pkgsrc-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: databases/p5-gdbm "Review"



2010/8/28 Greg Troxel <gdt%ir.bbn.com@localhost>:
>
> Jens Rehsack <rehsack%googlemail.com@localhost> writes:
>
>> 2010/8/27 Thomas Klausner <wiz%netbsd.org@localhost>:
>>> On Thu, Aug 26, 2010 at 01:04:03PM +0200, Jens Rehsack wrote:
>>>> 2010/8/26 Stoned Elipot <stoned.elipot%gmail.com@localhost>:
>>>> > Are you sure you want to add
>>>> > PERL_MM_USE_DEFAULT=1 to lang/perl5/module.mk's MAKE_ENV ?
>>>>
>>>> I'm pretty sure - otherwise each prompt() in Makefile.PL or Build.PL will 
>>>> wait
>>>> until a user answers. This is a no-go for pkg_rr or other, non-bulk-build
>>>> updates.
>>>
>>> Well, this should be caught by the package updater, shouldn't it? And
>>> I don't remember that we had problems because of this before.
>>
>> I had several problems with that on some perl packages, especially
>> while rebuilding all p5-* packages using pkg_rr. I run pkg_rr unattended
>> (probably pkg_rr should define sth. like "BATCH_MODE", and
>> PERL_MM_USE_DEFAULT should be set in BATCH_MODE only),
>> and had 3 or 4 packages where the Makefile.PL/Build.PL waits for user
>> input. So the pkg_rr runs the entire weekend instead of 18h.
>
> I see your point, but I don't think this is really about pkg_rr.
> Basically pkgsrc does not support the concept of interaction by a human
> with the build system of something in a pkg.  pkg_comp and 'make
> package-install' with dependencies both have the same problem as you are
> running into, so doing something to pkg_rr isn't the right answer.

It depends. When it's wanted that packagers stumble over those
questions by a module, they shouldn't test in BATCH_MODE.
I think, enabling pkg_rr to use an .pkg_rr.rc from users home (or
$prefix/pkg_rr.rc), would be enough, because it already supports
pass-though -D flag.

> Another way to say that is that what you wonder about putting in
> BATCH_MODE should just be the way it is.
>
> It seems the problem is that perl needs to have separate switches to not
> ask questions

Yes. That's what's I missed.

> and to not load from CPAN.  What I think we want is for
> any attempt to load from CPAN to simply result in immediate failure.
> python has the same problem.

That happens automatically by the faked home dir. As long as we don't
put a CPAN confiig in it, any cpan download attempt will fail immediately.
Further, neither ExtUtils::MakeMaker nor Module::Build download anything
automatically from CPAN (newer M::B could, but ask back before and
have 'N' as default answer). Module::Install needs "--skipdeps" to avoid
any downloading action. People work on a better solution overall, because
I explain our (pkgsrc) point of view regulary in #toolchain on irc.perl.org.

/Jens


Home | Main Index | Thread Index | Old Index