<qyliss>
not sure if my messages aren't getting through of if the log bot isn't picking them up...
<MichaelRaskin>
I see a link to an RFC proposal
<MichaelRaskin>
That got thhrough
<MichaelRaskin>
Maybe submit crate2nix even before RFC submission
<qyliss>
oh, cool, it worked
<qyliss>
cool
<qyliss>
I see. Messages are getting through but very delayed.
<qyliss>
Well this will be fun
<qyliss>
MichaelRaskin: I figure there must be a reason crate2nix isn't included already? Although I don't know what that could be.
<MichaelRaskin>
Looks like there is some soulseeking what to call 1.0, and it kind of works from checkout (but slowly), so nobody gets around to submit it
<MichaelRaskin>
I find some stuff packaged using crate2nix, and I see no attempts to submit it
<qyliss>
there will be resistence to that rfc fwiw
<tazjin>
qyliss: I suspect the resistance there stems more from how carnix generates disjunct dependency sets per application by default
<tazjin>
which should be less of an issue if we were to generate them more like we do for other languages and reuse
<tazjin>
btw, I'm fairly sure peter (from crate2nix) is at 36c3 so we could do a brainstorming session
<qyliss>
I want a quicker fix to this problem
<qyliss>
That would be rad in theory, but buildRustPackage is broken now
<tazjin>
it's not a large change
<tazjin>
fairly sure buildRustCrate already supports doing things that way and the generated files would only need to be slightly different
<tazjin>
the main question I have is about dealing with package versions, because providing a consistent set (like haskellPackages.*) is gonna be difficult for Rust (since there's no stackage or anything like that )
<qyliss>
Exactly
<qyliss>
I think that's a much bigger thing to try to introduce
<tazjin>
not if we populate it lazily
<tazjin>
e.g. crate2nix writes deps into their own derivations when people package things with missing deps
<tazjin>
but we don't preemptively generate a package set
<tazjin>
qyliss: I spoke to peter - would you be up for a chat about this some time tomorrow? (he's busy today)
<qyliss>
Even so, there's a policy question
<qyliss>
I don't think we should package every patch version of every Rust library, for example
<tazjin>
no, but all the ones that we need to build the software in nixpkgs
<tazjin>
and if we're doing that I'd like to avoid duplication
<tazjin>
otherwise we'll have 9 million copies of giant derivations for aho-corasick, serde etc.
<qyliss>
tazjin: I might have time tomorrow but difficult to say right now
<qyliss>
I think we'd additionally want to avoid packaging, e.g., serde 1.2.0 AND 1.2.1
<qyliss>
even if we have packages locked to each
<qyliss>
But this feels to me like a step we could take _after_ switching away from buildRustPackage. Having big Cargo.nix files is unsightly, but buildRustPackage is _broken_.
<tazjin>
qyliss: agree that we can dedup minor versions later, but per distinct version I'd like to avoid duplication
<tazjin>
let me know tomorrow if you have time, I'm trying to stay flexible! :)