Add InputRange... classes as a replacement for cppcoro::generator (…

…#1678)

The new classes (called `InputRange...`) and their requirements can be found in `src/util/iterators.h`. They can be used to replace our many uses of `cppcoro::generator`.

For an example, see `src/engine/Result.h`, where `using LazyResult = ad_utility::InputRangeTypeErased<IdTableVocabPair>` is defined as a replacement for `using Generator = cppcoro::generator<IdTableVocabPair>`, and the many places where now `LazyResult` is used instead of `Generator`.

Other replacements will take a little more work.

src/engine/CMakeLists.txt

@@ -1,6 +1,6 @@
 add_subdirectory(sparqlExpressions)
 add_library(SortPerformanceEstimator SortPerformanceEstimator.cpp)
-qlever_target_link_libraries(SortPerformanceEstimator)
+qlever_target_link_libraries(SortPerformanceEstimator parser)
 add_library(engine
         Engine.cpp QueryExecutionTree.cpp Operation.cpp Result.cpp LocalVocab.cpp
         IndexScan.cpp Join.cpp Sort.cpp

src/engine/Distinct.cpp

@@ -35,7 +35,7 @@ VariableToColumnMap Distinct::computeVariableToColumnMap() const {
 
 // _____________________________________________________________________________
 template <size_t WIDTH>
-Result::Generator Distinct::lazyDistinct(Result::Generator input,
+Result::Generator Distinct::lazyDistinct(Result::LazyResult input,
                                          bool yieldOnce) const {
   IdTable aggregateTable{subtree_->getResultWidth(), allocator()};
   LocalVocab aggregateVocab{};

src/engine/Distinct.h

@@ -64,7 +64,8 @@ class Distinct : public Operation {
   // if every `IdTable` from `input` should yield it's own `IdTable` or if all
   // of them should get aggregated into a single big `IdTable`.
   template <size_t WIDTH>
-  Result::Generator lazyDistinct(Result::Generator input, bool yieldOnce) const;
+  Result::Generator lazyDistinct(Result::LazyResult input,
+                                 bool yieldOnce) const;
 
   // Removes all duplicates from input with regards to the columns
   // in keepIndices. The input needs to be sorted on the keep indices,

src/engine/GroupBy.cpp

@@ -177,8 +177,6 @@ uint64_t GroupBy::getSizeEstimateBeforeLimit() {
     return _subtree->getMultiplicity(_subtree->getVariableColumn(var));
   };
 
-  // TODO<joka921> Once we can use `std::views` this can be solved
-  // more elegantly.
   float minMultiplicity = ql::ranges::min(
       _groupByVariables | ql::views::transform(varToMultiplicity));
   return _subtree->getSizeEstimate() / minMultiplicity;

src/engine/IndexScan.cpp

@@ -489,13 +489,13 @@ void IndexScan::updateRuntimeInfoForLazyScan(const LazyScanMetadata& metadata) {
 // resulting from the generator.
 struct IndexScan::SharedGeneratorState {
   // The generator that yields the tables to be joined with the index scan.
-  Result::Generator generator_;
+  Result::LazyResult generator_;
   // The column index of the join column in the tables yielded by the generator.
   ColumnIndex joinColumn_;
   // Metadata and blocks of this index scan.
   Permutation::MetadataAndBlocks metaBlocks_;
   // The iterator of the generator that is currently being consumed.
-  std::optional<Result::Generator::iterator> iterator_ = std::nullopt;
+  std::optional<Result::LazyResult::iterator> iterator_ = std::nullopt;
   // Values returned by the generator that have not been re-yielded yet.
   // Typically we expect only 3 or less values to be prefetched (this is an
   // implementation detail of `BlockZipperJoinImpl`).
@@ -648,7 +648,7 @@ Result::Generator IndexScan::createPrefilteredIndexScanSide(
 
