest
+)
+vespa_add_test(NAME searchsummary_attribute_tokens_dfw_test_app COMMAND searchsummary_attribute_tokens_dfw_test_app)
diff --git a/searchsummary/src/tests/docsummary/attribute_tokens_dfw/attribute_tokens_dfw_test.cpp b/searchsummary/src/tests/docsummary/attribute_tokens_dfw/attribute_tokens_dfw_test.cpp
new file mode 100644
index 00000000000..bac817077c4
--- /dev/null
+++ b/searchsummary/src/tests/docsummary/attribute_tokens_dfw/attribute_tokens_dfw_test.cpp
@@ -0,0 +1,97 @@
+// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#include <vespa/searchsummary/docsummary/attribute_tokens_dfw.h>
+#include <vespa/searchsummary/test/mock_attribute_manager.h>
+#include <vespa/searchsummary/test/mock_state_callback.h>
+#include <vespa/searchsummary/test/slime_value.h>
+#include <vespa/vespalib/gtest/gtest.h>
+
+#include <vespa/log/log.h>
+LOG_SETUP("attribute_tokens_dfw_test");
+
+using search::attribute::CollectionType;
+using search::docsummary::AttributeTokensDFW;
+using search::docsummary::GetDocsumsState;
+using search::docsummary::DocsumFieldWriter;
+using search::docsummary::test::MockAttributeManager;
+using search::docsummary::test::MockStateCallback;
+using search::docsummary::test::SlimeValue;
+
+class AttributeTokensDFWTest : public ::testing::Test {
+protected:
+    MockAttributeManager _attrs;
+    std::unique_ptr<DocsumFieldWriter> _writer;
+    MockStateCallback _callback;
+    GetDocsumsState _state;
+    std::shared_ptr<search::MatchingElementsFields> _matching_elems_fields;
+    vespalib::string _field_name;
+
+public:
+    AttributeTokensDFWTest()
+        : _attrs(),
+          _writer(),
+          _callback(),
+          _state(_callback),
+          _matching_elems_fields(),
+          _field_name()
+    {
+        _attrs.build_string_attribute("array_str", { {"This", "is", "A TEST"}, {} });
+        _attrs.build_string_attribute("cased_array_str", { {"CASING", "Matters here" }, {} }, CollectionType::ARRAY, false);
+        _attrs.build_string_attribute("wset_str", { {"This is", "b", "C"}, {} }, CollectionType::WSET);
+        _attrs.build_string_attribute("single_str", { {"Hello World"}, {} }, CollectionType::SINGLE);
+        _state._attrCtx = _attrs.mgr().createContext();
+    }
+    ~AttributeTokensDFWTest() {}
+
+    void setup(const vespalib::string& field_name) {
+        _writer = std::make_unique<AttributeTokensDFW>(field_name);
+        _writer->setIndex(0);
+        auto attr = _state._attrCtx->getAttribute(field_name);
+        EXPECT_TRUE(_writer->setFieldWriterStateIndex(0));
+        _state._fieldWriterStates.resize(1);
+        _field_name = field_name;
+        _state._attributes.resize(1);
+        _state._attributes[0] = attr;
+    }
+
+    void expect_field(const vespalib::string& exp_slime_as_json, uint32_t docid) {
+        vespalib::Slime act;
+        vespalib::slime::SlimeInserter inserter(act);
+        if (!_writer->isDefaultValue(docid, _state)) {
+            _writer->insertField(docid, nullptr, _state, inserter);
+        }
+
+        SlimeValue exp(exp_slime_as_json);
+        EXPECT_EQ(exp.slime, act);
+    }
+};
+
+TEST_F(AttributeTokensDFWTest, outputs_slime_for_array_of_string)
+{
+    setup("array_str");
+    expect_field("[ ['this' ], [ 'is' ], [ 'a test' ] ]", 1);
+    expect_field("null", 2);
+}
+
+TEST_F(AttributeTokensDFWTest, outputs_slime_for_cased_array_of_string)
+{
+    setup("cased_array_str");
+    expect_field("[ ['CASING' ], [ 'Matters here' ] ]", 1);
+    expect_field("null", 2);
+}
+
+TEST_F(AttributeTokensDFWTest, outputs_slime_for_wset_of_string)
+{
+    setup("wset_str");
+    expect_field("[ ['this is'], [ 'b' ], [ 'c' ] ]", 1);
+    expect_field("null", 2);
+}
+
+TEST_F(AttributeTokensDFWTest, single_string)
+{
+    setup("single_str");
+    expect_field("[ 'hello world' ]", 1);
+    expect_field("[ '' ]", 2);
+}
+
+GTEST_MAIN_RUN_ALL_TESTS()
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/CMakeLists.txt b/searchsummary/src/vespa/searchsummary/docsummary/CMakeLists.txt
index 0287517f830..84ceb18ef6f 100644
--- a/searchsummary/src/vespa/searchsummary/docsummary/CMakeLists.txt
+++ b/searchsummary/src/vespa/searchsummary/docsummary/CMakeLists.txt
