Re: [PATCH 2/3] test: improve rebase -q test

From: SZEDER Gábor <szeder@ira.uka.de>
Date: Jun 10, 2013, 20:19 UTC
Message-ID: <20130610201933.GI2091@goldbirke>
In-Reply-To: <51B623DA.4050802@kdbg.org>

On Mon, Jun 10, 2013 at 09:07:06PM +0200, Johannes Sixt wrote:

Show 10 quoted lines

> Am 10.06.2013 19:27, schrieb SZEDER Gábor:
> > My main motivation is that, like in your example, in the bash prompt
> > tests I only have to check a single line of output, but because of
> > debuggability I always did:
> > 
> >   echo "(master)" >expected
> >   __git_ps1 >actual
> >   test_cmp expected actual
> 
> Chained by &&, I presume.

Sure.

Show 9 quoted lines

> > With such a helper function this could be reduced to a single line:
> > 
> >   test_string_equal "(master)" "$(__git_ps1)"
> > 
> > without loss of functionality
> 
> Not quite: A non-zero exit code of the $(__git_ps1) is lost. (It
> probably doesn't matter here, but it certainly does if the command is
> $(git rev-parse foo) or similar.)

Ouch, indeed. Yeah, the exit code doesn't matter for the prompt tests (I mean for __git_ps1() tests, but maybe it does matter for some __gitdir() tests), but I suppose it does matter everywhere else where the same construct is used. We could still do

  actual="$(git foo)" &&
  test_string_equal "good" "$actual"

to preserve and check the exit code, and this is still one line shorter, but overall not that appealing anymore.

However. The git command's exit code is lost the same way in 'test good = $(git foo)' constructs, too, and plenty of such constructs are all over the test suite. Shouldn't we avoid using such constucts then?

Gábor