Vulkan-Hpp icon indicating copy to clipboard operation
Vulkan-Hpp copied to clipboard

Vulkan-Headers CI tests failing against current Vulkan-Docs branch

Open oddhack opened this issue 4 months ago • 9 comments

There are a bunch of errors raised in vulkan_structs.hpp about "vk::DeviceCreateInfo:: must be initialized by mem-initializer in 'constexpr' constructor". This is holding up the Vulkan-Headers update to match to spec update so would appreciate help @asuessenbach . See e,g,

https://github.com/KhronosGroup/Vulkan-Headers/actions/runs/16665329727/job/47170884324

oddhack avatar Aug 01 '25 12:08 oddhack

Seems related to #2230, where the deprecated member init was removed from the ctor to avoid the deprecation warning. #2235 reintroduces these, but might end up triggering said deprecation warnings.. we shall see

M2-TE avatar Aug 01 '25 13:08 M2-TE

@oddhack I think, I fixed this issue with #2236. But the build (cmake-unix (ubuntu-latest, 3.22.1) is still failing!

I don't know, how the generation of the Vulkan-Hpp files could be triggered there.

asuessenbach avatar Aug 04 '25 15:08 asuessenbach

I don't know, how the generation of the Vulkan-Hpp files could be triggered there.

The build is not triggered there AFAIK, unless some of the LunarG folks are doing something I wasn't aware of. The files are built from Vulkan-Docs as part of the release process and just copied into Vulkan-Headers as an aggregation and publication mechanism. I will try regenerating again.

oddhack avatar Aug 04 '25 15:08 oddhack

Compilation of the headers is now failing in the clang++-18 portion of the tests for Vulkan-Headers: https://github.com/KhronosGroup/Vulkan-Headers/actions/runs/18392928205/job/52406731689?pr=572 .

oddhack avatar Oct 10 '25 00:10 oddhack

I am going to go ahead and publish the Vulkan-Headers update as the problem (seems) confined to a fringe case of the experimental modules with an older C++ version.

oddhack avatar Oct 10 '25 00:10 oddhack

We will address CI asap, now that #2303 has been merged. Indeed, failure with module runs can be ignored for now!

M2-TE avatar Oct 10 '25 07:10 M2-TE

@oddhack this failure will be resolved by KhronosGroup/Vulkan-Headers/#569.

sharadhr avatar Oct 10 '25 11:10 sharadhr

Important to note: "resolved" as in, it won't test modules for a short while

M2-TE avatar Oct 10 '25 11:10 M2-TE

Indeed. I think it's best to keep testing/building within this repository, including the experimental UUID and module setup. All this will naturally be removed once CMake removes that UUID.

Vulkan-Headers will install the module interface files regardless.

sharadhr avatar Oct 10 '25 11:10 sharadhr