@@ -5,6 +5,7 @@ vespa_add_library(searchsummary_docsummary OBJECT
     array_attribute_combiner_dfw.cpp
     attribute_combiner_dfw.cpp
     attribute_field_writer.cpp
+    attribute_tokens_dfw.cpp
     attributedfw.cpp
     check_undefined_value_visitor.cpp
     copy_dfw.cpp
@@ -17,6 +18,7 @@ vespa_add_library(searchsummary_docsummary OBJECT
     document_id_dfw.cpp
     dynamicteaserdfw.cpp
     empty_dfw.cpp
+    empty_docsum_field_writer_state.cpp
     geoposdfw.cpp
     getdocsumargs.cpp
     juniper_dfw_query_item.cpp
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/attribute_tokens_dfw.cpp b/searchsummary/src/vespa/searchsummary/docsummary/attribute_tokens_dfw.cpp
new file mode 100644
index 00000000000..9e0dafc5e91
--- /dev/null
+++ b/searchsummary/src/vespa/searchsummary/docsummary/attribute_tokens_dfw.cpp
@@ -0,0 +1,179 @@
+// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#include "attribute_tokens_dfw.h"
+#include "docsumstate.h"
+#include "empty_docsum_field_writer_state.h"
+#include <vespa/searchcommon/attribute/iattributevector.h>
+#include <vespa/searchcommon/attribute/i_multi_value_attribute.h>
+#include <vespa/vespalib/data/slime/slime.h>
+#include <vespa/vespalib/text/lowercase.h>
+#include <vespa/vespalib/text/utf8.h>
+#include <vespa/vespalib/util/stash.h>
+
+using search::attribute::IAttributeVector;
+using search::attribute::BasicType;
+using search::attribute::IMultiValueAttribute;
+using search::attribute::IMultiValueReadView;
+using vespalib::LowerCase;
+using vespalib::Utf8Reader;
+using vespalib::Utf8Writer;
+using vespalib::slime::ArrayInserter;
+using vespalib::slime::Cursor;
+using vespalib::slime::Inserter;
+
+namespace search::docsummary {
+
+namespace {
+
+const IMultiValueReadView<const char*>*
+make_read_view(const IAttributeVector& attribute, vespalib::Stash& stash)
+{
+    auto multi_value_attribute = attribute.as_multi_value_attribute();
+    if (multi_value_attribute != nullptr) {
+        return multi_value_attribute->make_read_view(IMultiValueAttribute::MultiValueTag<const char*>(), stash);
+    }
+    return nullptr;
+}
+
+void
+insert_value(vespalib::stringref value, Inserter& inserter, vespalib::string& scratch, bool lowercase)
+{
+    Cursor& arr = inserter.insertArray(1);
+    ArrayInserter ai(arr);
+    if (lowercase) {
+        scratch.clear();
+        Utf8Reader r(value);
+        Utf8Writer w(scratch);
+        while (r.hasMore()) {
+            w.putChar(LowerCase::convert(r.getChar()));
+        }
+        ai.insertString(scratch);
+    } else {
+        ai.insertString(value);
+    }
+}
+
+}
+
+class MultiAttributeTokensDFWState : public DocsumFieldWriterState
+{
+    const IMultiValueReadView<const char*>* _read_view;
+    vespalib::string                        _lowercase_scratch;
+    bool                                    _lowercase;
+public:
+    MultiAttributeTokensDFWState(const IAttributeVector& attr, vespalib::Stash& stash);
+    ~MultiAttributeTokensDFWState() override;
+    void insertField(uint32_t docid, Inserter& target) override;
+};
+
+MultiAttributeTokensDFWState::MultiAttributeTokensDFWState(const IAttributeVector& attr, vespalib::Stash& stash)
+    : DocsumFieldWriterState(),
+      _read_view(make_read_view(attr, stash)),
+      _lowercase_scratch(),
+      _lowercase(attr.has_uncased_matching())
+{
+}
+
+MultiAttributeTokensDFWState::~MultiAttributeTokensDFWState() = default;
+
+void
+MultiAttributeTokensDFWState::insertField(uint32_t docid, Inserter& target)
+{
+    if (!_read_view) {
+        return;
+    }
+    auto elements = _read_view->get_values(docid);
+    if (elements.empty()) {
+        return;
+    }
+    Cursor &arr = target.insertArray(elements.size());
+    ArrayInserter ai(arr);
+    for (const auto & element : elements) {
+        insert_value(element, ai, _lowercase_scratch, _lowercase);
+    }
+}
+
+class SingleAttributeTokensDFWState : public DocsumFieldWriterState
+{
+    const IAttributeVector& _attr;
+    vespalib::string        _lowercase_scratch;
+    bool                    _lowercase;
+public:
+    SingleAttributeTokensDFWState(const IAttributeVector& attr);
+    ~SingleAttributeTokensDFWState() override;
