A first draft of adding triples for the langmatches optimization.

Signed-off-by: Johannes Kalmbach <[email protected]>

src/engine/sparqlExpressions/SparqlExpressionPimpl.h

@@ -99,6 +99,7 @@ class SparqlExpressionPimpl {
   struct LangFilterData {
     Variable variable_;
     std::string language_;
+    bool isLangmatches_ = false;
   };
   std::optional<LangFilterData> getLanguageFilterExpression() const;
 

src/engine/sparqlExpressions/StringExpressions.cpp

@@ -479,9 +479,28 @@ using EncodeForUriExpression =
       }
     };
 
-using LangMatches =
+using LangMatchesImpl =
     StringExpressionImpl<2, decltype(langMatching), StringValueGetter>;
 
+class LangMatches : public LangMatchesImpl {
+ public:
+  using LangMatchesImpl::LangMatchesImpl;
+  std::optional<LangFilterData> getLanguageFilterExpression() const override {
+    AD_CORRECTNESS_CHECK(children().size() == 2);
+    auto* var = dynamic_cast<const VariableExpression*>(children()[0].get());
+    auto* str =
+        dynamic_cast<const StringLiteralExpression*>(children()[1].get());
+    if (!(var && str)) {
+      return std::nullopt;
+    }
+    // TODO<joka921> We need to check whether the literal is plain. (no language
+    // tag or something else).
+    return LangFilterData{
+        var->value(),
+        std::string(asStringViewUnsafe(str->value().getContent())), true};
+  }
+};
+
 // STRING WITH LANGUAGE TAG
 [[maybe_unused]] inline auto strLangTag =
     [](std::optional<std::string> input,

src/index/IndexBuilderTypes.h

@@ -243,15 +243,15 @@ auto getIdMapLambdas(
     // the allocation and deallocation of these hash maps (that are newly
     // created for each batch) much cheaper (see `CachingMemoryResource.h` and
     // `IndexImpl.cpp`).
-    itemArray[j]->map_.map_.reserve(5 * maxNumberOfTriples / NumThreads);
+    itemArray[j]->map_.map_.reserve(6 * maxNumberOfTriples / NumThreads);
     // The LANGUAGE_PREDICATE gets the first ID in each map. TODO<joka921>
     // This is not necessary for the actual QLever code, but certain unit tests
     // currently fail without it.
     itemArray[j]->getId(TripleComponent{
         ad_utility::triple_component::Iri::fromIriref(LANGUAGE_PREDICATE)});
   }
   using OptionalIds =
-      std::array<std::optional<std::array<Id, NumColumnsIndexBuilding>>, 3>;
+      std::array<std::optional<std::array<Id, NumColumnsIndexBuilding>>, 4>;
 
   /* given an index idx, returns a lambda that
    * - Takes a triple and a language tag
@@ -278,6 +278,8 @@ auto getIdMapLambdas(
                 .iriOrLiteral_.getIri();
         auto langTaggedPredId = map.getId(TripleComponent{
             ad_utility::convertToLanguageTaggedPredicate(iri, lt.langtag_)});
+        auto langMatchesTaggedPredId = map.getId(TripleComponent{
+            ad_utility::convertToLangmatchesTaggedPredicate(iri, lt.langtag_)});
         auto& spoIds = *res[0];  // ids of original triple
         // TODO replace the std::array by an explicit IdTriple class,
         //  then the emplace calls don't need the explicit type.
@@ -299,6 +301,8 @@ auto getIdMapLambdas(
                                ad_utility::triple_component::Iri::fromIriref(
                                    LANGUAGE_PREDICATE)}),
                            langTagId, tripleGraphId});
+        res[3].emplace(
+            Arr{spoIds[0], langMatchesTaggedPredId, spoIds[2], tripleGraphId});
       }
       return res;
     };

src/parser/GraphPattern.h

@@ -43,8 +43,9 @@ class GraphPattern {
 
   // Modify query to take care of language filter. `variable` is the variable,
   // `languageInQuotes` is the language.
-  void addLanguageFilter(const Variable& variable,
-                         const std::string& languageInQuotes);
+  [[nodiscard]] bool addLanguageFilter(const Variable& variable,
+                                       const std::string& languageInQuotes,
+                                       bool isLangmatches = false);
 
   bool _optional;
 

src/parser/ParsedQuery.cpp

@@ -265,8 +265,9 @@ void ParsedQuery::registerVariableVisibleInQueryBody(const Variable& variable) {
 ParsedQuery::GraphPattern::GraphPattern() : _optional(false) {}
 
 // __________________________________________________________________________
-void ParsedQuery::GraphPattern::addLanguageFilter(const Variable& variable,
-                                                  const std::string& langTag) {
+bool ParsedQuery::GraphPattern::addLanguageFilter(const Variable& variable,
+                                                  const std::string& langTag,
+                                                  bool isLangmatches) {
   // Find all triples where the object is the `variable` and the predicate is
   // a simple `IRIREF` (neither a variable nor a complex property path).
   // Search in all the basic graph patterns, as filters have the complete
@@ -275,6 +276,10 @@ void ParsedQuery::GraphPattern::addLanguageFilter(const Variable& variable,
   // Subqueries etc.
   // TODO<joka921> Also support property paths (^rdfs:label,
   // skos:altLabel|rdfs:label, ...)
+
+  if (isLangmatches && langTag.find('-') != std::string::npos) {
+    return false;
+  }
   std::vector<SparqlTriple*> matchingTriples;
   using BasicPattern = parsedQuery::BasicGraphPattern;
   namespace ad = ad_utility;
@@ -295,14 +300,20 @@ void ParsedQuery::GraphPattern::addLanguageFilter(const Variable& variable,
 
