Bug #72254 for Module-Build: Missing deps should have a non-zero exit status

Mon Nov 07 13:47:06 2011 ovid [...] cpan.org - Ticket created

Subject:

Missing deps should have a non-zero exit status

When I use Makefile.PL, I rely on this: perl Makefile.PL && make && make test In writing up an example for basic Module::Build usage, I did the same thing: perl Build.PL && ./Build && ./Build test However, when I tried to build Weather::Google which lists a dependency on XML::Simple, it correctly reported that XML::Simple was not installed, but it went on to run (and fail) all of the tests. ./Build should return a non-zero status to ensure that the && ./Build test does not run. Cheers, Ovid

Mon Nov 07 14:13:25 2011 mschwern [...] cpan.org - Correspondence added

I agree.

Mon Nov 07 14:13:26 2011 The RT System itself - Status changed from 'new' to 'open'

Mon Nov 07 14:26:20 2011 dagolden [...] cpan.org - Correspondence added

I think a non-zero exit status will convince CPAN clients to abort, which means they won't then attempt to resolve prerequisites. Historically, missing prereqs were advisory -- as many modules might still work with version X even when the prerequisite was listed as Y (>X). Thus, a failure due to a prerequisite would manifest during an actual test, not during Build.PL. I would be willing to see a non-zero exit *only* for Perl 5.10.1 or greater and only when we don't have the magic PERL_CPAN_IS_RUNNING (or PERL_CPANPLUS_IS_RUNNING) or whatever the actual magic environment variables are. (Off the top of my head, I think CPAN/CPANPLUS as of 5.10.1 both set the variables correctly, which is why I would set that minimum.) I don't have time to check the exact behaviors of the clients, but I hope that gives the general parameters for what a solution would involve. -- David

Mon Nov 07 14:26:39 2011 dagolden [...] cpan.org - Severity Unimportant added

Mon Nov 07 14:51:56 2011 schwern [...] pobox.com - Correspondence added

Subject:	Re: [rt.cpan.org #72254] Missing deps should have a non-zero exit status
Date:	Mon, 07 Nov 2011 11:51:44 -0800
To:	bug-Module-Build [...] rt.cpan.org
From:	Michael G Schwern <schwern [...] pobox.com>

Sorry, my +1 was hasty. This doesn't work and AFAIK never did: perl Makefile.PL && make && make test The Makefile.PL does not exit with non-zero on prereq failure nor does make. It runs through to the test. In isolation it would be nice if they did exit non-zero, I agree with David that the behaviors of the CPAN clients will probably be affected.

Mon Nov 07 19:41:56 2011 LEONT [...] cpan.org - Correspondence added

On Mon Nov 07 14:26:20 2011, DAGOLDEN wrote: Show quoted text

> I think a non-zero exit status will convince CPAN clients to abort, > which means they won't then attempt to resolve prerequisites.

Well, he's suggesting to return non-zero during build, not configuration, so the CPAN client should have already installed the requirements by then. I'm not seeing a problem there. Show quoted text

> Historically, missing prereqs were advisory -- as many modules might > still work with version X even when the prerequisite was listed as Y > (>X). Thus, a failure due to a prerequisite would manifest during an > actual test, not during Build.PL.

The documentation has warned the users since 2002 (version 0.06) that it may do hard require. I'd say that's more than enough warning. Leon

Mon Nov 07 19:53:10 2011 dagolden [...] cpan.org - Correspondence added

Subject:	Re: [rt.cpan.org #72254] Missing deps should have a non-zero exit status
Date:	Mon, 7 Nov 2011 19:52:39 -0500
To:	bug-Module-Build [...] rt.cpan.org
From:	David Golden <dagolden [...] cpan.org>

On Mon, Nov 7, 2011 at 7:41 PM, Leon Timmermans via RT <bug-Module-Build@rt.cpan.org> wrote: Show quoted text

> Well, he's suggesting to return non-zero during build, not > configuration, so the CPAN client should have already installed the > requirements by then. I'm not seeing a problem there.

I missed that detail, but I would disagree with it. "make/Build" has nothing to do conceptually with prerequisites so I wouldn't use it for a general check of that sort. I could imagine a "hard prereqs" mode where "make/Build" fails if build_requires are missing and likewise a failure of "make test/Build test" if test_requires are missing (which begs the question of what to do with "runtime_requires", aka, plain old "requires"). Ultimately, I think that's just messy. The right place to fail is at configuration time during Makefile.PL/Build.PL and only under a flag that enables that (e.g. something like PREREQ_FATAL for EU::MM, perhaps call it "--prereq_fatal" for M::B) or else "automagically" under the rules I laid out where it is safe to do so without screwing up CPAN clients. -- David