+    void insertField(uint32_t docid, Inserter& target) override;
+};
+
+SingleAttributeTokensDFWState::SingleAttributeTokensDFWState(const IAttributeVector& attr)
+    : DocsumFieldWriterState(),
+      _attr(attr),
+      _lowercase_scratch(),
+      _lowercase(attr.has_uncased_matching())
+{
+}
+
+SingleAttributeTokensDFWState::~SingleAttributeTokensDFWState() = default;
+
+void
+SingleAttributeTokensDFWState::insertField(uint32_t docid, Inserter& target)
+{
+    auto s = _attr.get_raw(docid);
+    insert_value(vespalib::stringref(s.data(), s.size()), target, _lowercase_scratch, _lowercase);
+}
+
+DocsumFieldWriterState*
+make_field_writer_state(const IAttributeVector& attr, vespalib::Stash& stash)
+{
+    auto type = attr.getBasicType();
+    switch (type) {
+    case BasicType::Type::STRING:
+        if (attr.hasMultiValue()) {
+            return &stash.create<MultiAttributeTokensDFWState>(attr, stash);
+        } else {
+            return &stash.create<SingleAttributeTokensDFWState>(attr);
+        }
+    default:
+        ;
+    }
+    return &stash.create<EmptyDocsumFieldWriterState>();
+}
+
+AttributeTokensDFW::AttributeTokensDFW(const vespalib::string& input_field_name)
+    : DocsumFieldWriter(),
+      _input_field_name(input_field_name)
+{
+}
+
+AttributeTokensDFW::~AttributeTokensDFW() = default;
+
+const vespalib::string&
+AttributeTokensDFW::getAttributeName() const
+{
+    return _input_field_name;
+}
+
+bool
+AttributeTokensDFW::isGenerated() const
+{
+    return true;
+}
+
+bool
+AttributeTokensDFW::setFieldWriterStateIndex(uint32_t fieldWriterStateIndex)
+{
+    _state_index = fieldWriterStateIndex;
+    return true;
+}
+
+void
+AttributeTokensDFW::insertField(uint32_t docid, const IDocsumStoreDocument*, GetDocsumsState& state, vespalib::slime::Inserter& target) const
+{
+    auto& field_writer_state = state._fieldWriterStates[_state_index];
+    if (!field_writer_state) {
+        const auto& attr = *state.getAttribute(getIndex());
+        field_writer_state = make_field_writer_state(attr, state.get_stash());
+    }
+    field_writer_state->insertField(docid, target);
+}
+
+}
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/attribute_tokens_dfw.h b/searchsummary/src/vespa/searchsummary/docsummary/attribute_tokens_dfw.h
new file mode 100644
index 00000000000..53bb3d73290
--- /dev/null
+++ b/searchsummary/src/vespa/searchsummary/docsummary/attribute_tokens_dfw.h
@@ -0,0 +1,30 @@
+// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#pragma once
+
+#include "docsum_field_writer.h"
+
+namespace search::docsummary {
+
+/*
+ * Class for writing values from a string attribute vector as arrays
+ * containing the tokens. The string values are not split but they are
+ * lowercased if the string attribute vector uses uncased matching.
+ */
+class AttributeTokensDFW : public DocsumFieldWriter
+{
+private:
+    vespalib::string            _input_field_name;
+    uint32_t _state_index; // index into _fieldWriterStates in GetDocsumsState
+
+protected:
+    const vespalib::string & getAttributeName() const override;
+public:
+    AttributeTokensDFW(const vespalib::string& input_field_name);
+    ~AttributeTokensDFW() override;
+    bool isGenerated() const override;
+    bool setFieldWriterStateIndex(uint32_t fieldWriterStateIndex) override;
+    void insertField(uint32_t docid, const IDocsumStoreDocument* doc, GetDocsumsState& state, vespalib::slime::Inserter& target) const override;
+};
+
+}
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/attributedfw.cpp b/searchsummary/src/vespa/searchsummary/docsummary/attributedfw.cpp
index 4ec406b7cf0..0fc26387c00 100644
--- a/searchsummary/src/vespa/searchsummary/docsummary/attributedfw.cpp
+++ b/searchsummary/src/vespa/searchsummary/docsummary/attributedfw.cpp
@@ -4,6 +4,7 @@
 #include "docsumwriter.h"
 #include "docsumstate.h"
 #include "docsum_field_writer_state.h"
+#include "empty_docsum_field_writer_state.h"
 #include <vespa/eval/eval/value.h>
 #include <vespa/eval/eval/value_codec.h>
 #include <vespa/searchcommon/attribute/i_multi_value_attribute.h>
@@ -129,14 +130,6 @@ make_read_view(const IAttributeVector& attribute, vespalib::Stash& stash)
     return nullptr;
 }
 
