Fix bug in join algorithm in case of many empty blocks

src/engine/Filter.cpp

@@ -75,7 +75,9 @@ ProtoResult Filter::computeResult(bool requestLaziness) {
               for (auto& [idTable, localVocab] : subRes->idTables()) {
                 IdTable result = self->filterIdTable(subRes->sortedBy(),
                                                      idTable, localVocab);
-                co_yield {std::move(result), std::move(localVocab)};
+                if (!result.empty()) {
+                  co_yield {std::move(result), std::move(localVocab)};
+                }
               }
             }(std::move(subRes), this),
             resultSortedOn()};

src/util/JoinAlgorithms/JoinAlgorithms.h

@@ -801,6 +801,10 @@ struct BlockZipperJoinImpl {
     for (size_t numBlocksRead = 0; it != end && numBlocksRead < 3;
          ++it, ++numBlocksRead) {
       if (ql::ranges::empty(*it)) {
+        // We haven't read a block, so we have to adapt the counter.
+        // NOTE: If `numBlocksRead == 0`, the underflow is no problem because
+        // the next operation is `++`.
+        --numBlocksRead;
         continue;
       }
       if (!eq((*it)[0], currentEl)) {

test/FilterTest.cpp

@@ -87,13 +87,11 @@ TEST(Filter, verifyPredicateIsAppliedCorrectlyOnLazyEvaluation) {
   auto referenceTable1 =
       makeIdTableFromVector({{true}, {true}, {true}}, asBool);
   auto referenceTable2 = makeIdTableFromVector({{true}}, asBool);
-  IdTable referenceTable3{0, ad_utility::makeUnlimitedAllocator<Id>()};
 
   auto m = matchesIdTable;
   EXPECT_THAT(
       toVector(std::move(generator)),
-      ElementsAre(m(referenceTable1), m(referenceTable2), m(referenceTable3),
-                  m(referenceTable3), m(referenceTable2)));
+      ElementsAre(m(referenceTable1), m(referenceTable2), m(referenceTable2)));
 }
 
 // _____________________________________________________________________________

test/JoinAlgorithmsTest.cpp

@@ -182,6 +182,18 @@ TEST(JoinAlgorithms, JoinWithBlocksExactlyFourBlocksPerElement) {
   testJoin(a, b, expectedResult);
 }
 
+// Test the handling of many empty blocks.
+TEST(JoinAlgorithms, JoinWithEmptyBlocks) {
+  NestedBlock a{{{42, 1}, {42, 2}}, {{42, 3}}};
+  // The join has to handle all the entries with `42` in the first column at
+  // once. In `b` there are more than 3 empty blocks between blocks with this
+  // entry. There was previously a bug in this case.
+  NestedBlock b{{{42, 16}}, {}, {}, {}, {}, {}, {}, {}, {}, {{42, 23}}};
+  JoinResult expectedResult{{42, 1, 16}, {42, 1, 23}, {42, 2, 16},
+                            {42, 2, 23}, {42, 3, 16}, {42, 3, 23}};
+  testJoin(a, b, expectedResult);
+}
+
 // ________________________________________________________________________________________
 TEST(JoinAlgorithms, JoinWithBlocksMultipleBlocksPerElementBothSides) {
   NestedBlock a{{{42, 0}}, {{42, 1}, {42, 2}}, {{42, 3}, {67, 0}}};