Hello! I sense a lot of impatience in your message :-), and I also see many questions. It is up to us all to answer them, I’ll just reply selectively here.
zimoun <zimon.touto...@gmail.com> skribis: > On Thu, 7 Oct 2021 at 18:07, Ludovic Courtès <ludovic.cour...@inria.fr> wrote: [...] >> The second-best solution is to improve our tooling so we can actually >> keep source code in a more controlled way. That’s what I had in mind >> with <https://ci.guix.gnu.org/jobset/source>. We have storage space for >> that on berlin, but it’s not infinite. > > If Berlin has space, why so much derivations are missing when running > time-machine? That’s not related to the question at hand, but it would be worth investigating, first by trying to quantify that. For the record, the ‘guix publish’ config on berlin is here: https://git.savannah.gnu.org/cgit/guix/maintenance.git/tree/hydra/modules/sysadmin/services.scm#n485 If I read that correctly, nars have a TTL of 180 days (this is the time a nar is retained after the last time it has been requested, so it’s a lower bound.) >> Another approach is to use ‘git-fetch’ more, at least for non-Autotools >> packages (that’s the case for Scotch, for instance.) > > This is what I suggested when opening this thread [1] more than one > year ago. Reading the discussion and keeping in mind the inertia, I > do not think it is a viable path. For instance, you know all the > pitfalls and you updated Scotch without switching to git-fetch -- no > criticism :-) just a realistic matter of facts to have good coverage. > > <https://lists.gnu.org/archive/html/guix-devel/2020-05/msg00224.html> Right, and I agree Scotch is a package that can definitely use ‘git-fetch’ (there are bootstrapping considerations of packages low in the stack, for instance you wouldn’t want to have Git fetched over ‘git-fetch’, but for packages like this there’s no reason not to use ‘git-fetch’.) Thanks, Ludo’.