-class EmptyWriterState : public DocsumFieldWriterState
-{
-public:
-    EmptyWriterState() = default;
-    ~EmptyWriterState() = default;
-    void insertField(uint32_t, Inserter&) override { }
-};
-
 template <typename MultiValueType>
 class MultiAttrDFWState : public DocsumFieldWriterState
 {
@@ -302,7 +295,7 @@ make_field_writer_state(const vespalib::string& field_name, const IAttributeVect
     default:
         ;
     }
-    return &stash.create<EmptyWriterState>();
+    return &stash.create<EmptyDocsumFieldWriterState>();
 }
 
 void
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.cpp b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.cpp
index 2ac5d1babbf..d3fc71b3173 100644
--- a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.cpp
+++ b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.cpp
@@ -7,6 +7,7 @@ namespace search::docsummary::command {
 const vespalib::string abs_distance("absdist");
 const vespalib::string attribute("attribute");
 const vespalib::string attribute_combiner("attributecombiner");
+const vespalib::string attribute_tokens("attribute-tokens");
 const vespalib::string copy("copy");
 const vespalib::string documentid("documentid");
 const vespalib::string dynamic_teaser("dynamicteaser");
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.h b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.h
index d53351d8b04..d77416f2df5 100644
--- a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.h
+++ b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_commands.h
@@ -13,6 +13,7 @@ namespace search::docsummary::command {
 extern const vespalib::string abs_distance;
 extern const vespalib::string attribute;
 extern const vespalib::string attribute_combiner;
+extern const vespalib::string attribute_tokens;
 extern const vespalib::string copy;
 extern const vespalib::string documentid;
 extern const vespalib::string dynamic_teaser;
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_factory.cpp b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_factory.cpp
index 2f7d9acdb65..28a2c34ca87 100644
--- a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_factory.cpp
+++ b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_factory.cpp
@@ -1,6 +1,7 @@
 // Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "attribute_combiner_dfw.h"
+#include "attribute_tokens_dfw.h"
 #include "copy_dfw.h"
 #include "docsum_field_writer_commands.h"
 #include "docsum_field_writer_factory.h"
@@ -16,8 +17,10 @@
 #include "tokens_dfw.h"
 #include <vespa/searchlib/common/matching_elements_fields.h>
 #include <vespa/vespalib/util/exceptions.h>
+#include <vespa/vespalib/util/issue.h>
 
