From: Jeff King <peff@peff.net>
Date: Fri, 29 Feb 2008 23:01:56 GMT
Subject: Re: [PATCH] t6024-recursive-merge.sh: hide spurious output when not running verbosely
Message-ID: <20080229230156.GA8742@coredump.intra.peff.net>
In-Reply-To: <20080229225816.GA24892@glandium.org>

On Fri, Feb 29, 2008 at 11:58:16PM +0100, Mike Hommey wrote:

> > > -git reset --hard
> > > +git reset --hard >&3 2>&4
> > >  test_expect_success 'refuse to merge binary files' '
> > >  	printf "\0" > binary-file &&
> > >  	git add binary-file &&
> > 
> > Should this perhaps just be:
> > 
> >   test_expect_success 'reset state' 'git reset --hard'
> 
> Is it really about testing git reset ?

No, of course not. Nor is t0003 about testing 'setup'. But it is a
convention[1] in git tests to do everything inside a test_expect_success
wrapper, which is beneficial because:

  - you get your output handled properly automagically :)
  - if the setup step fails, we notice. Which means:
     - it cannot silently cause a later test to succeed (e.g., setup
       changes a file, then the real test reverts it)
     - it is easier to debug failing tests, because you are notified
       of the _first_ failure

-Peff

[1] By convention, I don't necessarily mean that there is a written
rule, nor that we always do it that way, but only that in my experience
we usually do. Somebody like Junio could say more definitely what is
preferred.