>>>>> "Andrew" == Andrew Gierth <andrew(at)tao11(dot)riddles(dot)org(dot)uk> writes:
Andrew> It seems possible that this could be fixed by simply setting
Andrew> argisrow=false for all the null tests generated in such cases.
This turns out to be more somplicated than I thought. A lot of places
look at argisrow to distinguish "simple" null tests from the
standard-required logic for composite values. In some of the cases I've
looked at, fixing the bug actually looks like it improves things (for
example, in (row(1,a) IS NOT NULL), a can be deduced to be not null in
the "not the null value" sense); but some places like
match_clause_to_indexcol I'm less sure of.
--
Andrew (irc:RhodiumToad)