vulkan-sdk: 1.3.239 -> 1.3.243

[PedroHLC]

Description of changes

• Bumps to the latest tags as seen in https://vulkan.lunarg.com/doc/sdk/1.3.243.0/linux/release_notes.html

• One of the GLSLANG patches needed to be rebased, I've included it, and added comments mentioning the upstream PR (Use CMAKE_INSTALL_FULL_LIBDIR in compat cmake files KhronosGroup/glslang#3067).

Things done

On an x86_64-linux: Tested building vulkan-tools, which built vulkan-headers, vulkan-loader, spirv-headers, spirv-tools and glslang. Building vulkan-tools-lunarg builds vulkan-validation-layers. I've built vulkan-extension-layer, then spirv-cross and finished by running vkcube-wayland.

I didn't test the moltenvk update (Cc @reckenrode).

[Kranzes]

Do you have a script that updates all of these at once?

[PedroHLC]

Do you have a script that updates all of these at once?

Nope, are you considering writing one?

[Ralith]

LGTM. Wouldn't mind seeing shaderc 2023.3 in here as well, but not a blocker.

[Kranzes]

Do you have a script that updates all of these at once?

Nope, are you considering writing one?

I wasn't planning but I did end up writing one, check this out: https://github.com/Kranzes/nixpkgs/tree/vulkan-sdk-update-script

Tell me if you would like me to create a PR.

[PedroHLC]

Tell me if you would like me to create a PR.

You can do it, but my suggestions:

1. Don't push the version bumps, only the script. This way you can target master directly;
2. Instead of removing the version assertions, move them to meta.broken (e.g.: broken = version != spirv-headers.version;);
3. Add the script to vulkan-header's passthru.updateScript. Since it's the core of the version assertions;
4. Any other review/suggestion related to the script, we can do on said PR/draft to avoid delaying this PR though staging interactions.

[reckenrode]

Did some testing of MoltenVK this morning on Darwin. The changes LGTM.

[trofi]

Bisect says 6cc536e vulkan-headers: 1.3.239.0 -> 1.3.243.0 broke libplacebo (mpv depend) in staging as:

$ nix build --no-link -f. libplacebo -L
...
libplacebo> src/utils_gen.c: In function 'vk_res_str':
libplacebo> src/utils_gen.c:32:10: error: 'VK_ERROR_VALIDATION_FAILED' undeclared (first use in this function); did you mean 'VK_ERROR_VALIDATION_FAILED_EXT'?
libplacebo>    32 |     case VK_ERROR_VALIDATION_FAILED: return "VK_ERROR_VALIDATION_FAILED";
libplacebo>       |          ^~~~~~~~~~~~~~~~~~~~~~~~~~
libplacebo>       |          VK_ERROR_VALIDATION_FAILED_EXT

[Kranzes]

libplacebo seems to be pretty outdated, we could try updating it.

Edit: #224795

[PedroHLC]

Bisect says 6cc536e vulkan-headers: 1.3.239.0 -> 1.3.243.0 broke libplacebo (mpv depend) in staging as:

$ nix build --no-link -f. libplacebo -L
...
libplacebo> src/utils_gen.c: In function 'vk_res_str':
libplacebo> src/utils_gen.c:32:10: error: 'VK_ERROR_VALIDATION_FAILED' undeclared (first use in this function); did you mean 'VK_ERROR_VALIDATION_FAILED_EXT'?
libplacebo>    32 |     case VK_ERROR_VALIDATION_FAILED: return "VK_ERROR_VALIDATION_FAILED";
libplacebo>       |          ^~~~~~~~~~~~~~~~~~~~~~~~~~
libplacebo>       |          VK_ERROR_VALIDATION_FAILED_EXT

Sorry, this is not the first time libplace broke with a vulkan-headers bump, I should have tested it.

[vcunat]

Also mangohud regression looks like caused by this PR: https://hydra.nixos.org/build/215271636
Maintainers: @kira-bruneau, @zeratax