[RFC 140 2a] pkgs/by-name: Enforce for new packages
#275539
Conversation
infinisil
force-pushed
the
by-name-enforce
branch
from
c53ec7b
to
9d746df
Compare
infinisil
added
the
significant
infinisil
force-pushed
the
by-name-enforce
branch
5 times, most recently
from
1c31376
to
41b1926
Compare
|
This PR is now theoretically ready, but since the changes got a bit big, I split off everything not directly related to this PR into #278805, which should be merged first. |
infinisil
force-pushed
the
by-name-enforce
branch
from
41b1926
to
6ff113c
Compare
pkgs/by-name/README.md
Outdated
| Packages found in the named-based structure do not need to be explicitly added to the | ||
| Packages found in the name-based structure must not be added to the | ||
| `top-level/all-packages.nix` file unless they require overriding the default value | ||
| of an implicit attribute (see below). |
There was a problem hiding this comment.
I think it is still better to lead with the positive part. Something like
Packages found in the name-based structure are automatically included, without needing to be added in a file like
all-packages.nix. They must only occur in file if they require overriding the default value of an implicit attribute (see below).
There was a problem hiding this comment.
I changed the wording a little bit, but this is essentially applied now :)
|
This pull request has been mentioned on NixOS Discourse. There might be relevant details there: https://discourse.nixos.org/t/2024-01-09-nixpkgs-architecture-team-meeting-47/38037/1 |
infinisil
force-pushed
the
by-name-enforce
branch
from
6ff113c
to
6e192d8
Compare
| if let Some(path) = &call_package_path { | ||
| format!("./{}", path.display()) | ||
| } else { | ||
| "...".into() |
There was a problem hiding this comment.
For my notes: pkgs/test/nixpkgs-check-by-name/tests/new-package-non-by-name/expected is the test of this.
| // This is a bit of an odd case: We don't even _know_ whether this attribute | ||
| // would qualify for using pkgs/by-name. We can either: | ||
| // - Assume it's not using pkgs/by-name, which has the problem that if a | ||
| // package evaluation gets broken temporarily, the fix can remove it from | ||
| // pkgs/by-name again | ||
| // - Assume it's using pkgs/by-name already, which has the problem that if a | ||
| // package evaluation gets broken temporarily, fixing it requires a move to | ||
| // pkgs/by-name |
There was a problem hiding this comment.
Future ratchets might try to drive this number down, I might imagine.
There was a problem hiding this comment.
Oh yeah, I thought about implementing an eval-successful ratchet, but I think this should be a future PR. Then also it's also not pkgs/by-name-specific anymore and the tool could use a rename.
There was a problem hiding this comment.
One only needs to look at the list of things detected by NixpkgsProblems to realize we've backed ourselves into a damn fine nixpkgs linter framework. 🚀
In a future commit this will be extended
Not that important, but nice. Also adds a nice test case show-casing the two current ratchet checks at once.
This was an oversight in NixOS#275539 not accounted for, which would've failed in CI
This was an oversight in NixOS#275539 not accounted for, which would've failed in CI
|
Now that nixos-unstable contains the latest fixes, here's the final follow-up PR: #281835 which starts enabling the enforcement! |
This was an oversight in NixOS/nixpkgs#275539 not accounted for, which would've failed in CI
Note
An open call for the final review and merge was held on 2024-01-15, see #275539 (comment)
Context
This is part of the effort to implement RFC 140, aka
pkgs/by-name. While previous PRs already enabled users to define packages usingpkgs/by-name, there's still thousands of packages still usingall-packages.nix, and new ones added every day.This PR is part of the RFC 140 migration plan, building on top of #272395 and #274591.
This work is sponsored by Antithesis and Tweag ✨
Changes
This PR prevents new packages using
pkgs.callPackagefrom being defined not inpkgs/by-name. Instead thepkgs/by-namedirectory has to be used instead. This the A-side of Part 2 of the RFC.This doesn't affect packages defined using alternate
callPackage's likelibsForQt5.callPackageorcallPackages, whichpkgs/by-namecannot handle (yet?).Note
The effects of this PR will only be seen once the nixos-unstable channel updates, since its pre-built version of the tooling is used for CI.
Things done
callPackagebut not usingpkgs/by-namethrow an errorpkgs/by-namethrows an errorcallPackagedoesn't give an error.Add a 👍 reaction to pull requests you find important.