[Nix-dev] Re: Another approach at parallelizing in-build jobs

[Lluís Batlle i Rossell]

Hello Marc,

On Sun, Jun 20, 2010 at 02:15:21PM +0200, Marc Weber wrote:
> > That my fear is related to people that may have written bad the Makefiles, and
> > that fear makes me simply against the opt-out approach, but I sure favour the
> > opt-in.
> 
> No matter in which way you look at something. Which is the policy to
> opt-out/opt-in packages?
> 
> Can we talk about ideas to find out whether there is a solution at all
> other than "I trust the comitter that he had a good feeling?"
Having opt-in allows knowing who to ask, if there is a problem on parallel
building. And I'd go for trusting the commiter, and seeing the commit message
in nix-commits, so if anyone doubts about parallelizing that package, others can
notice and discuss.
The committer can write in the commit log, why he thinks that the parallel build
will work. Some upstream packages are explicitly maintained such parallel builds
work, for example (openoffice is one of those).

> Because I don't have an answer for that I used a global opt-in and
> wanted to use this feature on my computer only while using hydra
> binaries.
I think that the kind of 'global opt-in' can be done through substituters, which
already convey the meaning of "same out path, but got in some other way the user
trusts as good enough".

> I'm for a perfect patch. But I don't know how to do it unless we
> discuss this topic in depth.
Good!

> That's why I didn't even try to make a perfect patch.
That lighted the fire the discussion.

Regards,
Lluís.