Re: [gentoo-dev] Re: Re: RFC: auto-detection of unpack dependencies

2008-07-15 Thread Alec Warner
On Tue, Jul 15, 2008 at 1:58 PM, Tiziano Müller <[EMAIL PROTECTED]> wrote:
> Patrick Börjesson wrote:
>
>> On 2008-07-15 21:40, Tiziano Müller uttered these thoughts:
>>> Marius Mauch wrote:
>>>
>>> > As a result of Cardoes earlier mail we talked a bit about possible
>>> > solutions in #gento-portage, and I suggested to let portage
>>> > automatically inject the deps based on SRC_URI pattern matching.
>>> > A mapping of extensions and their unpack deps would be kept in the tree
>>> > (e.g. mapping '.tar.bz2' to '( app-arch/tar app-arch/bzip2 )'
>> [snip]
>>> > So, is this something ebuild maintainers would like in general, or does
>>> > such a feature cause you nightmares?
>>>
>>> Yes. I think that's something which should be done manually.
>>
>> Indeed, the correct solution would be to state the deps manually in each
>> ebuild that requires the dep. But in this case it would mean adjusting
>> the DEPEND string of pretty much the entire tree. Until such measures
>> are stated required, this would be a good middle ground, no?
> no. How about just introducing the new deps on their next version or
> revision bump? (I assume that more than half of the packages would be fixed
> within the next half year and that's more than fast enough).

Why would you do a bunch of manual work when you can script it easily
with few false positives?
Doubly so when false positives are relatively cheap; adding an extra
dep on a package you probably
already have installed won't hurt much and those it does hurt can file
bugs for the RESTRICT and/or
explicitly state the dep.

We could alternatively just use this automatic system to automatically
modify DEPEND in the majority of ebuilds
and work the bugs out that way.

I think doing this entirely manually is just a waste of your time though ;)

-Alec


>
>>
>> The same thing would apply to gcc if all "real" depends were to be
>> required in all ebuilds, but that would pretty much have to be manually
>> stated since the PM wouldn't be able to judge that by automatic
>> measures. This, on the other hand, can (at least partially) be handled
>> automatically for the ebuild-devs on the PM side of things.
> That's a different thing:
> A dependency on gcc just ensures that gcc is installed not that it is
> actually used to build a package.
> And for such a dependency we'd need new ways to express deps since gcc is
> only needed when building packages not when it gets installed from a
> binpkg.
> But this is not an argument for an automagic dep.
>
> --
> gentoo-dev@lists.gentoo.org mailing list
>
>
--
gentoo-dev@lists.gentoo.org mailing list



Re: [gentoo-dev] Re: Re: RFC: auto-detection of unpack dependencies

2008-07-15 Thread Patrick Börjesson
On 2008-07-15 22:58, Tiziano Müller uttered these thoughts:
> Patrick Börjesson wrote:
> > On 2008-07-15 21:40, Tiziano Müller uttered these thoughts:
> > The same thing would apply to gcc if all "real" depends were to be
> > required in all ebuilds, but that would pretty much have to be manually
> > stated since the PM wouldn't be able to judge that by automatic
> > measures. 
> That's a different thing:
> A dependency on gcc just ensures that gcc is installed not that it is
> actually used to build a package.

Not quite sure what you mean here. I'm just saying that if you want to
go the route of stating all deps explicitly, you have to state in the
ebuild (DEPENDS) that gcc is needed to build the package, if that's the
case. I'm not against this at all (I'm not an ebuild-maintainer), i just
gave an example for when there's no sane way for the PM to automatically
inject a dependency. 

> And for such a dependency we'd need new ways to express deps since gcc is
> only needed when building packages not when it gets installed from a
> binpkg.

Portage (or whichever PM you want) uses it's own way of packaging
binpkgs, so for it to be able to extract those binpkgs, a RDEPEND on the
applications used for that specific task has to be stated in the _PM_
itself. It isn't the ebuild deciding which format it's gonna be
packaged down into. 
I'm far from sure about this, but DEPENDS aren't really taken into
consideration when installing from a binpkg, so stating (f.ex) gcc in
DEPENDS wouldn't draw it in when you install the package from a binpkg. 

It is however known to the ebuild-maintainer and/or the PM which format 
the source is packaged in, so that's a sane thing to put in DEPEND,
whether by manual editing of the ebuild/eclass, or by automation in the PM. 

Patrick B

-- 
()  The ASCII Ribbon Campaign - against HTML Email
/\  and proprietary formats.


pgpXag0IuOJCu.pgp
Description: PGP signature