Re: Malconnected machines & guix pull

[Ricardo Wurmus]

Pjotr Prins <address@hidden> writes:

> On Thu, Mar 10, 2016 at 04:44:30PM +0100, Ricardo Wurmus wrote:
>> 
>> Marco van Zwetselaar <address@hidden> writes:
>> 
>> > In fact, what I am really looking for is a --substitute-urls option for 
>> > 'guix pull', or a way to separate the pull and the subsequent build, so 
>> > that I can make it connect to a 'guix publish'-ing server.
>> 
>> Instead of “guix pull” may I suggest using git instead?  Especially,
>> when you have multiple machines on which Guix should run, it may make
>> sense to just clone the Guix git repository once from the Internet,
>> clone it to the other machines via local network, and then update once
>> in a while via “git pull” (not “guix pull”).
>> 
>> You can create a link “~/.config/guix/latest” pointing to, say,
>> “~/dev/guix” (if that’s the directory holding the cloned repository).
>> 
>> When you update with “git pull” you’ll only have to download recent
>> changes, not the whole Guix code.
>> 
>> Does this make sense?
>> 
>> ~~ Ricardo
>
> This has the additional benefit of 'fixating' the dependencies and
> having some control over status of packages. This is why we have a
> gn-stable checkout of guix which is shared among deployments. And it
> can save a lot of downloads. 

Yeah, this achieves what in traditional package management scenarios is
done by a repository mirror.  In combination with local build hosts you
can ensure that all packages at the current checkout are available as
binary substitutes.

~~ Ricardo