greg@turnstep.com writes:
> Some very minimal checking could be put into place, but it would just be
> a small subset of the full checking that occurs later on, so it seemed
> better to leave all that logic in one place.
Then why not move the rangechecks on month to the full check code too?
(It sure looks like you've lost the defense against month=0, btw.)
BTW, it seems to me that part of the original thread about this issue
included a demonstration that Postgres would sometimes take out-of-range
dates. That may have just been with respect to to_date() ... but are
you sure there are no other paths for parsing a date spec in the main
date/time code?
regards, tom lane