Disallow physical index scans in Orca to have predicate condition with subquery

Physical operators should not have subqueries, by this point they should
have been converted into correlated joins or expressions. While we would
consider these invalid plans this wouldn't ever be executed, we assumed
in one case that we could retrieve an exact expression from a group that
contained a subquery when deriving the distribution spec, which caused a
crash.

Now, we'll be more defensive and disallow creating physical index scans
if the predicate contains a subquery. This isn't a problem for generic
gets, as they don't have a scalar predicate.

src/backend/gporca/libgpopt/src/xforms/CXformDynamicIndexGet2DynamicIndexScan.cpp

@@ -66,6 +66,14 @@ CXformDynamicIndexGet2DynamicIndexScan::Transform(CXformContext *pxfctxt,
 	// create/extract components for alternative
 	CName *pname = GPOS_NEW(mp) CName(mp, popIndexGet->Name());
 
+	// extract components
+	CExpression *pexprIndexCond = (*pexpr)[0];
+	if (pexprIndexCond->DeriveHasSubquery())
+	{
+		return;
+	}
+	pexprIndexCond->AddRef();
+
 	CTableDescriptor *ptabdesc = popIndexGet->Ptabdesc();
 	ptabdesc->AddRef();
 
@@ -88,10 +96,6 @@ CXformDynamicIndexGet2DynamicIndexScan::Transform(CXformContext *pxfctxt,
 	COrderSpec *pos = popIndexGet->Pos();
 	pos->AddRef();
 
-	// extract components
-	CExpression *pexprIndexCond = (*pexpr)[0];
-	pexprIndexCond->AddRef();
-
 	// create alternative expression
 	CExpression *pexprAlt = GPOS_NEW(mp) CExpression(
 		mp,

src/backend/gporca/libgpopt/src/xforms/CXformIndexGet2IndexOnlyScan.cpp

@@ -86,6 +86,13 @@ CXformIndexGet2IndexOnlyScan::Transform(CXformContext *pxfctxt,
 	CIndexDescriptor *pindexdesc = pop->Pindexdesc();
 	CTableDescriptor *ptabdesc = pop->Ptabdesc();
 
+	// extract components
+	CExpression *pexprIndexCond = (*pexpr)[0];
+	if (pexprIndexCond->DeriveHasSubquery())
+	{
+		return;
+	}
+
 	CMDAccessor *md_accessor = COptCtxt::PoctxtFromTLS()->Pmda();
 	const IMDRelation *pmdrel = md_accessor->RetrieveRel(ptabdesc->MDId());
 	const IMDIndex *pmdindex = md_accessor->RetrieveIndex(pindexdesc->MDId());
@@ -127,8 +134,7 @@ CXformIndexGet2IndexOnlyScan::Transform(CXformContext *pxfctxt,
 	GPOS_ASSERT(NULL != pos);
 	pos->AddRef();
 
-	// extract components
-	CExpression *pexprIndexCond = (*pexpr)[0];
+
 
 	// addref all children
 	pexprIndexCond->AddRef();

src/backend/gporca/libgpopt/src/xforms/CXformIndexGet2IndexScan.cpp

@@ -80,6 +80,13 @@ CXformIndexGet2IndexScan::Transform(CXformContext *pxfctxt,
 	CIndexDescriptor *pindexdesc = pop->Pindexdesc();
 	CTableDescriptor *ptabdesc = pop->Ptabdesc();
 
+	// extract components
+	CExpression *pexprIndexCond = (*pexpr)[0];
+	if (pexprIndexCond->DeriveHasSubquery())
+	{
+		return;
+	}
+
 	pindexdesc->AddRef();
 	ptabdesc->AddRef();
 
@@ -91,9 +98,6 @@ CXformIndexGet2IndexScan::Transform(CXformContext *pxfctxt,
 	GPOS_ASSERT(NULL != pos);
 	pos->AddRef();
 
-	// extract components
-	CExpression *pexprIndexCond = (*pexpr)[0];
-
 	// addref all children
 	pexprIndexCond->AddRef();
 

src/backend/gporca/server/CMakeLists.txt

@@ -324,6 +324,7 @@ Distinct-LegacyOpfamily;
 
 CSubquery2Test:
 Subq2PartialDecorrelate Subq2CorrSQInLOJOn Subq2NotInWhereLOJ Subq2OuterRef2InJoin Subq2OuterRefMultiLevelInOn
+Index-Join-With-Subquery-In-Pred
 ")
 
 set(mdp_dir "../data/dxl/minidump/")