   // Replace all the matching triples.
   for (auto* triplePtr : matchingTriples) {
-    triplePtr->p_._iri = ad_utility::convertToLanguageTaggedPredicate(
-        triplePtr->p_._iri, langTag);
+    triplePtr->p_._iri = isLangmatches
+                             ? ad_utility::convertToLangmatchesTaggedPredicate(
+                                   triplePtr->p_._iri, langTag)
+                             : ad_utility::convertToLanguageTaggedPredicate(
+                                   triplePtr->p_._iri, langTag);
   }
 
   // Handle the case, that no suitable triple (see above) was found. In this
   // case a triple `?variable ql:langtag "language"` is added at the end of
   // the graph pattern.
   if (matchingTriples.empty()) {
+    if (isLangmatches) {
+      return false;
+    }
     LOG(DEBUG) << "language filter variable " + variable.name() +
                       " did not appear as object in any suitable "
                       "triple. "
@@ -326,6 +337,7 @@ void ParsedQuery::GraphPattern::addLanguageFilter(const Variable& variable,
                         langEntity);
     t.push_back(std::move(triple));
   }
+  return true;
 }
 
 // ____________________________________________________________________________

src/parser/sparqlParser/SparqlQleverVisitor.cpp

@@ -703,8 +703,12 @@ GraphPattern Visitor::visit(Parser::GroupGraphPatternContext* ctx) {
       if (auto langFilterData =
               filter.expression_.getLanguageFilterExpression();
           langFilterData.has_value()) {
-        const auto& [variable, language] = langFilterData.value();
-        pattern.addLanguageFilter(variable, language);
+        const auto& [variable, language, isLangmatches] =
+            langFilterData.value();
+        if (!pattern.addLanguageFilter(variable, language, isLangmatches)) {
+          // TODO<joka921> Code duplication.
+          pattern._filters.push_back(std::move(filter));
+        }
       } else {
         pattern._filters.push_back(std::move(filter));
       }

src/util/Conversions.cpp

@@ -22,21 +22,39 @@
 namespace ad_utility {
 
 // _________________________________________________________
-triple_component::Iri convertLangtagToEntityUri(const string& tag) {
+triple_component::Iri convertLangtagToEntityUri(std::string_view tag) {
   return triple_component::Iri::fromIriref(makeQleverInternalIri("@", tag));
 }
 
 // _________________________________________________________
-std::string convertToLanguageTaggedPredicate(const string& pred,
-                                             const string& langtag) {
+std::string convertToLanguageTaggedPredicate(std::string_view pred,
+                                             std::string_view langtag) {
   return absl::StrCat("@", langtag, "@", pred);
 }
 
+static std::string_view getPrimaryLanguage(std::string_view language) {
+  return language.substr(0, language.find('-'));
+}
+
 // _________________________________________________________
 triple_component::Iri convertToLanguageTaggedPredicate(
-    const triple_component::Iri& pred, const std::string& langtag) {
+    const triple_component::Iri& pred, std::string_view langtag) {
   return triple_component::Iri::fromIriref(absl::StrCat(
       "@", langtag, "@<", asStringViewUnsafe(pred.getContent()), ">"));
 }
 
+// _________________________________________________________
+std::string convertToLangmatchesTaggedPredicate(std::string_view pred,
+                                                std::string_view langtag) {
+  return absl::StrCat("@@", getPrimaryLanguage(langtag), "@@", pred);
+}
+
+// _________________________________________________________
+triple_component::Iri convertToLangmatchesTaggedPredicate(
+    const triple_component::Iri& pred, std::string_view langtag) {
+  return triple_component::Iri::fromIriref(
+      absl::StrCat("@@", getPrimaryLanguage(langtag), "@@<",
+                   asStringViewUnsafe(pred.getContent()), ">"));
+}
+
 }  // namespace ad_utility

src/util/Conversions.h

@@ -17,9 +17,15 @@ constexpr std::string_view languageTaggedPredicatePrefix = "@";
 // TODO<joka921> The overload that takes and returns `std::string` can be
 // removed as soon as we also store strongly-typed IRIs in the predicates of the
 // `SparqlTriple` class.
-triple_component::Iri convertLangtagToEntityUri(const std::string& tag);
-std::string convertToLanguageTaggedPredicate(const std::string& pred,
-                                             const std::string& langtag);
+triple_component::Iri convertLangtagToEntityUri(std::string_view tag);
+std::string convertToLanguageTaggedPredicate(std::string_view pred,
+                                             std::string_view langtag);
 triple_component::Iri convertToLanguageTaggedPredicate(
-    const triple_component::Iri& pred, const std::string& langtag);
+    const triple_component::Iri& pred, std::string_view langtag);
+
+// TODO<joka921> Comment.
+std::string convertToLangmatchesTaggedPredicate(std::string_view pred,
+                                                std::string_view langtag);
+triple_component::Iri convertToLangmatchesTaggedPredicate(
+    const triple_component::Iri& pred, std::string_view langtag);
 }  // namespace ad_utility