On Jul 12, 2008, at 14:57, Tom Lane wrote:
> 4. A lot of the later test cases are equally uselessly testing whether
> piggybacking over text functions works. The odds of ever finding
> anything with those tests are not distinguishable from zero (unless
> the
> underlying text function is busted, which is not your responsibility
> to
> test). So I don't see any point in putting them into the standard
> regression package. (What maybe *would* be useful to test, but you
> didn't, is whether the result of a function is considered citext
> rather
> than text.)
I'd like to keep these tests, since they ensure not just that the
functions work but that they work with citext. Given what we found
with length() and friends not working when there was an implicit cast
to bpchar, I think it's valuable to continue to ensure that these
functions work as expected with citext. Otherwise someone in the
future might come along and make the cast to bpchar implicit again,
and no tests would fail to tell him/her otherwise.
These tests make good regressions.
Thanks,
David