Andrew Gierth <andrew(at)tao11(dot)riddles(dot)org(dot)uk> writes:
> I propose that it should not ignore rowMarks, per the attached patch or
> something similar.
+1 for not ignoring rowMarks, but this patch seems like a hack to me.
Why didn't you just add RowMarkClause as one of the known alternative
expression node types? There's no advantage to hard-wiring such
restrictive assumptions about where it can appear.
regards, tom lane