blender: 5.0.1 -> 5.1.0

[frozolotl]

Changelog

Blender documentation on library changes:

• https://projects.blender.org/blender/blender/pulls/149852
• https://projects.blender.org/blender/blender/issues/147201

Things done

• Built on platform:
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• Tested, as applicable:
  • NixOS tests in nixos/tests.
  • Package tests at passthru.tests.
    Appears to segfault independently of this change.
  • Tests in lib/tests or pkgs/test for functions and "core" functionality.
• Ran nixpkgs-review on this PR. See nixpkgs-review usage.
• Tested basic functionality of all binary files, usually in ./result/bin/.
• Nixpkgs Release Notes
  • Package update: when the change is major or breaking.
• NixOS Release Notes
  • Module addition: when adding a new NixOS module.
  • Module update: when the change is significant.
• Fits CONTRIBUTING.md, pkgs/README.md, maintainers/README.md and other READMEs.

nixpkgs-review result

Generated using nixpkgs-review.

Command: nixpkgs-review
Commit: 70118d4b59ca11af7a2e4b75fc18310db2108e99

---

x86_64-linux

✅ 2 packages built:

• blender
• blendfarm

[amarshall]

Successfully tested CUDA build and render on x86_64-linux, but ROCm is broken.

[frozolotl]

Successfully tested CUDA build and render on x86_64-linux, but ROCm is broken.

Thanks. ROCm is fixed now, as far as I can tell.
My machine does not support HIPRT, so that might be something to check still.

[drupol]

Diff LGTM, I was wondering if submitting the patch upstream would make sense.

[amarshall]

Were you able to perform an HIP (non-RT) Cycles GPU render? It segfaults for me, but it also segfaults on 5.0.1 so…

Blender HIP RT has been broken for a while in Nixpkgs (it can’t properly detect that it’s supported).

[frozolotl]

Were you able to perform an HIP (non-RT) Cycles GPU render? It segfaults for me, but it also segfaults on 5.0.1 so…

Oh, that's good to know! I actually am too suffering from that same segfault you are talking about. I originally assumed it was an issue with my GPU because no one else seemed to have that issue from what I could tell.
I managed to test this PR because system-wide I still had ROCm 7.1.1 in use, which Blender did use.

I have been debugging this a bit since I published this PR. Let me share what I know:
I know that ROCm 7.1.1 works, while 7.2.0 does not. ROCm 7.2.1 doesn't fix the issue (I will create a PR once I was able to make things work).

Looking at the backtrace, it seems that rocm-comgr calls rocm-llvm, which—surprisingly—calls LLVM-lib 19 (even though it should be LLVM 22.0.0!). The reason for this is that OpenShadingLanguage imports LLVM-lib 19.
Switching to LLVM 22.1.0 does not fix the issue. However, disabling OpenShadingLanguage does!

I'll be back later. Hopefully I'll find a fix that does not involve disabling OSL.

---

If anyone needs a temporary fix, switching true to false here should be all that's needed (of course, at the cost of some functionality).
Alternatively, reverting #484792 would likely also do it.

[amarshall]

I think this PR then is fine since it’s not breaking anything “new”. It might make sense to add an additional commit to e.g.

diff --git a/pkgs/by-name/bl/blender/package.nix b/pkgs/by-name/bl/blender/package.nix
index 91df7b38b283..0605cf7712fe 100644
--- a/pkgs/by-name/bl/blender/package.nix
+++ b/pkgs/by-name/bl/blender/package.nix
@@ -176,7 +176,7 @@ stdenv'.mkDerivation (finalAttrs: {
     (lib.cmakeBool "WITH_CYCLES_DEVICE_ONEAPI" false)
     (lib.cmakeBool "WITH_CYCLES_DEVICE_OPTIX" cudaSupport)
     (lib.cmakeBool "WITH_CYCLES_EMBREE" embreeSupport)
-    (lib.cmakeBool "WITH_CYCLES_OSL" true)
+    (lib.cmakeBool "WITH_CYCLES_OSL" (!rocmSupport))
     (lib.cmakeBool "WITH_SYSTEM_GLOG" true)
     (lib.cmakeBool "WITH_HYDRA" openUsdSupport)
     (lib.cmakeBool "WITH_INSTALL_PORTABLE" false)

to at least unbreak HIP.

[drupol]

Perhaps we could do it in a new PR so we can move on in this PR faster?

[frozolotl]

I issued #504149, which solves the segfault.

[amarshall]

Wait, why did this target staging? It can go to master.

[frozolotl]

Wait, why did this target staging? It can go to master.

Oops, that was not my intention 😅

[frozolotl]

A quick note on the Eigen patch: there has been a PR to Blender that would fix the build.
https://projects.blender.org/blender/blender/pulls/156417

[amarshall]

Cherry-pick to master #505352