Re: [PATCH v2 1/2] rev-parse tests: add tests executed from a subdirectory

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

Hi Junio,

On Fri, 10 Feb 2017, Junio C Hamano wrote:

> Johannes Schindelin <johannes.schindelin@gmx.de> writes:
> 
> > diff --git a/t/t1700-split-index.sh b/t/t1700-split-index.sh
> > index 292a0720fcc..1d6e27a09d8 100755
> > --- a/t/t1700-split-index.sh
> > +++ b/t/t1700-split-index.sh
> > @@ -200,4 +200,21 @@ EOF
> >  	test_cmp expect actual
> >  '
> >  
> > +test_expect_failure 'rev-parse --shared-index-path' '
> > +	rm -rf .git &&
> > +	test_create_repo . &&
> > +	git update-index --split-index &&
> > +	ls -t .git/sharedindex* | tail -n 1 >expect &&
> > +	git rev-parse --shared-index-path >actual &&
> > +	test_cmp expect actual &&
> > +	mkdir work &&
> > +	test_when_finished "rm -rf work" &&
> > +	(
> > +		cd work &&
> > +		ls -t ../.git/sharedindex* | tail -n 1 >expect &&
> > +		git rev-parse --shared-index-path >actual &&
> > +		test_cmp expect actual
> > +	)
> 
> This looks iffy.

Indeed.

> If we expect multiple sharedindex* files, the first output from "ls -t"
> may or may not match the real one in use (multiple things do happen
> within a single second or whatever your filesystem's time granularity
> is).  Two "ls -t" run in this test would (hopefully) give stable
> results, but I suspect that the chance the first line in the output
> matches what --shared-index-path reports is 50% if we expect to have two
> sharedindex* files.
> 
> On the other hand, if we do not expect multiple sharedindex* files,
> use of "ls" piped to "tail" is simply misleading.
> 
> If this test can be written in such a way that there is only one
> such file that match the pattern, it would be the cleanest to
> understand and explain.  As there is only a single invocation of
> "update-index --split-index" immediately after a new repository is
> created, I suspect that the expectation to see only one sharedindex*
> file already holds (because its name is unpredictable, we still need
> to catch it with wildcard), and if that is the case, we can just
> lose "-t" and pipe to tail, i.e. "ls .git/sharedindex* >expect".

Indeed. We can expect only one sharedindex file to be present, and we do
not even have to call out to `ls` but can get away with calling `echo`
(which is a builtin in Bash).

Ciao,
Johannes