4 ms·
> This sounds in some ways pretty similar to something I'd been imagining for Nix, particularly for the purpose of grabbing package definition sources which hav
by tomberek 4y ago
> This sounds in some ways pretty similar to something I'd been imagining for Nix, particularly for the purpose of grabbing package definition sources which have been known to work for a given version or range. (The use case is automatically grabbing or generating packages for given semantic versions instead of by Nixpkgs hash.)
Yes, this is something users often expect. We make this more efficient than grabbing a unique Nixpkgs hash for each package, which would mean lots of downloading and disk usage for each one - instead our catalog lets you bypass that need.
> I'm interested to see what all they've done with the catalog, and I wonder if it will ever be brought to Nix proper.
Some of the ideas need design iteration before it would be proper to bring into Nix. Some building blocks have already been incorporated (eg: eval-store, impure-derivations and fetch-closure were funded and inspired by flox).
What portions are of greatest interest?
- pxc 4y ago> What portions are of greatest interest? I will have a better idea of this after I've had a chance to play with the software! Unfortunately that may be in about a week. Naively: I think a DB of successful builds and what Nix functions/files produced the working packages and at what versions could be really useful to things like dream2nix or nix-init. (I think that CLI for one-off installs has considerable value itself.) For my present use cases (perhaps not future ones at my org!) the cases enabling distribution of binary artifacts without publishing sources are not of great interest. For me, those fall into the enterprise-y bucket.