I'm okay with this change if it's necessary although if so maybe it would be better to just use the older version universally.

However, I'd like to understand why tbb_2022 isn't working and maybe if there is an issue about it in the tbb repo or some other report about it somewhere.

We tested with this version upstream on macOS in the BLAKE3 repo on CI (since homebrew uses 2022) and it seemed to work fine there.

@Mic92 Can you tell me how to reproduce this? What commands are you running?

I see from your comment here that you are seeing crashes during early initialization even on master, aside from the BLAKE3 stuff.

I just tried to get nix compiling on my ARM macOS system and also am getting crashes (on master) just from trying to enter the dev environment with nix develop. So I can't seem to get to a point where I can even try to debug this.

EDIT: The issue I mentioned was unrelated; now resolved.

@Mic92 I can't seem to reproduce this on my system, running on macOS 15.3.2.

What I tried is adding tbb_2022_0 as a new package in packaging/dependencies.nix:

  tbb_2022_0 = pkgs.tbb_2021_11.overrideAttrs (attrs: rec {
    version = "2022.0.0";
    src = pkgs.fetchFromGitHub {
      owner = "oneapi-src";
      repo = "oneTBB";
      tag = "v${version}";
      hash = "sha256-XOlC1+rf65oEGKDba9N561NuFo1YJhn3Q1CTGtvkn7A=";
    };
  });

Then compiling nix on the blake3-tbb branch as usual, and running the hash command:

nix --extra-experimental-features blake3-hashes hash file --type blake3 1G.bin

This doesn't give me any errors or crashes. So I'm not sure what is causing the problem you are seeing.

It is very reproducible when switching nixpkgs in the nix flake. It breaks the test suite:

$ uname
Darwin evo 24.1.0 Darwin Kernel Version 24.1.0: Thu Oct 10 21:02:26 PDT 2024; root:xnu-11215.41.3~2/RELEASE_ARM64_T8122 arm64 arm Darwin
$ commit 866e193fbcd284a2385352a03dc57b040c762fb0
Author: Jörg Thalheim <[email protected]>
Date:   Wed Mar 26 08:25:39 2025 +0100

    switch to nixpkgs master

diff --git flake.lock flake.lock
index ce484a67a..377d7e88b 100644
--- flake.lock
+++ flake.lock
@@ -63,16 +63,16 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1734359947,
-        "narHash": "sha256-1Noao/H+N8nFB4Beoy8fgwrcOQLVm9o4zKW1ODaqK9E=",
+        "lastModified": 1742972947,
+        "narHash": "sha256-o4PBZUmfxqxc1OzyWY/N4ftJgD5WhHPbKlBNwtP3P6g=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "48d12d5e70ee91fe8481378e540433a7303dbf6a",
+        "rev": "c019b5d0a53e2f1bcefdb4a631b997ca2adb8200",
         "type": "github"
       },
       "original": {
         "owner": "NixOS",
-        "ref": "release-24.11",
+        "ref": "master",
         "repo": "nixpkgs",
         "type": "github"
       }
diff --git flake.nix flake.nix
index 094df9776..d8bf23936 100644
--- flake.nix
+++ flake.nix
@@ -1,7 +1,7 @@
 {
   description = "The purely functional package manager";
 
-  inputs.nixpkgs.url = "github:NixOS/nixpkgs/release-24.11";
+  inputs.nixpkgs.url = "github:NixOS/nixpkgs/master";
 
   inputs.nixpkgs-regression.url = "github:NixOS/nixpkgs/215d4d0fd80ca5163643b03a33fde804a29cc1e2";
   inputs.nixpkgs-23-11.url = "github:NixOS/nixpkgs/a62e6edd6d5e1fa0329b8653c801147986f8d446";


$ nix build github:Mic92/nix-1/test-unstable
error: builder for '/nix/store/5jzhlzvcz53ivyckg4zi1a3r8g4y74z7-nix-util-tests-run.drv' failed with exit code 139;
       last 25 log lines:
       >
       > [----------] 7 tests from LevenshteinDistance/LevenshteinDistanceTest
       > [ RUN      ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/0
       > [       OK ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/0 (0 ms)
       > [ RUN      ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/1
       > [       OK ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/1 (0 ms)
       > [ RUN      ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/2
       > [       OK ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/2 (0 ms)
       > [ RUN      ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/3
       > [       OK ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/3 (0 ms)
       > [ RUN      ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/4
       > [       OK ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/4 (0 ms)
       > [ RUN      ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/5
       > [       OK ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/5 (0 ms)
       > [ RUN      ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/6
       > [       OK ] LevenshteinDistance/LevenshteinDistanceTest.CorrectlyComputed/6 (0 ms)
       > [----------] 7 tests from LevenshteinDistance/LevenshteinDistanceTest (0 ms total)
       >
       > [----------] Global test environment tear-down
       > [==========] 351 tests from 76 test suites ran. (61 ms total)
       > [  PASSED  ] 351 tests.
       >
       >   YOU HAVE 1 DISABLED TEST
       >
       > /private/tmp/nix-build-nix-util-tests-run.drv-0/.attr-0l2nkwhif96f51f4amnlf414lhl4rv9vh8iffyp431v6s28gsr90: line 2: 45373 Segmentation fault: 11  /nix/store/a49zhkfkvmy6wqdxnchr8wnc1hqnw2sd-exec/bin/exec /nix/store/382danbkwfh584vy6000ym88w8pnjsii-nix-util-tests-2.28.0pre/bin/nix-util-tests
       For full logs, run:
         nix log /nix/store/5jzhlzvcz53ivyckg4zi1a3r8g4y74z7-nix-util-tests-run.drv
error: 1 dependencies of derivation '/nix/store/55bpiab3q86ns78w1imr9djnszr8ibnb-nix-2.28.0pre20250326_866e193.drv' failed to build

It is very reproducible when switching nixpkgs in the nix flake. It breaks the test suite:

Thanks for the additional info. I will try to reproduce this locally.

