Re: t7610-mergetool.sh test failure

From: Jeff King <peff@peff.net>
Date: Jun 22, 2016, 17:12 UTC
Message-ID: <20160622171202.GB10056@sigill.intra.peff.net>
In-Reply-To: <CALR6jEisU=ZCZADcTx685vSrz96jUaxfAbKMEPQn0KRFqtVw=A@mail.gmail.com>

On Wed, Jun 22, 2016 at 06:53:40PM +0200, Armin Kunaschik wrote:

Show 13 quoted lines

> Another thread I'm trying to revive... the discussion went away quite a bit
> from the suggested patch to conditionally run this one test only when mktemp
> is available.
> 
> I'll create a patch when there are chances it is accepted.
> 
> I could think of a way to replace mktemp with a perl one-liner (or
> small script)...
> conditionally, when mktemp is not available... maybe in the build process?
> As far as I can see, perl is absolutely necessary and can therefore be used to
> "solve" the mktemp problem...
> 
> ...or maybe I should stop bringing this up again :-)

I think perl is necessary for the test suite, but not for git-mergetool itself. And this is a problem in the script itself, IIRC; so it really is broken on your system (albeit in a really tiny way), and not just a test portability thing.

So the viable solutions to me are one of:

  1. Accept that this little part of mergetool doesn't work on systems
     without "mktemp", make sure we fail gracefully (we seem to), and make
     sure the test suite can handle this case (which was the earlier
     patch, I think).

  2. Implement our own git-mktemp (or similar abstraction) in C. We
     already have all the code, so it really is just a thin wrapper.

I don't mind (2), but given the lack of people clamoring for a fix to mergetool itself, I'm perfectly happy with (1).

-Peff