postgresql(dot)org(at)tgice(dot)com writes:
> I would submit that in that situation, it would be
> reasonable for a user to expect my suggested syntax to still use the
> indicated indexes.
The only thing that will make that work is if "indexed_col IS NULL" were
an indexable condition, which it isn't because the PG index API only
supports "indexed_col operator something" as an indexable condition
(IS NULL is not an operator, and even if it were, there's no "something"
on its righthand side). Fixing this has been on the radar screen for
awhile, but it's not done, largely for lack of agreement about a
reasonably clean way to change that API.
regards, tom lane