[Nix-dev] Re: NixOS on embedded systems

[Lluís Batlle i Rossell]

On Sat, Jul 17, 2010 at 03:32:04PM +0200, Ludovic Courtès wrote:
> Hi Lluís!
> 
> Lluís Batlle i Rossell <viriketo at gmail.com> writes:
> 
> > I have the idea that NixOS expressions usually don't need a compiler, but NixOS
> > relies on the nixpkgs expressions that need a compiler.
> 
> Which are “the nixpkgs expressions that need a compiler” you’re
> referring to?
> In theory, if everything you need is available pre-compiled, you don’t
> need a compiler.  Am I overlooking cases where this isn’t true?

Do you think that hydra builds every nixpkgs package you may need? It builds a
lot, but not all (1st point).

Additionally, it is a heavy operation for a computer to evaluate the current
nixpkgs (2nd point).

Having those nixpkgs pointing mostly to fetchurl derivations instead of usual
stdenv mkDerivations, it makes additionally a fence against building natively or
cross-building those packages (3rd point). Then you could have x86_64 hydra
cross-building for mipsel, and evaluate a nixos in that mipsel (instead of
evaluating a cross-nixos - whose expression still does not exist - in x86_64),
this nixos taking the packages from hydra (so being able to type nixos-rebuild
in your embedded).

I wonder if the current nixos implementation (namely, the modules) make nixos a
heavy evaluation, even with light nixpkgs. Eelco Dolstra may know that better,
as he made the new evaluator with nixos modules in mind I think. The embedded
system may need the garbage collector Eelco talked about from time to time.

Regards,
Lluís.