Git's tests have depended on Perl since at least 2006

This is just an interesting datapoint, but Git's test suite currently crashes and burns if you don't have a perl in your $PATH, no matter if you've compiled with NO_PERL=YesPlease or not. This has been the case since at least 2006, or v1.3.0-rc1~13^2~34.

The affected tests are:

    ./t0000-basic.sh ./t0025-crlf-auto.sh ./t0080-vcs-svn.sh
    ./t0020-crlf.sh ./t1010-mktree.sh ./t1300-repo-config.sh
    ./t3300-funny-names.sh ./t4012-diff-binary.sh
    ./t4020-diff-external.sh ./t4014-format-patch.sh
    ./t4029-diff-trailing-space.sh ./t4031-diff-rewrite-binary.sh
    ./t4030-diff-textconv.sh ./t4043-diff-rename-binary.sh
    ./t4103-apply-binary.sh ./t4116-apply-reverse.sh ./t4200-rerere.sh
    ./t5303-pack-corruption-resilience.sh ./t5300-pack-object.sh
    ./t6011-rev-list-with-bad-commit.sh
    ./t6013-rev-list-reverse-parents.sh ./t6003-rev-list-topo-order.sh

I have a WIP patch series to tackle this issue:

    http://github.com/avar/git/compare/git:pu...in-progress-dont-depend-on-perl

It adds a PERL dependency to things and changes perl to "$PERL_PATH". Many things were implicitly depending on perl due to this bit in test-lib.sh:

    q_to_nul () {
        perl -pe 'y/Q/\000/'
    }

And many other bits do other things with perl but aren't declaring a PERL dependency.

Anyway, I just thought I'd mention it, that this hasn't been fixed yet would suggest that nearly everyone has perl anyway, and that maybe we shouldn't think twice about introducing a perl dependency to things when it's convenient.

Ævar Arnfjörð Bjarmason wrote:

Yes, I think this was known. Development machines (which is where people tend to run tests most) tend to have perl but some weird production machines did not.

Reasonable.

Is that necessary? As you mentioned before, a lot of uses of perl are for run-of-the-mill one-liners and do not care much which version they get.

On the other hand, even if it is not necessary, it might still be a good idea to just stick to one perl version for simplicity.

Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:

I'd suggest you not to go overboard on this. NO_PERL=YesPlease is primarily about the deployment environment. The person who is building and packaging git may have perl on her development box.

On Fri, Aug 13, 2010 at 22:45, Junio C Hamano <gitster@pobox.com> wrote:

I promise not to go overboard :)

The aim of this and some recent patch serieses has been to improve our smoker support. I'd like people to be able to run a smoker without a perl without sending false positives in their smoke reports, that's all.

Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:

I had an impression that your smoker report supports builds on tap harness that in turn builds on perl.

On Fri, Aug 13, 2010 at 23:30, Junio C Hamano <gitster@pobox.com> wrote:

It does, but I was considering cases like:

    PATH=/a/limited/test/environment /usr/bin/prove ...

But actually getting rid of the smoker Perl dependency would be easy, it's just running each test, saving the output to a file and tar-ing it all up. It'd be very easy to provide a shellscript that did that.