 using vespalib::IllegalArgumentException;
+using vespalib::Issue;
 
 namespace search::docsummary {
 
@@ -91,6 +94,21 @@ DocsumFieldWriterFactory::create_docsum_field_writer(const vespalib::string& fie
         } else {
             throw_missing_source(command);
         }
+    } else if (command == command::attribute_tokens) {
+        if (!source.empty()) {
+            if (has_attribute_manager()) {
+                auto ctx = getEnvironment().getAttributeManager()->createContext();
+                const auto* attr = ctx->getAttribute(source);
+                if (attr == nullptr) {
+                    Issue::report("No valid attribute vector found: field='%s', command='%s', source='%s'",
+                                  field_name.c_str(), command.c_str(), source.c_str());
+                } else {
+                    fieldWriter = std::make_unique<AttributeTokensDFW>(source);
+                }
+            }
+        } else {
+            throw_missing_source(command);
+        }
     } else if (command == command::abs_distance) {
         if (has_attribute_manager()) {
             fieldWriter = AbsDistanceDFW::create(source.c_str(), getEnvironment().getAttributeManager());
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_state.h b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_state.h
index efacd0b1a49..ce151d82fc0 100644
--- a/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_state.h
+++ b/searchsummary/src/vespa/searchsummary/docsummary/docsum_field_writer_state.h
@@ -2,6 +2,8 @@
 
 #pragma once
 
+#include <cstdint>
+
 namespace vespalib::slime { struct Inserter; }
 
 namespace search::docsummary {
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/empty_docsum_field_writer_state.cpp b/searchsummary/src/vespa/searchsummary/docsummary/empty_docsum_field_writer_state.cpp
new file mode 100644
index 00000000000..b279430a367
--- /dev/null
+++ b/searchsummary/src/vespa/searchsummary/docsummary/empty_docsum_field_writer_state.cpp
@@ -0,0 +1,18 @@
+// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#include "empty_docsum_field_writer_state.h"
+
+using vespalib::slime::Inserter;
+
+namespace search::docsummary {
+
+EmptyDocsumFieldWriterState::EmptyDocsumFieldWriterState() = default;
+
+EmptyDocsumFieldWriterState::~EmptyDocsumFieldWriterState() = default;
+
+void
+EmptyDocsumFieldWriterState::insertField(uint32_t, Inserter&)
+{
+}
+
+}
diff --git a/searchsummary/src/vespa/searchsummary/docsummary/empty_docsum_field_writer_state.h b/searchsummary/src/vespa/searchsummary/docsummary/empty_docsum_field_writer_state.h
new file mode 100644
index 00000000000..ff8319b52d1
--- /dev/null
+++ b/searchsummary/src/vespa/searchsummary/docsummary/empty_docsum_field_writer_state.h
@@ -0,0 +1,21 @@
+// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#pragma once
+
+#include "docsum_field_writer_state.h"
+
+namespace search::docsummary {
+
+/*
+ * Class used as fallback when no suitable field writer state could be
+ * instantiated. insertField() is a noop.
+ */
+class EmptyDocsumFieldWriterState : public DocsumFieldWriterState
+{
+public:
+    EmptyDocsumFieldWriterState();
+    ~EmptyDocsumFieldWriterState() override;
+    void insertField(uint32_t, vespalib::slime::Inserter&) override;
+};
+
+}
diff --git a/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.cpp b/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.cpp
index 557ba8f4cde..4d73420b523 100644
--- a/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.cpp
+++ b/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.cpp
@@ -20,9 +20,13 @@ template <typename AttributeType, typename ValueType>
 void
 MockAttributeManager::build_attribute(const vespalib::string& name, BasicType type,
                                       CollectionType col_type,
-                                      const std::vector<std::vector<ValueType>>& values)
+                                      const std::vector<std::vector<ValueType>>& values,
+                                      std::optional<bool> uncased)
 {
     Config cfg(type, col_type);
+    if (uncased.has_value()) {
+        cfg.set_match(uncased.value() ? Config::Match::UNCASED  : Config::Match::CASED);
+    }
     auto attr_base = AttributeFactory::createAttribute(name, cfg);
     assert(attr_base);
     auto attr = std::dynamic_pointer_cast<AttributeType>(attr_base);
@@ -55,9 +59,10 @@ MockAttributeManager::~MockAttributeManager() = default;
 void
 MockAttributeManager::build_string_attribute(const vespalib::string& name,
                                              const std::vector<std::vector<vespalib::string>>& values,
-                                             CollectionType col_type)
+                                             CollectionType col_type,
+                                             std::optional<bool> uncased)
 {
-    build_attribute<StringAttribute, vespalib::string>(name, BasicType::Type::STRING, col_type, values);
+    build_attribute<StringAttribute, vespalib::string>(name, BasicType::Type::STRING, col_type, values, uncased);
 }
 
