2023-10-16 Nix team meeting minutes #95

fricklerhandwerk · October 20, 2023, 5:38pm

Attendees: @thufschmitt @tomberek @ericson2314 @fricklerhandwerk @edolstra @infinisil

Agenda

Assessing concerns with moving forward with flakes stabilisation
Reviews

Blockers for stabilizing `fetchTree`

Making the fetchTree stabilisation depend on rewriting `fetchGit` with `libgit`

@roberth raised the issue last time
Fear that we’ll want to change the API a bit based on that, which should happen before the stabilisation
- We can decide to fix forward until the next release
We shouldn’t make the stabilisation wait forever if that doesn’t happen quickly enough
Concerns for breaking changes:
- git config
- ssh config
  - libgit works fine fetching over https and the legacy git protocol, but .ssh/config will be ignored
  - @edolstra: libssh2 is supported, but it does not seem to have the right features
    - there is a patch but not upstream yet
  - we could start with adding NixOS tests to prepare the migration to libgit, and then see step by step whether it has all the features we need
Advantages of using libgit
- Accessing files in source trees efficiently
- Lazy tree uses it for caching
Decision: it’s worth doing but it’s not a blocker for stabilising fetchTree

Registry

Not part of the stabilisation.

URI syntax

Not intended to be part of the first stable fetchTree.
The structured attribute version is easier to handle as it has fewer edge-cases.

Agreement on not stabilising the URI syntax for now.

Documentation

Needs a rewrite to reflect reality of stable features.

@fricklerhandwerk will take care of that.

“Fancy” fetchers (`mercurial`, `github`, `gitlab`, `sourcehut`)

mercurial already exists through fetchMercurial, so we should stabilize
fetchTarball and fetchUrl are already builtins
gitHub is widely used (and useful) for efficiently fetching Nixpkgs
Don’t stabilise gitlab and sourcehut yet
- Those are not as commonly used and it gives us time to fix remaining issues

Git smudging

Smudging allows running arbitrary stuff at fetch time. Do we want to have it run by default? Or at all?

Current behavior:

fetchGit doesn’t run smudging
github fetches tarballs, and GitHub applies smudging to those

Examples:

$ nix build github:tomberek/export-magic -L
text> /nix/store/8fdx2h960yf292czd4ik2f6fza63wlqh-.version
text> HEAD -> main
text>  (HEAD -> main)
text> f0fe26485056d92e78d84de3daf14a8875548783
text> Tue, 21 Feb 2023 19:51:28 -0500
text> 2023-02-21 19:51:28 -0500
text> 2023-02-21T19:51:28-05:00
text> HEAD -> main
text>  (HEAD -> main)
text> f0fe26485056d92e78d84de3daf14a8875548783
text> Tue, 21 Feb 2023 19:51:28 -0500
text> 2023-02-21 19:51:28 -0500
text> 2023-02-21T19:51:28-05:00

But also:

$ nix build git+ssh://git@github.com/tomberek/export-magic -L
text> /nix/store/8fdx2h960yf292czd4ik2f6fza63wlqh-.version
text> HEAD -> main
text>  (HEAD -> main)
text> f0fe26485056d92e78d84de3daf14a8875548783
text> Tue, 21 Feb 2023 19:51:28 -0500
text> 2023-02-21 19:51:28 -0500
text> 2023-02-21T19:51:28-05:00
text> HEAD -> main
text>  (HEAD -> main)
text> f0fe26485056d92e78d84de3daf14a8875548783
text> Tue, 21 Feb 2023 19:51:28 -0500
text> 2023-02-21 19:51:28 -0500
text> 2023-02-21T19:51:28-05:00

@thufschmitt and @tomberek will look into it.