schemachanger: create index should fail storing column that is already in pk about cockroach

[spilchen]

I think the DSC check here needs a small tweak:

cockroach/pkg/sql/schemachanger/scbuild/internal/scbuildstmt/create_index.go

Lines 545 to 552 in 7346b9b

	for _, storingNode := range n.Storing {
	colName := storingNode.String()
	if _, found := columnRefs[colName]; found {
	panic(pgerror.Newf(pgcode.InvalidObjectDefinition,
	"index %q already contains column %q", n.Name, colName))
	}
	columnRefs[colName] = struct{}{}
	}

If I take the original repro and rename the column "col\u000b7ͪ%q_w0_10" to "col7_w0_10", removing any UTF elements, the check works as expected. I believe the code needs to account for UTF characters. Maybe using something like colName := storingNode.Normalize() would help.

Another aspect of this issue is that with the DSC, we seem to allow virtual columns in the STORING clause. I think this should be blocked to be consistent with the legacy schema changer.

[annrpom]

We should block store + PK in the DSC like we do in the legacy

[rafiss]

The check should be made in the builder, before the schema change begins executing.

[rafiss]

The legacy schema changer has a check explicitly for the primary index:

cockroach/pkg/sql/catalog/tabledesc/structured.go

Lines 732 to 740 in d1b3921

	if primaryColIDs.Contains(col.GetID()) && idx.GetEncodingType() == catenumpb.SecondaryIndexEncoding {
	// If the primary index contains a stored column, we don't need to
	// store it - it's already part of the index.
	err = pgerror.Newf(pgcode.DuplicateColumn,
	"index %q already contains column %q", idx.GetName(), col.GetName())
	err = errors.WithDetailf(err,
	"column %q is part of the primary index and therefore implicit in all indexes", col.GetName())
	return err
	}

A similar check is needed in the declarative schema changer, around here:

cockroach/pkg/sql/schemachanger/scbuild/internal/scbuildstmt/create_index.go

Lines 545 to 552 in 7346b9b

	for _, storingNode := range n.Storing {
	colName := storingNode.String()
	if _, found := columnRefs[colName]; found {
	panic(pgerror.Newf(pgcode.InvalidObjectDefinition,
	"index %q already contains column %q", n.Name, colName))
	}
	columnRefs[colName] = struct{}{}
	}