Re: [PATCH 0/2] Add Windows support to the new RUNTIME_PREFIX design

From: Junio C Hamano <gitster@pobox.com>
Date: Mar 28, 2018, 17:12 UTC
Message-ID: <xmqqa7us86pi.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <CAD1RUU9XK837mdRwicMwM5qVApzz8o2e4Eg=B0LH3SRtLqG9WQ@mail.gmail.com>

Daniel Jacques <dnj@google.com> writes:

Show 9 quoted lines

> A simple grep suggests that the current test suite doesn't seem to have any
> RUNTIME_PREFIX-specific tests. When I've been running the test suites, I've
> been doing it with a "config.mak" file that explicitly enables
> RUNTIME_PREFIX to get the runtime prefix code tested against the standard
> Git testing suites.
>
> From a Git maintainer's perspective, would such a test be a prerequisite
> for landing this patch series, or is this a good candidate for follow-up
> work to improve our testing coverage?

It would be a nice-to-have follow-up, I would say, but as you two seem to be working well together and it shouldn't be too involved to have the minimum test that makes sure the version of "git" being tested thinks things should be where we think they should be, with something like...

	test_expect_success RUNTIME_PREFIX 'runtime-prefix basics' '
		(
			# maybe others
			safe_unset GIT_EXEC_PATH &&
			git --exec-path >actual
		) &&
		# compute the expected value -- we know the first
		# element of $PATH is where we find "git", so things
		# should be computable relative to that, perhaps?
		echo >expect "${PATH%%:*}/..." &&
		# then compare
		test_cmp expect actual		
	'

so I am hoping such a minimum test to be in the series when it graduate to 'master' and become a part of a release.

On the other hand, "make a whole test install and try running it" may actually be easier but that probably can be done using existing GIT_TEST_INSTALLED framework? In short, you would probably do

 - make RUNTIME_PREFIX=YesPlease
 - make RUNTIME_PREFIX=YesPlease DESTDIR=...some..where... install
 - GIT_TEST_INSTALLED=...some..where.../bin make test

or something like that.