Re: [OE-core] [PATCH 3/3] vulkan-validation: Add the official Vulkan validation layers

[Jordan Crouse <jorcrous@amazon.com>]

On Fri, Nov 11, 2022 at 09:03:02AM +0100, Alexander Kanavin wrote:
> On Thu, 10 Nov 2022 at 22:30, Jordan Crouse <jorcrous@amazon.com> wrote:
> > ERROR: vulkan-validation-1.3.231.1-r0 do_package_qa: QA Issue: -dev package vulkan-validation-dev contains non-symlink .so '/usr/lib/libVkLayer_khronos_validation.so' [dev-elf]
> >
> > I borrowed the QA inhibit from vulkan-loader (since they are built
> > mostly the same) but if there is a different path forward, that would be
> > great.
> 
> Thanks. I am not seeing QA inhibits or any special handling in the
> current vulkan-loader:
> https://git.yoctoproject.org/poky/tree/meta/recipes-graphics/vulkan/vulkan-loader_1.3.231.1.bb?h=master-next

That was my mistake, I guess I don't clearly remember why these were
added in our downstream tree - clearly a poorly executed effort at
getting rid of the error.

> Can you find out why the /usr/lib/libVkLayer_khronos_validation.so is
> built unversioned and where in the source tree that decision is made?

These libraries (plugins, really) are dynamically loaded by the vulkan-loader at runtime. 

I couldn't immediately find an standard online but a cursory glance of the vulkan
drivers/layers from other distros use a .so format, for example:

https://packages.debian.org/buster/amd64/mesa-vulkan-drivers/filelist

And from mesa in OE:

https://git.openembedded.org/openembedded-core/tree/meta/recipes-graphics/mesa/mesa.inc#n304

I agree that the existing INSANE_SKIP was the wrong way to go about it -
perhaps the better option would be to do this:

FILES:${PN} = "${libdir}/*.so ${datadir}/vulkan"

I think ${PN}-dev would nominally be empty anyway - the two
outputs of this package are .so layers and a .manifest file for the
loader.

Jordan