Re: [PATCH v4] Makefile: replace perl/Makefile.PL with simple make rules

From: Todd Zullinger <tmz@pobox.com>
Date: Dec 20, 2017, 06:15 UTC
Message-ID: <20171220061545.GF3693@zaya.teonanacatl.net>
In-Reply-To: <20171219235759.27570-1-avarab@gmail.com>

Hi Ævar,

Ævar Arnfjörð Bjarmason wrote:

Show 12 quoted lines

> Here's a hopefully final version. The only difference with v3 is:
> 
>     -    local @_ = ($caller, @_);
>     +    unshift @_, $caller;
> 
> As it turns out localizing @_ isn't something that worked properly
> until
> https://github.com/Perl/perl5/commit/049bd5ffd62b73325d4b2e75e59ba04b3569137d
> 
> That commit isn't part of the 5.16.3 version that ships with CentOS 7,
> which explains why Michael J Gruber had issues with it. I've tested
> this on CentOS 7 myself, it passes all tests now.

Thanks for tracking this down!

FWIW, I applied this version to next and tested it with CentOS 6 and 7. The tests pass on both (though there are some unrelated failures on CentOS 6 in t5700-protocol-v1, which I haven't looked into further yet).

I also applied this patch to 2.15.1 and ran the tests in the Fedora build system for all fedora and epel releases, which also passed (though with some spurious git-svn failures on x86_64 in fedora 28, AKA rawhide).

The .pmc extensions seem to cause rpm to fail to parse the files for rpm 'provides' as it normally would. This causes scripts like git-send-email which generates a 'requires' on Git::Error to fail to find anything which provides it.

I'm not familiar with the .pmc extenstion. Searching the fedora repositories, there is only one other package - and one file within it - which has a .pmc extension.

(The package is perl-test, the file is /usr/libexec/perl5-tests/perl-tests/t/run/flib/t2.pmc.)

Perhaps it's a bug in rpm's perl dependency generator, but I'd like to think that git wouldn't be the first package to find it.

Is the .pmc extension important to ensure these files are loaded in the right order? Since they're all in the Git namespace, I don't imagine there should be anything else in @INC which would be provided by the system or another package.

Pardon my ignorance if I've missed the obvious (I haven't fully read "perldoc -f require" which you referenced in the commit message).

-- 
Todd
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Suppose I were a member of Congress, and suppose I were an idiot. But,
I repeat myself.
    -- Mark Twain