fix(cudf): Guard hash join build debug logging against empty inputs (#17221)

Summary:
When the build side of a hash join has no data (e.g. anti-join with
empty result), inputs_ is empty. With a debug build, the logging at
`noMoreInput()` accesses `inputs_[0]` without bound checking,
which could result in a `SIGSEGV`.

This fixes Q16 (TPC-H) which uses NOT IN (anti-join) where the build
side can be empty for some partitions.

Pull Request resolved: #17221

Reviewed By: kagamiori

Differential Revision: D101671299

Pulled By: peterenescu

fbshipit-source-id: 0971d85671e67810789d8148ce3891e757faf4f7

Author: pramodsatya

velox/experimental/cudf/exec/CudfHashJoin.cpp

@@ -204,9 +204,11 @@ void CudfHashJoinBuild::doNoMoreInput() {
   };
 
   if (CudfConfig::getInstance().debugEnabled) {
-    VLOG(1) << "CudfHashJoinBuild: build batches";
-    VLOG(1) << "Build batches number of columns: "
-            << inputs_[0]->getTableView().num_columns();
+    VLOG(1) << "CudfHashJoinBuild: build batches count: " << inputs_.size();
+    if (!inputs_.empty()) {
+      VLOG(1) << "Build batches number of columns: "
+              << inputs_[0]->getTableView().num_columns();
+    }
     for (auto i = 0; i < inputs_.size(); i++) {
       VLOG(1) << "Build batch " << i
               << ": number of rows: " << inputs_[i]->getTableView().num_rows();
@@ -224,7 +226,7 @@ void CudfHashJoinBuild::doNoMoreInput() {
   for (auto const& tbl : tbls) {
     VELOX_CHECK_NOT_NULL(tbl);
   }
-  if (CudfConfig::getInstance().debugEnabled) {
+  if (CudfConfig::getInstance().debugEnabled && !tbls.empty()) {
     VLOG(1) << "Build table number of columns: " << tbls[0]->num_columns();
     for (auto i = 0; i < tbls.size(); i++) {
       VLOG(1) << "Build table " << i

velox/experimental/cudf/tests/HashJoinTest.cpp

@@ -8493,4 +8493,22 @@ DEBUG_ONLY_TEST_F(HashJoinTest, hashTableCleanupAfterProbeFinish) {
   ASSERT_TRUE(tableEmpty);
 }
 
+TEST_F(HashJoinTest, emptyBuildWithDebugEnabled) {
+  cudf_velox::CudfConfig::getInstance().debugEnabled = true;
+  SCOPE_EXIT {
+    cudf_velox::CudfConfig::getInstance().debugEnabled = false;
+  };
+
+  HashJoinBuilder(*pool_, duckDbQueryRunner_, driverExecutor_.get())
+      .injectSpill(false)
+      .numDrivers(1)
+      .keyTypes({BIGINT()})
+      .probeVectors(100, 5)
+      .buildVectors(0, 5)
+      .referenceQuery(
+          "SELECT t_k0, t_data, u_k0, u_data FROM t, u WHERE t_k0 = u_k0")
+      .checkSpillStats(false)
+      .run();
+}
+
 } // namespace