Re: [PATCH v2 2/5] lib-gpg: allow tests with GPGSM or GPGSSH prereq first

From: Todd Zullinger <tmz@pobox.com>
Date: Oct 11, 2025, 02:14 UTC
Message-ID: <aOm9fLW-8_oJQZy9@teonanacatl.net>
In-Reply-To: <xmqqbjmeafqm.fsf@gitster.g>

Junio C Hamano wrote:

Show 14 quoted lines

> Todd Zullinger <tmz@pobox.com> writes:
> 
>> I agree.  But it is ugly that any tests we have which rely
>> on the GPG2 prereq simply never run.  That should be fixed
>> and, if it were me, I'd do so by dropping the flaky tests in
>> t1016 initially.  Someone who cares about those tests
>> running could debug it more and hopefully fix the problem.
> 
> Let me queue your two patches as-is to leave the tip of 'seen'
> broken for a few days to see if anybody bites ;-).  After that, we
> may do "s|test_expect_success|test_expect_failure|" on those tests
> that you call "flaky".  Are they flaky in the sense that they
> sometimes pass sometimes fail depending on the timing, or just
> simply buggy and always fail?

In my recollection, they fail all (or nearly all?) of the time in our CI runs and when I was building git for Fedora infrastructure, they failed consistently on the Fedora builders as well.

They fail rarely (if ever) when I run them locally, even with --stress options. That made it rather difficult to work out the issue. I thought that it was a timing problem for a while, but I wasn't able to find a way to demonstrate that.

Thanks for the willingness to suffer some test breakage to see if it can flush out a fix. :)

I suspect there are folks here who know the test suite and code being tested well enough that it may be really obvious to them. Whether there is an intersection of those folks and spare "round tuits" is another matter.

-- 
Todd