Git v2.21.0-rc0 broken on *BSD, maybe others

["Ævar Arnfjörð Bjarmason" <avarab@gmail.com>]

On Thu, Feb 07 2019, Junio C Hamano wrote:

> An early preview release Git v2.21.0-rc0 is now available for
> testing at the usual places.  It is comprised of 426 non-merge
> commits since v2.20.0, contributed by 57 people, 13 of which are
> new faces.

As seen on https://gitlab.com/git-vcs/git-ci/branches there are
regressions since 2.19.0, e.g. we have hard compile errors on BSD now,
and maybe AIX. I haven't dug into it.

I probably won't have the time for this myself before 2.21, but am
sending this message in case others are interested. I don't expect these
to be serious issues, just portability nits that'll either be fixed
before 2.21, or alternatively by people packaging for those
platforms. If they don't care enough to pursue it pro-actively I'm not
very inclined to spend time on it myself...

Johannes: We didn't have a chance to finish this conversation up at Git
Merge, but the useful bit of CI platforms basically devolves into
test_for(SHA1, params), with the hard part driving it so it's done
constantly, N branches are tested with smart notices of "XYZ broken in A
but not B" etc.

My impression is that you're *way* ahead with the Azure stuff for
solving those hard parts. Is there any way we could ditch most of this
ad-hoc GitLab CI setup I have on the GCC compile farm, and make it be
driven by what you have, and at some point it would just do something
like:

    ssh <some compile farm host> "compile-and-test.sh SHA1 <params>"

Then we could all display it in the same UI, and get feedback on "pu"
topics before they land et.c