Re: [PERFORM] Feature request: smarter use of

On 3/6/04 11:07 PM, Tom Lane wrote:
> Neil Conway <neilc(at)samurai(dot)com> writes:
>> Uh, you get test failures on OS X 10.3.2? On my OS X machine with the
>> latest CVS HEAD code, the tests run without any failures.
>
> I was wondering about that too. I get one failure --- a one-liner
> zero-versus-minus-zero difference in geometry.out. Anything more seems
> cause for suspicion.

I know, I get that same 1 failure in 7.4.1 on my Mac at work which is
(seemingly) identical to my Mac at home. I just deleted my CVS repository
and then re-checked-out the whole thing, rebuilt it (no problems there, as
usual), and ran make check again. This time I got 9 failures!

When I got the 5 failures with an unmodified CVS checkout earlier, I didn't
even bother to look at the diffs because I just accepted that as my baseline
(it is CVS, after all, and things are in flux). But this time, a quick look
at the diffs showed lines like this in the regression diffs:

! psql: could not fork new process for connection: Resource temporarily
unavailable

I then recalled that my Mac at work has all sorts of kernel params and
resource limits tweaked to make Postgres happier. Just running "unlimit" in
my shell was all I needed to do to make "make check" work at home, and I got
the same old single geometry test failure (-0 vs. 0) both with and without
the patch. Yay.

-John