Btw.: afair the largest issues in terms of download stability came from source repos, e.g. the gitlab from MPI-SWS hosting IRIS and also the INRIA one was not entirely stable. But it was not that bad that I bothered to create statistics. 2 years back it was overall much worse.
this might be an option in the future: https://hannes.robur.coop/Posts/OpamMirror
(at least I would prefer predictable hosting of the crucial source archives)
Nice! Does it also mirror the coq repos? All issues I had came from coq opam packages (afair).
I think the mirroring includes Git repositories, if that's what the underlying packages in an opam repo use. Is that what you mean?
there is some "dump git" code here, so Git support seems plausible: https://git.robur.io/robur/opam-mirror/src/branch/main/mirage/unikernel.ml#L736
My question was if they mirror just the Ocaml opam repo + referenced tar balls (which I expect) or also the two coq opam repos (released and extra-dev) + referenced tar balls.
I mean, the current deployment mentioned in the blog post only mirrors the general OCaml opam repo. But the software they use/develop can support any opam repo, like released
and extra-dev
. This is my understanding, at least.
That's cool! BTW, another path forward would be to rely on Software Heritage for archiving tarballs. IIRC this has already been implemented for Guix and is a WIP for Nix. Doing the same for opam would seem natural.
@Karl Palmskog : It just means that we need servers to host this.
the ideal is probably to combine the mirroring with hosting on Software Heritage, e.g., we just host the opam repo metadata, which points all packages to SH-hosted tarballs
Yes, this would make sense. Still I want to repeat that the reliability seen last year doesn't call for action.
Last updated: Oct 12 2024 at 13:01 UTC