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

[Marc Weber]

Excerpts from Jeff Johnson's message of Mon Jun 21 00:27:14 +0200 2010:
> Or do I still miss something?
You're right.. It doesn't make sense to pass
N=1, N=10 or N=8 or N=4 to the builder because you'd get 4 different hashes

However you can map it to a boolean values:

POTENTIALLY_BAD = N > 1;

This way you have what you want: You separated the maybe bad packages
from good ones.

This POTENTIALLY_BAD could also be used to non chroot builds etc.

Disadvantage: Because hashes change you have two systems in /nix/store.
If you want to get binaries from Hydra Hydra has to build each package
twice. If you want to find broken packages you even have to run Hydra
below max load to ensure that parallel building within a package takes
place.

I care about that the hash changes if you either opt-out a package
or remove the opt-in so that you can recover if you found a broken build
easily.

Marc Weber