>>>>> "Tom" == Tom Lane <tgl@sss.pgh.pa.us> writes:
>> Here is a draft patch along those lines; the intent of this one is
>> that no existing walker or mutator should need to change (the change
>> to the dependency code is basically cosmetic I believe, just avoids
>> walking some things twice).
Tom> Hmm. I think this is a reasonable direction to go in, but
Tom> what about groupingSets and rowMarks?
groupingSets ultimately contains nothing but numbers which are
meaningless without reference to the matching groupClause list. So
anything that cares about those is really going to have to process them
in its Query case in the walker function in order to get at both
clauses.
Similarly, rowMarks contains indexes into the rangetable (and no
recursive substructure at all), so it's likewise better processed at the
Query level.
Tom> Also, in HEAD I'd be inclined to add assertions about utilityStmt
Tom> being NULL.
Yup.
--
Andrew (irc:RhodiumToad)