Fix the extract values function sigature when used with filter proces…

…s in selective reader (#10956) Summary: Pull Request resolved: #10956 The processFilter depends on the exact value callback type match to detect if the filter to drop value or not. If not drop value, it will add value to the result buffer of the selective reader. For null filter type, it add nulls to the null result buffer of the selective reader. The recent refactor changes the extract value callback from T to const T& which breaks the comparison. It tries to add value to the null result buffer which is not initialized in the prepare read of the selective reader as the reader is a filter only reader. This PR reverts the signature change and verified with both failed query shadow and a new unit test to repro this. We will refactor this later to avoid a copy extract value callback later Reviewed By: Yuhta Differential Revision: D62394606 fbshipit-source-id: 5137e2c9c2f935cca1c08a16d42c8d392e0e7827
facebookincubator · Sep 10, 2024 · ef251d4 · ef251d4
1 parent 55931e7
commit ef251d4
Show file tree

Hide file tree

Showing 6 changed files with 65 additions and 22 deletions.
diff --git a/velox/dwio/common/ColumnVisitors.h b/velox/dwio/common/ColumnVisitors.h
@@ -162,7 +162,7 @@ class ColumnVisitor {
       TFilter& filter,
       SelectiveColumnReader* reader,
       const RowSet& rows,
-      const ExtractValues& values)
+      ExtractValues values)
       : filter_(filter),
         reader_(reader),
         allowNulls_(!TFilter::deterministic || filter.testNull()),
@@ -719,7 +719,7 @@ class DictionaryColumnVisitor
       TFilter& filter,
       SelectiveColumnReader* reader,
       const RowSet& rows,
-      const ExtractValues& values)
+      ExtractValues values)
       : ColumnVisitor<T, TFilter, ExtractValues, isDense>(
             filter,
             reader,

diff --git a/velox/dwio/common/SelectiveByteRleColumnReader.h b/velox/dwio/common/SelectiveByteRleColumnReader.h
@@ -60,7 +60,7 @@ class SelectiveByteRleColumnReader : public SelectiveColumnReader {
   void readHelper(
       velox::common::Filter* filter,
       const RowSet& rows,
-      const ExtractValues& extractValues);
+      ExtractValues extractValues);
 
   template <typename Reader, bool kEncodingHasNulls>
   void readCommon(
@@ -77,7 +77,7 @@ template <
 void SelectiveByteRleColumnReader::readHelper(
     velox::common::Filter* filter,
     const RowSet& rows,
-    const ExtractValues& extractValues) {
+    ExtractValues extractValues) {
   reinterpret_cast<Reader*>(this)->readWithVisitor(
       rows,
       ColumnVisitor<int8_t, TFilter, ExtractValues, isDense>(

diff --git a/velox/dwio/common/SelectiveFloatingPointColumnReader.h b/velox/dwio/common/SelectiveFloatingPointColumnReader.h
@@ -56,8 +56,10 @@ class SelectiveFloatingPointColumnReader : public SelectiveColumnReader {
       typename TFilter,
       bool isDense,
       typename ExtractValues>
-  void
-  readHelper(velox::common::Filter* filter, RowSet rows, ExtractValues values);
+  void readHelper(
+      velox::common::Filter* filter,
+      const RowSet& rows,
+      ExtractValues values);
 
   template <
       typename Reader,
@@ -66,11 +68,11 @@ class SelectiveFloatingPointColumnReader : public SelectiveColumnReader {
       typename ExtractValues>
   void processFilter(
       velox::common::Filter* filter,
-      RowSet rows,
+      const RowSet& rows,
       ExtractValues extractValues);
 
   template <typename Reader, bool isDense>
-  void processValueHook(RowSet rows, ValueHook* hook);
+  void processValueHook(const RowSet& rows, ValueHook* hook);
 };
 
 template <typename TData, typename TRequested>
@@ -81,7 +83,7 @@ template <
     typename ExtractValues>
 void SelectiveFloatingPointColumnReader<TData, TRequested>::readHelper(
     velox::common::Filter* filter,
-    RowSet rows,
+    const RowSet& rows,
     ExtractValues extractValues) {
   reinterpret_cast<Reader*>(this)->readWithVisitor(
       rows,
@@ -97,7 +99,7 @@ template <
     typename ExtractValues>
 void SelectiveFloatingPointColumnReader<TData, TRequested>::processFilter(
     velox::common::Filter* filter,
-    RowSet rows,
+    const RowSet& rows,
     ExtractValues extractValues) {
   if (filter == nullptr) {
     readHelper<Reader, velox::common::AlwaysTrue, isDense>(
@@ -144,7 +146,7 @@ void SelectiveFloatingPointColumnReader<TData, TRequested>::processFilter(
 template <typename TData, typename TRequested>
 template <typename Reader, bool isDense>
 void SelectiveFloatingPointColumnReader<TData, TRequested>::processValueHook(
-    RowSet rows,
+    const RowSet& rows,
     ValueHook* hook) {
   switch (hook->kind()) {
     case aggregate::AggregationHook::kDoubleSum:

diff --git a/velox/dwio/common/SelectiveIntegerColumnReader.h b/velox/dwio/common/SelectiveIntegerColumnReader.h
@@ -48,7 +48,7 @@ class SelectiveIntegerColumnReader : public SelectiveColumnReader {
       typename ExtractValues>
   void processFilter(
       velox::common::Filter* filter,
-      const ExtractValues& extractValues,
+      ExtractValues extractValues,
       const RowSet& rows);
 
