Re: t9001 failures on 32-bit Linux?

From: Junio C Hamano <gitster@pobox.com>
Date: Dec 10, 2017, 22:28 UTC
Message-ID: <xmqqfu8i5ibs.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <42367c9f-debb-8ea2-e1ea-1ca513853ded@ramsayjones.plus.com>

Ramsay Jones <ramsay@ramsayjones.plus.com> writes:

Show 17 quoted lines

> Hi Junio,
>
> I noticed the revert of the 'ab/simplify-perl-makefile' branch on
> top of 'pu'. So, I fired up my 32-bit Linux and attempted to see
> if I could debug this t9001 test failure.
>
> Unfortunately, I could not get it to fail. :(
>
> Both of the 'pu' (@77e921d77d) and 'pu~1' (@cfef1ebefd) builds pass
> the full test-suite and, likewise, running t9001 in a loop for a
> count of 100 (about 45 minutes run-time each).
>
> [If it makes any difference, I don't have sendmail installed (no
> /usr/sbin/sendmail or /usr/lib/sendmail), or any sendmail in my
> $PATH.]
>
> Sorry I couldn't help with this! :(

I suspect that the difference is not about send(e)mail per-se, but the differences between the set of perl modules installed in the linux32 Travis environment and the other one. Perhaps the use of deprecated Error.pm is the culprit?

I do appreciate that the environments we use on Travis are not monoculture and we end up testing more variations (in this case, the cause likely does not have anything to do with 64- vs 32-bit, but Perl modules). It sometimes gets confusing, though ;-)

Another difference I noticed while scanning the failing log is that the Linux32 environment seem to lack libsvn-perl and all tests are skipped there. I do not think that this particular variation is helping us to diversify our test base, though ;-)

Thanks.