Re: Porting to anything

[Ludovic Courtès]

Andreas Enge <address@hidden> skribis:

> The currently unsatisfying situation, according to "HACKING", is that 
> bootstrapping guix essentially starts by installing (and thus in the worst 
> case porting) nix.

Agreed.

> I have spent the evening working on a shell script automating the 
> compilation of the bootstrap binaries. Unsurprisingly, I end up 
> reprogramming in shell part of make-bootstrap.scm. (In particular, it cost 
> me some time to understand how to obtain statically linked binaries of the 
> binutils.) And I am only halfway through, and maybe made some errors in the 
> process that will only reveal themselves later.

Isn’t it equally unsatisfying?  :-)

> So I wonder whether the dependency on nix is essential in make-
> bootstrap.scm. Could one not renounce at working in a chroot with defined 
> inputs from nix, and execute the exact same code in the current 
> environment, using the programs available in the current system from the 
> user's path? Then guix could be bootstrapped easily on any gnu/linux 
> system.

Yes.  I outlined a plan in
<http://lists.gnu.org/archive/html/bug-guix/2013-01/msg00099.html>.

However, I really think cross-compilation is the way to go.  I have a
tentative set of packages for the cross-toolchain that’s not fully baked
yet, and then we’ll have to adjust gnu-build-system to support
cross-compilation.

So that won’t happen overnight, but we could work toward that goal for
the next month or so.

WDYT?

Ludo’, who sympathizes with the feeling of frustration.  :-/