 // _____________________________________________________________________________
 std::pair<Result::Generator, Result::Generator> IndexScan::prefilterTables(
-    Result::Generator input, ColumnIndex joinColumn) {
+    Result::LazyResult input, ColumnIndex joinColumn) {
   AD_CORRECTNESS_CHECK(numVariables_ <= 3 && numVariables_ > 0);
   auto metaBlocks = getMetadataForScan();
 

src/engine/IndexScan.h

@@ -106,7 +106,7 @@ class IndexScan final : public Operation {
   // there are undef values, the second generator represents the full index
   // scan.
   std::pair<Result::Generator, Result::Generator> prefilterTables(
-      Result::Generator input, ColumnIndex joinColumn);
+      Result::LazyResult input, ColumnIndex joinColumn);
 
  private:
   // Implementation detail that allows to consume a generator from two other

src/engine/Join.cpp

@@ -40,8 +40,10 @@ using LazyInputView =
 // Convert a `generator<IdTableVocab>` to a `generator<IdTableAndFirstCol>` for
 // more efficient access in the join columns below and apply the given
 // permutation to each table.
-LazyInputView convertGenerator(Result::Generator gen,
-                               OptionalPermutation permutation = {}) {
+CPP_template(typename Input)(
+    requires ad_utility::SameAsAny<Input, Result::Generator,
+                                   Result::LazyResult>) LazyInputView
+    convertGenerator(Input gen, OptionalPermutation permutation = {}) {
   for (auto& [table, localVocab] : gen) {
     applyPermutation(table, permutation);
     // Make sure to actually move the table into the wrapper so that the tables

src/engine/Result.cpp

@@ -124,7 +124,7 @@ void Result::applyLimitOffset(
                   limitOffset);
     limitTimeCallback(limitTimer.msecs(), idTable());
   } else {
-    auto generator = [](Generator original, LimitOffsetClause limitOffset,
+    auto generator = [](LazyResult original, LimitOffsetClause limitOffset,
                         auto limitTimeCallback) -> Generator {
       if (limitOffset._limit.value_or(1) == 0) {
         co_return;
@@ -160,7 +160,7 @@ void Result::assertThatLimitWasRespected(const LimitOffsetClause& limitOffset) {
     auto limit = limitOffset._limit;
     AD_CONTRACT_CHECK(!limit.has_value() || numRows <= limit.value());
   } else {
-    auto generator = [](Generator original,
+    auto generator = [](LazyResult original,
                         LimitOffsetClause limitOffset) -> Generator {
       auto limit = limitOffset._limit;
       uint64_t elementCount = 0;
@@ -192,7 +192,7 @@ void Result::checkDefinedness(const VariableToColumnMap& varColMap) {
     AD_EXPENSIVE_CHECK(performCheck(
         varColMap, std::get<IdTableSharedLocalVocabPair>(data_).idTable_));
   } else {
-    auto generator = [](Generator original,
+    auto generator = [](LazyResult original,
                         [[maybe_unused]] VariableToColumnMap varColMap,
                         [[maybe_unused]] auto performCheck) -> Generator {
       for (IdTableVocabPair& pair : original) {
@@ -212,7 +212,7 @@ void Result::runOnNewChunkComputed(
         onNewChunk,
     std::function<void(bool)> onGeneratorFinished) {
   AD_CONTRACT_CHECK(!isFullyMaterialized());
-  auto generator = [](Generator original, auto onNewChunk,
+  auto generator = [](LazyResult original, auto onNewChunk,
                       auto onGeneratorFinished) -> Generator {
     // Call this within destructor to make sure it is also called when an
     // operation stops iterating before reaching the end.
@@ -254,7 +254,7 @@ const IdTable& Result::idTable() const {
 }
 
 // _____________________________________________________________________________
-Result::Generator& Result::idTables() const {
+Result::LazyResult& Result::idTables() const {
   AD_CONTRACT_CHECK(!isFullyMaterialized());
   const auto& container = std::get<GenContainer>(data_);
   AD_CONTRACT_CHECK(!container.consumed_->exchange(true));

src/engine/Result.h

@@ -33,16 +33,23 @@ class Result {
         : idTable_{std::move(idTable)}, localVocab_{std::move(localVocab)} {}
   };
 
