Re: glib2 on NetBSD/i386 3.0 using pkgsrc-2011Q1

[Stefan Schaeckeler <schaecsn%gmx.net@localhost>]

Hi all,

> On Freitag, 10. Juni 2011 at 19:26, Stefan Schaeckeler wrote:
> 
> > That makes me wonder how packages and patches are tested. Are committers 
> > only testing them on their plattform and commit if they work there?
> seems so
> > 
> > Wouldn't it make sense to ask (or force?) committers to test packages and 
> > patches in VMs for operating systems that are easy to set up such as intel 
> > linux and intel open solaris?

> In theory â?? yes. In practice, I don't think so.


> Which platforms should be the "core"?

Details. Guess, besides NetBSD, perhaps Dragonfly, one intel Linux distribution 
(=> likely to work on many other linux distributions), intel open solaris (=> 
likely to work on sparc solaris).


> What about software that just doesn't work on all platforms, because
> it needs to be ported? But maybe just works on some platforms?

That should be explictly set with ONLY_FOR_PLATFORMS=.....


> IMHO you can not require testing on all the pkgsrc platforms,
> nothing would ever get done. Porting something like Firefox is a
> hell of a lot of work, for every single platform.

Things that work at one point in time randomly break later on non-NetBSD. I 
assume that, if the committer knows that a new patch breaks some plattforms, he 
will often instantly be able to easily fix it.

I really would love to see *such* cases to be cought.

When we come back to firefox, the build error message is the same as 1 1/2 
years ago.


> I use pkgsrc on DragonFly, and I've come across a few places where
> programs don't work. I've generally found that any fixes I submitted
> through the PR system have been warmly welcomed, which is really
> encouraging.

Not all my sparc/solaris patches are accepted :P


> On the other hand, in very many instances software that has probably
> not been tested on DFly by the maintainers worked out of the box.

.. because Dragonfly is very close to NetBSD?

 
> So from this experience, I'd rather have packages in pkgsrc that may or may 
> not work on non-NetBSD platforms than only packages guaranteed to work 
> everywhere.

 
> Of course, if it *is* possible to test on other platforms before committing 
> that is very desirable, and everyone encountering portability problems is 
> dearly encouraged to fix them and submit the fixes.


> >  How about setting up a build farm of various plattforms
>
> that would be cool!

That would make a great Google Summer of Code project :)


> >  and require commits to successfully build on these machines?
>
> hmm, I'd say: no. maybe "successfully on those it built before,
> unless there is good reason"?

Ok, that's perhaps too demanding.

 Stefan

-- 

Scotty: Captain, we din' can reference it!
Kirk:   Analysis, Mr. Spock?
Spock:  Captain, it doesn't appear in the symbol table.
Kirk:   Then it's of external origin?
Spock:  Affirmative.
Kirk:   Mr. Sulu, go to pass two.
Sulu:   Aye aye, sir, going to pass two.