Re: [PATCH 21/25] t5000-t5999: fix broken &&-chains

[Jeff King <peff@peff.net>]

On Thu, Jul 12, 2018 at 01:44:49PM -0400, Eric Sunshine wrote:

> On Thu, Jul 12, 2018 at 8:37 AM SZEDER Gábor <szeder.dev@gmail.com> wrote:
> > The change below should be squashed into this patch to fix a
> > previously unnoticed broken &&-chain.  I think you missed it, because
> > this test script is rather expensive and you didn't run it with
> > GIT_TEST_CLONE_2GB=YesPlease.
> >
> > diff --git a/t/t5608-clone-2gb.sh b/t/t5608-clone-2gb.sh
> > @@ -23,7 +23,7 @@ test_expect_success CLONE_2GB 'setup' '
> > -               echo "M 100644 :$i $i" >> commit
> > +               echo "M 100644 :$i $i" >> commit &&
> 
> Thanks for finding this. I tried to get as much coverage as possible
> by installing packages I don't normally have installed (Apache, cvs,
> cvsps, Subversion, Perforce, etc.) and even temporarily modified a
> script or two to force it run when I simply couldn't meet some
> prerequisite, thus reducing the "skipped" messages to a minimum, but I
> wasn't even aware of this prerequisite since I never saw a "skipped"
> message for it.

I think in theory we should be able to lint _every_ test, even if we're
not running it. After all, the point is that a proper linting runs zero
commands.

That said, it may not be worth the implementation effort. The linting
happens at a pretty low-level, and we've already decided to skip tests
long before then (even for single prereqs, let alone skip_all cases
where we exit the script early).

> Looking at it more closely, I think the reason it didn't come to my
> attention is that this script doesn't use the standard skip_all="..."
> mechanism for skipping the tests but instead "rolls its own", and
> apparently 'prove' simply swallowed (or was unable to produce) an
> overall "skipped" message for this script.

Yeah, that is a bit funny. For a whole-test skip like this, I think
skip_all is easier to read, as it is immediately apparent to the reader
that nothing that _doesn't_ meet that prereq should be in the file. So
I'd be happy to see it switched (though it's not _that_ big a deal, so
leaving it is fine, too).

By the way, "prove --directives" can help with finding individual
skipped tests.

-Peff