@shlevy, thanks for your PR! By analyzing the history of the files in this pull request, we identified @edolstra, @peti and @Profpatsch to be potential reviewers.

I'm pretty sure this is right and my tests pass, but just wanted a second opinion.

Nope, it's wrong, fix incoming

OK, should be working now

I think something like that should be reviewed before being merged. cc @peti @nbp

It would also be nice to have tests for this function, in lib/tests.nix.

OK, I'll add some tests

I also agree with @Profpatsch that we should be adding tests to our lib functions now (see lib/tests directory).

For function tests, lib/tests.nix is the correct file. The lib/tests/ folder is only set up for module tests.

Hah I just recently wrote this not knowing it was added. It took me a couple attempts, but then I fixed it by proving the monoid https://gist.github.com/Ericson2314/ab120ccce04ac41b2ec2bec11fda3879 . I'm not sure which of machine-checked tests or hand-written proof is better for inclusion in Nixpkgs, but rest assured this PR is correct 😀 .

@Ericson2314 There are tests already, but if you want to formalize nix in coq and check in a machine-checked proof I'm 💯 👅

Maybe it would be more practical to compose lists of extensions directly. We already use

let # ...
    toFix = lib.foldl' (lib.flip lib.extends) (self: {}) someListOfExtensions;
in lib.fix toFix

@vcunat More practical in what sense? We've been using this as-is for a while now.

Just a random thought. It surely depends on particular use case (and the number of extensions you usually combine).

@vcunat on the contrary I rather do a fold1 with composeExtensions and fix with a fixExtension = f: fix (f {}).

Yes, using folds explicitly would be a more idiomatic way, at least if we had a forldr1 or foldl1, but it's easy enough to write those.

Well, while it's extra needless work, the monoid identity can always provide the base case for a normal fold too.