Can you confirm that if you only create the override for tbb_2022_0 like I described that it does work?

I suspect what is happening is not due to an issue with tbb_2022_0 itself (since it works locally and works upstream on BLAKE3 CI) but some sort of configuration or linking issue that is only being triggered through master nixpkgs.

In this comment you mention this detail:

I added some patches that I needed to get this build with current nixpkgs unstable. However I think we are currently ending up with two tbb versions somehow?

Can you elaborate on the part about there being two versions of tbb?

If in fact there are multiple versions of tbb (or even maybe multiple separate copies of the same version) being linked somewhere along the way for nix, that could potentially cause a problem like this.

That could lead to multiple separate tbb schedulers involved during execution with unexpected results.

The solution may be to statically link tbb: uxlfoundation/oneTBB#297

I believe this should be possible by setting BUILD_SHARED_LIBS=OFF as an extra CMake option to libblake3, which should propagate to the tbb dependency.

Can you give that a try and see if makes a difference?

The downside to statically linking tbb is that it's usually not recommended to do this for libraries because of the possibility of duplicate schedulers again (if a consuming library or program also uses tbb). For an executable that's not an issue, and is what mold does.

It is very reproducible when switching nixpkgs in the nix flake. It breaks the test suite:

Thanks for the additional info. I will try to reproduce this locally.

Can you confirm that if you only create the override for tbb_2022_0 like I described that it does work?

Do you mean on top of your open pull request (NixOS/nix#12676)? if I applied to my unstable-nixpkgs branch, it has no effect. I also tried your pull request with the same result.

I suspect what is happening is not due to an issue with tbb_2022_0 itself (since it works locally and works upstream on BLAKE3 CI) but some sort of configuration or linking issue that is only being triggered through master nixpkgs.

Yes. If I apply this commit here, even is fixed again.

In this comment you mention this detail:

I added some patches that I needed to get this build with current nixpkgs unstable. However I think we are currently ending up with two tbb versions somehow?

Can you elaborate on the part about there being two versions of tbb?

I am not so sure, but I saw the old tbb version at some point in the stack trace. Haven't checked yet where it's injected.

If in fact there are multiple versions of tbb (or even maybe multiple separate copies of the same version) being linked somewhere along the way for nix, that could potentially cause a problem like this.

That could lead to multiple separate tbb schedulers involved during execution with unexpected results.

The solution may be to statically link tbb: uxlfoundation/oneTBB#297

I believe this should be possible by setting BUILD_SHARED_LIBS=OFF as an extra CMake option to libblake3, which should propagate to the tbb dependency.

Can you give that a try and see if makes a difference?

The downside to statically linking tbb is that it's usually not recommended to do this for libraries because of the possibility of duplicate schedulers again (if a consuming library or program also uses tbb). For an executable that's not an issue, and is what mold does.

It is very reproducible when switching nixpkgs in the nix flake. It breaks the test suite:

Thanks for the additional info. I will try to reproduce this locally.
Can you confirm that if you only create the override for tbb_2022_0 like I described that it does work?

Do you mean on top of your open pull request (NixOS/nix#12676)? if I applied to my unstable-nixpkgs branch, it has no effect. I also tried your pull request with the same result.

Oops, you are right. I forgot to add the override to the libblake3 override.

I made that change and now you can test with tbb_2022_0 without switching to unstable using this branch: https://github.com/silvanshade/nix/tree/blake3-tbb_2022_0

Once I do that, I get segfaults like you reported. So it a problem with the tbb_2022_0 not related to the nixpkgs channel.

I investigated further and went back and compiled the examples from the BLAKE3 repo with this configuration:

cmake --fresh -S c -B c/build -DBLAKE3_USE_TBB=1 -DBLAKE3_EXAMPLES=1 -DCMAKE_INSTALL_PREFIX=c/build/install -DCMAKE_EXPORT_COMPILE_COMMANDS=1 -DCMAKE_VERBOSE_MAKEFILE=1

I found that if I had my environment set up to use tbb_2022_0 installed from nix, I would get a segfault just with c/example.c, which doesn't even use the tbb update like c/example_tbb.c.

If I installed tbb instead through homebrew (which uses 2022), no segfaults with either c/example.c or c/example_tbb.c.

That indicates there is something wrong with the nixpkgs tbb_2022_0.

It may have something to do with the custom patches being applied but I haven't had a chance to test further.

I don't know how easy it will be to figure out what is wrong there so I'm fine with seeing this merged until a proper solution can be found.

But I think it might be better to switch to tbb_2021_11 for all platforms just to minimize any other potential issues.

nixpkgs-review result

Generated using nixpkgs-review.

Command: nixpkgs-review pr 393048

aarch64-darwin

Backport failed for release-24.11, because it was unable to cherry-pick the commit(s).

Please cherry-pick the changes locally and resolve any conflicts.

git fetch origin release-24.11
git worktree add -d .worktree/backport-393048-to-release-24.11 origin/release-24.11
cd .worktree/backport-393048-to-release-24.11
git switch --create backport-393048-to-release-24.11
git cherry-pick -x ee0fc3559eef448209dc7de4ec7344f7524bf98e

Backport: #393691