Hello! I'll take a look at this when I'm able, but don't let me approval be a blocker.

I've noticed a lot of PRs targeting staging recently dealing with various aspects of support for __structuredAttrs -- is there a tracking issue for better support I can use to look at recent changes? For example, I'm not familiar with the function being used in this PR, concatTo.

I've noticed a lot of PRs targeting staging recently dealing with various aspects of support for __structuredAttrs -- is there a tracking issue for better support I can use to look at recent changes? For example, I'm not familiar with the function being used in this PR, concatTo.

There is no tracking issue, the discussion around this all happened in #318614.

I've noticed a lot of PRs targeting staging recently dealing with various aspects of support for __structuredAttrs -- is there a tracking issue for better support I can use to look at recent changes? For example, I'm not familiar with the function being used in this PR, concatTo.

There is no tracking issue, the discussion around this all happened in #318614.

I updated an old tracking issue: #205690. If you do additional PRs, would you mind referencing it? Makes it easier to track related work (which will come in handy when I eventually need to update or troubleshoot the setup hooks the CUDA ecosystem uses).

I haven't looked at it yet, but will try to do it by tomorrow. I originally wrote the tests for #272752 - I don't think they were supposed to fail, but I can't be 100% sure either honestly as it's been some time. Will at least take a quick look at the output of nix-build -A tests.auto-patchelf-hook

@wolfgangwalther Thanks for pinging me. The test is broken on master, and fails at [auto-patchelf-hook-test]: Check that the runpath has the right number of entries, which differs between the actual number of entries in the RUNPATH and the expected number of entries.

I suspect it's a change in stdenv or something, but this simple patch should fix the test (at least it does on master and it's unrelated to what we're interested in testing):

--- a/pkgs/test/auto-patchelf-hook/package.nix
+++ b/pkgs/test/auto-patchelf-hook/package.nix
@@ -60,7 +60,6 @@ stdenv.mkDerivation {
   # Additional phase performing the actual test.
   installCheckPhase =
     let allDeps = runtimeDependencies ++ [
-        (lib.getLib stdenv.cc.libc)
         (lib.getLib freetype)
       ];
     in

If you want to make sure your change didn't break the test, given the one line diff, you can include the fix in this PR I guess.

If you want to make sure your change didn't break the test, given the one line diff, you can include the fix in this PR I guess.

Did that, thanks!

The Zig build hook appears to have been broken by these changes.

The Zig build hook appears to have been broken by these changes.

Do you mean the autoPatchelfHook ones specifically, or a previous PR @jcollie?

The Zig build hook appears to have been broken by these changes.

Do you mean the autoPatchelfHook ones specifically, or a previous PR @jcollie?

I'm not exactly sure which PR broke what, but I see this now when trying to build a Zig package:

/nix/store/3bwyb6sw8qzir9prnh3609anzw9nvp6g-zig-hook/nix-support/setup-hook: line 15: concatTo: command not found

The Zig build hook appears to have been broken by these changes.

Do you mean the autoPatchelfHook ones specifically, or a previous PR @jcollie?

I'm not exactly sure which PR broke what, but I see this now when trying to build a Zig package:

/nix/store/3bwyb6sw8qzir9prnh3609anzw9nvp6g-zig-hook/nix-support/setup-hook: line 15: concatTo: command not found

This is very likely unrelated to this PR here. The zig hooks were changed in #335666. This is already on master. From what I can tell, building packages with zig still works.

Where (on which branch?) and how (any specific package to build?) can we reproduce this?

Where (on which branch?) and how (any specific package to build?) can we reproduce this?

This is unfortunately in a package that isn't open source (yet), but it was on unstable. I'll see if I can reproduce it tonight in another package that is open source.