 void
@@ -65,7 +70,7 @@ MockAttributeManager::build_float_attribute(const vespalib::string& name,
                                             const std::vector<std::vector<double>>& values,
                                             CollectionType col_type)
 {
-    build_attribute<FloatingPointAttribute, double>(name, BasicType::Type::DOUBLE, col_type, values);
+    build_attribute<FloatingPointAttribute, double>(name, BasicType::Type::DOUBLE, col_type, values, std::nullopt);
 }
 
 void
@@ -73,14 +78,14 @@ MockAttributeManager::build_int_attribute(const vespalib::string& name, BasicTyp
                                           const std::vector<std::vector<int64_t>>& values,
                                           CollectionType col_type)
 {
-    build_attribute<IntegerAttribute, int64_t>(name, type, col_type, values);
+    build_attribute<IntegerAttribute, int64_t>(name, type, col_type, values, std::nullopt);
 }
 
 void
 MockAttributeManager::build_raw_attribute(const vespalib::string& name,
                                           const std::vector<std::vector<std::vector<char>>>& values)
 {
-    build_attribute<SingleRawAttribute, std::vector<char>>(name, BasicType::Type::RAW, CollectionType::SINGLE, values);
+    build_attribute<SingleRawAttribute, std::vector<char>>(name, BasicType::Type::RAW, CollectionType::SINGLE, values, std::nullopt);
 }
 
 }
diff --git a/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.h b/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.h
index 69762a37f95..ae105025826 100644
--- a/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.h
+++ b/searchsummary/src/vespa/searchsummary/test/mock_attribute_manager.h
@@ -2,6 +2,7 @@
 
 #include <vespa/searchcommon/attribute/basictype.h>
 #include <vespa/searchlib/attribute/attributemanager.h>
+#include <optional>
 
 namespace search::docsummary::test {
 
@@ -15,7 +16,8 @@ private:
     template <typename AttributeType, typename ValueType>
     void build_attribute(const vespalib::string& name, search::attribute::BasicType type,
                          search::attribute::CollectionType col_type,
-                         const std::vector<std::vector<ValueType>>& values);
+                         const std::vector<std::vector<ValueType>>& values,
+                         std::optional<bool> uncased);
 
 public:
     MockAttributeManager();
@@ -24,7 +26,8 @@ public:
 
     void build_string_attribute(const vespalib::string& name,
                                 const std::vector<std::vector<vespalib::string>>& values,
-                                search::attribute::CollectionType col_type = search::attribute::CollectionType::ARRAY);
+                                search::attribute::CollectionType col_type = search::attribute::CollectionType::ARRAY,
+                                std::optional<bool> uncased = std::nullopt);
     void build_float_attribute(const vespalib::string& name,
                                const std::vector<std::vector<double>>& values,
                                search::attribute::CollectionType col_type = search::attribute::CollectionType::ARRAY);