   // Switches based on the type of ValueHook between different readWithVisitor
@@ -65,7 +65,7 @@ class SelectiveIntegerColumnReader : public SelectiveColumnReader {
   void readHelper(
       velox::common::Filter* filter,
       const RowSet& rows,
-      const ExtractValues& extractValues);
+      ExtractValues extractValues);
 
   // The common part of integer reading. calls the appropriate
   // instantiation of processValueHook or processFilter based on
@@ -82,7 +82,7 @@ template <
 void SelectiveIntegerColumnReader::readHelper(
     velox::common::Filter* filter,
     const RowSet& rows,
-    const ExtractValues& extractValues) {
+    ExtractValues extractValues) {
   switch (valueSize_) {
     case 2:
       reinterpret_cast<Reader*>(this)->Reader::readWithVisitor(
@@ -124,7 +124,7 @@ template <
     typename ExtractValues>
 void SelectiveIntegerColumnReader::processFilter(
     velox::common::Filter* filter,
-    const ExtractValues& extractValues,
+    ExtractValues extractValues,
     const RowSet& rows) {
   if (filter == nullptr) {
     readHelper<Reader, velox::common::AlwaysTrue, isDense>(

diff --git a/velox/dwio/dwrf/reader/SelectiveStringDictionaryColumnReader.h b/velox/dwio/dwrf/reader/SelectiveStringDictionaryColumnReader.h
@@ -68,16 +68,14 @@ class SelectiveStringDictionaryColumnReader
   void readWithVisitor(const RowSet& rows, TVisitor visitor);
 
   template <typename TFilter, bool isDense, typename ExtractValues>
-  void readHelper(
-      common::Filter* filter,
-      const RowSet& rows,
-      const ExtractValues& values);
+  void
+  readHelper(common::Filter* filter, const RowSet& rows, ExtractValues values);
 
   template <bool isDense, typename ExtractValues>
   void processFilter(
       common::Filter* filter,
       const RowSet& rows,
-      const ExtractValues& extractValues);
+      ExtractValues extractValues);
 
   // Fills 'values' from 'data' and 'lengthDecoder'. The count of
   // values is in 'values.numValues'.
@@ -129,7 +127,7 @@ template <typename TFilter, bool isDense, typename ExtractValues>
 void SelectiveStringDictionaryColumnReader::readHelper(
     common::Filter* filter,
     const RowSet& rows,
-    const ExtractValues& values) {
+    ExtractValues values) {
   readWithVisitor(
       rows,
       dwio::common::
@@ -141,7 +139,7 @@ template <bool isDense, typename ExtractValues>
 void SelectiveStringDictionaryColumnReader::processFilter(
     common::Filter* filter,
     const RowSet& rows,
-    const ExtractValues& extractValues) {
+    ExtractValues extractValues) {
   if (filter == nullptr) {
     readHelper<common::AlwaysTrue, isDense>(
         &dwio::common::alwaysTrue(), rows, extractValues);

diff --git a/velox/exec/tests/TableScanTest.cpp b/velox/exec/tests/TableScanTest.cpp
@@ -3153,6 +3153,49 @@ TEST_F(TableScanTest, mapIsNullFilter) {
       "SELECT * FROM tmp WHERE c0 is null");
 }
 
+TEST_F(TableScanTest, stringIsNullFilter) {
+  constexpr int kSize = 1000;
+  const char* baseStrings[] = {
+      "qwertyuiopasdfghjklzxcvbnm",
+      "qazwsxedcrfvtgbyhnujmikolp",
+  };
+  auto indices = allocateIndices(kSize, pool_.get());
+  for (int i = 0; i < kSize; ++i) {
+    indices->asMutable<vector_size_t>()[i] = i % 2;
+  }
+  auto nullBuffer =
+      AlignedBuffer::allocate<bool>(kSize, pool_.get(), bits::kNotNull);
+  auto* rawNullBuffer = nullBuffer->asMutable<uint64_t>();
+  for (int i = 0; i < kSize; i += 100) {
+    bits::setNull(rawNullBuffer, i);
+  }
+  auto dict = BaseVector::wrapInDictionary(
+      nullBuffer,
+      indices,
+      kSize,
+      makeFlatVector<std::string>({baseStrings[0], baseStrings[1]}));
+
+  auto rows = makeRowVector({"c0", "c1"}, {dict, dict});
+  auto tableType = asRowType(rows->type());
+  auto file = TempFilePath::create();
+  writeToFile(file->getPath(), {rows});
+  createDuckDbTable({rows});
+
+  const auto outputType = ROW({"c1"}, {VARCHAR()});
+  auto makePlan = [&](const std::vector<std::string>& filters) {
+    return PlanBuilder()
+        .tableScan(outputType, filters, "", tableType)
+        .planNode();
+  };
+
+  assertQuery(
+      makePlan({"c0 is not null"}),
+      {file},
+      "SELECT c1 FROM tmp WHERE c0 is not null");
+  assertQuery(
+      makePlan({"c0 is null"}), {file}, "SELECT c1 FROM tmp WHERE c0 is null");
+}
+
 TEST_F(TableScanTest, compactComplexNulls) {
   constexpr int kSize = 10;
   auto iota = makeFlatVector<int64_t>(kSize, folly::identity);