+  // The current implementation of (most of the) lazy results. Will be replaced
+  // in the future to make QLever compatible with C++17 again.
   using Generator = cppcoro::generator<IdTableVocabPair>;
+  // The lazy result type that is actually stored. It is type-erased and allows
+  // explicit conversion from the `Generator` above.
+  using LazyResult = ad_utility::InputRangeTypeErased<IdTableVocabPair>;
 
  private:
   // Needs to be mutable in order to be consumable from a const result.
   struct GenContainer {
-    mutable Generator generator_;
+    mutable LazyResult generator_;
     mutable std::unique_ptr<std::atomic_bool> consumed_ =
         std::make_unique<std::atomic_bool>(false);
-    explicit GenContainer(Generator generator)
+    explicit GenContainer(LazyResult generator)
         : generator_{std::move(generator)} {}
+    explicit GenContainer(Generator generator)
+        : generator_{Generator{std::move(generator)}} {}
   };
 
   using LocalVocabPtr = std::shared_ptr<const LocalVocab>;
@@ -155,7 +162,7 @@ class Result {
 
   // Access to the underlying `IdTable`s. Throw an `ad_utility::Exception`
   // if the underlying `data_` member holds the wrong variant.
-  Generator& idTables() const;
+  LazyResult& idTables() const;
 
   // Const access to the columns by which the `idTable()` is sorted.
   const std::vector<ColumnIndex>& sortedBy() const { return sortedBy_; }

src/engine/Service.cpp

@@ -564,8 +564,8 @@ void Service::precomputeSiblingResult(std::shared_ptr<Operation> left,
   // Creates a `Result::Generator` from partially materialized result data.
   auto partialResultGenerator =
       [](std::vector<Result::IdTableVocabPair> pairs,
-         Result::Generator prevGenerator,
-         Result::Generator::iterator it) -> Result::Generator {
+         Result::LazyResult prevGenerator,
+         std::ranges::iterator_t<Result::LazyResult> it) -> Result::Generator {
     for (auto& pair : pairs) {
       co_yield pair;
     }

src/parser/ParsedQuery.cpp

@@ -265,7 +265,7 @@ void ParsedQuery::GraphPattern::addLanguageFilter(const Variable& variable,
   std::vector<SparqlTriple*> matchingTriples;
   using BasicPattern = parsedQuery::BasicGraphPattern;
   namespace ad = ad_utility;
-  namespace stdv = std::views;
+  namespace stdv = ql::views;
   for (BasicPattern* basicPattern :
        _graphPatterns | stdv::transform(ad::getIf<BasicPattern>) |
            stdv::filter(ad::toBool)) {

src/util/BackgroundStxxlSorter.h

@@ -102,7 +102,7 @@ class BackgroundStxxlSorter {
   /// Transition from the input phase, where `push()` may be called, to the
   /// output phase and return a generator that yields the sorted elements. This
   /// function may be called exactly once.
-  [[nodiscard]] cppcoro::generator<value_type> sortedView() {
+  [[nodiscard]] auto sortedView() {
     setupSort();
     return bufferedAsyncView(outputGeneratorImpl(), _numElementsInRun);
   }

src/util/Generators.h

@@ -19,9 +19,10 @@ namespace ad_utility {
 // returns false. If the `aggregator` returns false, the cached value is
 // discarded. If the cached value is still present once the generator is fully
 // consumed, `onFullyCached` is called with the cached value.
-template <typename T>
+template <typename InputRange,
+          typename T = std::ranges::range_value_t<InputRange>>
 cppcoro::generator<T> wrapGeneratorWithCache(
-    cppcoro::generator<T> generator,
+    InputRange generator,
     InvocableWithExactReturnType<bool, std::optional<T>&, const T&> auto
         aggregator,
     InvocableWithExactReturnType<void, T> auto onFullyCached) {