Re: [PATCH 2/3] submodule tests: reset "trace.out" between "grep" invocations

[Taylor Blau <me@ttaylorr.com>]

On Mon, Oct 31, 2022 at 01:50:07PM +0100, Ævar Arnfjörð Bjarmason wrote:
> >>  	(
> >>  		cd downstream &&
> >>  		GIT_TRACE=$(pwd)/trace.out git fetch &&
> >>  		grep "1 tasks" trace.out &&
> >> +		>trace.out &&
> >> +
> >
> > I fail to see why these hunks are necessary. If we specify GIT_TRACE,
> > and don't have a test_must_fail around the execution, then why should we
> > feel obligated to clean up the trace.out after every execution?
>
> Because the trace file isn't clobbered by each git command that
> specifies GIT_TRACE, so these tests are basically doing:
>
> 	(echo foo; echo bar) >>trace &&
> 	grep foo trace &&
>
>         (echo bar) >>trace &&
> 	grep bar trace
>
> Now, it just so happens that the earlier command isn't echoing "bar" to
> the file, so this is currently working out.

Ah, nicely explained. This is new to me, since I thought the behavior of
both GIT_TRACE and GIT_TRACE2 was to clobber the file.

> But it's a bad pattern to be pretending as though you care about the
> last output (which was the intent of the test), when really what you're
> testing is the combined output of all preceding commands.
>
> This would also be a potenital landmine with "test_must_fail", just
> because the command failed we're not guaranteed to have written nothing
> to the log (and usually we'd get as far as to write something).

Yeah, both make sense, thanks.

Thanks,
Taylor