[Fwd: Re: PM post on Makefile.PL FAIL for missing libraries]

CPAN testers requests the opinion of someone with MakeMaker / C compiler fu.
They foolishly asked me. :)

Anyone have a thought?


-------- Original Message --------
Subject: Re: PM post on Makefile.PL FAIL for missing libraries
Date: Mon, 24 Sep 2007 19:03:17 -0400
From: David Golden <[email protected]>
To: CPAN Testers Discuss <[email protected]>
CC: Michael G Schwern <[email protected]>
References: <[email protected]>
<[email protected]>
<[email protected]>
<[email protected]>

On 9/24/07, Barbie <[email protected]> wrote:
> >   die "Required library not installed: libfoo\n";
>
> I would prefer the author didn't force that text, as it implies they are
> doing their own checking for a library, which is not necessarily the
> same as how a compiler would load the library. The better way is to be
> able to parse the compiler output and detect the warning that the
> library is missing.

That makes sense, but requires parsing compiler output for error
messages for missing libraries -- and actually, a missing library is
probably going to fail compilation due to missing headers anyway.  And
then how do you tell if a missing header is really a library problem
or just a typo?

  #include <sdtio.h>

> > >   m/No library found for -l([-\w]+)/g
> >
> > Should I add that to CPAN::Reporter as well?  If so, I'll squeak it in
> > before 1.00.  How portable is that across compilers?
>
> It was working for gcc and Visual Studio, I don't have others so can't
> say whether it is truly portable. If we encourage this it might throw up
> other formats.

Looking into it -- that seems to be a Makefile.PL output from
libraries in LIBS not being found -- generated from a lot of deep
magic buried in ExtUtils::LibList::Kid.  (Which is probably why it's
portable.  And I don't think that Module::Build offers any comparable
capability, either.

Further, the warning is actually this:
    Note (probably harmless): No library found for -lfoo

So is it harmless or not?  I guess it depends on whether make fails,
which might or might not be due to a missing library.  Should a
harmless missing library and a .xs syntax error cause a report to be
thrown away?

Also, at least for CPAN::Reporter, I'd have to add capabilities to
CPAN and CPAN::Reporter to retain output from each phase to the next
so that if "make" fails, I can go back and parse PL output for the "no
library" string.  That's not terribly hard (Andreas and I have already
discussed tacking it onto the CPAN::Distribution object), but it's
more than the trivial addition of a quick regex that I had
anticipated.

> I think we are in agreement that these reports should not be sent, but
> need to find a suitable method to detect the reason for the failure and
> accepting that missing libraries are legitimate reasons for abandoning
> the report.

I've copied Schwern since I'm not sure if he's on the
cpan-testers-discuss list and I think we need the opinion of someone
with more EUMM/compiler-fu...

David


-- 
I have a date with some giant cartoon robots and booze.