Sounds good to me based on your investigation in #375745.

nixpkgs-review result

Generated using nixpkgs-review.

Command: nixpkgs-review pr 375850

x86_64-linux

The key thing that still seems to be failing to build now is rocblas.

ollama-rocm, hipblas, hipsolver, rocalution,rocsolver were fine before this issue came up, but are downstream from rocblas.

The others were already broken, and should maybe be marked as such, or not, since 6.3 seems to be coming up anyways.

I have attached the build log for rocblas here: rocmPackages.rocblas-x86_64-linux.log.

I'm still not sure yet if skipping that test is the right fix. Adding a mechanism to bypass the warning would be another possibility.

I have marked this PR as ready for review, not because I actively want it reviewed or merged before looking into what's up with rocblas, but because if someone can confirm that it fixes their use case I think merging this for now might make sense.

The log output of the rocblas build is full of output produced by this line (probably not from that exact commit though):
https://github.com/ROCm/hipBLASLt/blob/a11ccf64efcd818106dbe37768f69dfcc0a7ff22/tensilelite/Tensile/SolutionStructs.py#L1297

Thanks for this, being that it addresses the only test that fails on my system, I'll apply this locally even if it isn't merged:

********************
Failed Tests (1):
  llvm-libc++-shared.cfg.in :: libcxx/transitive_includes.sh.cpp


Testing Time: 560.01s

Total Discovered Tests: 7869
  Unsupported      :  428 (5.44%)
  Passed           : 7400 (94.04%)
  Expectedly Failed:   40 (0.51%)
  Failed           :    1 (0.01%)
...
error: builder for '/nix/store/35a4dhaa1b27dcscz6qnj97a15qma2bx-rocm-llvm-libcxx-6.0.2.drv' failed with exit code 1

This works for me too.

I'd say let's merge this then and fix the issue with rocblas is in another PR.

Does anybody have an idea how we could turn off/patch out that echo from the stdenv we use in ROCm?

Quoting myself from the PR mentioned above

I'm guessing reverting this or adding an environment variable to turn this off in some places would not be a practical fix because it would cause a mass rebuild and therefore have a really long turnaround time?

I have marked this PR as ready for review, not because I actively want it reviewed or merged before looking into what's up with rocblas, but because if someone can confirm that it fixes their use case I think merging this for now might make sense.

I'd guess a large proportion of people hitting this issue probably care only about gaming, and I've verified Steam works with this patch applied. It seems like a harmless workaround until #370435 is ready.

Will this be merged into the stable as well ?

Will this be merged into the stable as well ?

I think what you might be waiting for is just for this to land in nixpkgs-unstable, where it has not quite landed yet as of now, because it takes some time for channels to advance. A few days at most, unless something is seriously 'blocking' the channel. You can keep track whether a PR has landed in a channel yet here: https://nixpk.gs/pr-tracker.html?pr=375850

What people usually do to get away from breakage like this is stay on the last working version until the fix is ready, or move forward to a newer commit on master, which has not landed yet. Though I have no idea how that stuff works without flakes, besides picking the previous generation in the bootloader.

It does not look like the change to the wrapper that caused this build failure was backported into a release, and the package seems to be building fine on 24.11:
https://hydra.nixos.org/build/284989140

Will this be merged into the stable as well ?

I think what you might be waiting for is just for this to land in nixpkgs-unstable, where it has not quite landed yet as of now, because it takes some time for channels to advance. A few days at most, unless something is seriously 'blocking' the channel. You can keep track whether a PR has landed in a channel yet here: https://nixpk.gs/pr-tracker.html?pr=375850

What people usually do to get away from breakage like this is stay on the last working version until the fix is ready, or move forward to a newer commit on master, which has not landed yet. Though I have no idea how that stuff works without flakes, besides picking the previous generation in the bootloader.

It does not look like the change to the wrapper that caused this build failure was backported into a release, and the package seems to be building fine on 24.11: https://hydra.nixos.org/build/284989140

Looks like if I use 24.11 and do not mix stable and unstable packages I'm all good. I was running ollama from unstable with rocm packages from unstable as well. Thanks for the update I will just need to wait then. Cheers

Looks like if I use 24.11 and do not mix stable and unstable packages I'm all good. I was running ollama from unstable with rocm packages from unstable as well. Thanks for the update I will just need to wait then. Cheers

Note that the present PR will not fix ollama-rocm, since rocblas still doesn't build (see #375850 (comment)).