Re: Mesa vulkan layer path fix for core-updates

[Kaelyn]

Hi,

------- Original Message -------
On Tuesday, May 9th, 2023 at 4:51 AM, John Kehayias 
<john.kehayias@protonmail.com> wrote:

> 
> Hello,
> 
> On Tue, Apr 25, 2023 at 04:40 PM, Kaelyn wrote:
> 
> > Hi,
> > 
> > ------- Original Message -------
> > On Tuesday, April 25th, 2023 at 2:15 PM, Andreas Enge andreas@enge.fr wrote:
> > 
> > > Hello Kaelyn,
> > > 
> > > thanks for your research!
> > 
> > You're welcome! :)
> > 
> > > Am Wed, Apr 19, 2023 at 04:07:51PM +0000 schrieb Kaelyn:
> > > 
> > > > * https://issues.guix.gnu.org/62176 can be closed when
> > > > core-updates is merged, since core-updates contains mesa 22.2.4
> > > > * Though not exactly mesa-related,
> > > > https://issues.guix.gnu.org/61364 can possibly be closed now, and
> > > > almost certainly once the core-updates merge is completed. (The
> > > > ticket is a number of workarounds the user applied in early Feb to
> > > > be able to build their system profile using core-updates, to pick
> > > > up Mesa 22 for newer hardware support. I'm not sure if any of the
> > > > patches are still relevant.)
> > > 
> > > I just closed these two.
> > > 
> > > > * https://issues.guix.gnu.org/58887 looks like an alternate way of
> > > > solving the layer path issues that
> > > > https://issues.guix.gnu.org/59453 also addresses. Additionally, it
> > > > adds two new nonstandard VK_*_PATH variables to vulkan-loader,
> > > > with at least VK_ILAYER_PATH seeming very similar in functionality
> > > > to VK_LAYER_PATH and VK_ADD_LAYER_PATH described at
> > > > https://github.com/KhronosGroup/Vulkan-Loader/blob/main/docs/LoaderInterfaceArchitecture.md
> > > > * https://issues.guix.gnu.org/58251 would be fixed by
> > > > https://issues.guix.gnu.org/59453
> > 
> > I feel these two can be closed once 59453 lands, as then the manifests
> > will have the store path to their corresponding shared objects. I also
> > think having the full paths in the manifests will lead to fewer
> > cross-version/cross-package mixing of objects, compared to setting and
> > using environment variables of directories to search.
> 
> 
> I haven't looked at the details, but I'm guessing these can all be
> closed now?

Yes they can be. As a quick smoke test, prior to these patches landing, 
"vulkaninfo > /dev/null" would consistently print out vulkan loader errors 
about not being able to open libVkLayer_MESA_device_select.so. (I call it a 
smoke test because that layer not being found isn't fatal; I first encountered 
the errors in a breaking way when packaging vulkan-validationlayers for use 
with the tutorial at https://vulkan-tutorial.com/.)

Slightly off-topic speaking of mesa bugs: another one that can be closed is 
https://issues.guix.gnu.org/43849 as the issue with mesa's reproducibility was 
fixed with a patch that landed in meson 0.59.0 (which Guix picked up in commit 
b15c3dd9b0 from July 2021).

> 
> > > > * https://issues.guix.gnu.org/62313 might need a modification to
> > > > mesa e.g. to add VDPAU_DRIVER_PATH as a native-search-path (one
> > > > possible solution; in my home profile I made VDPAU work by setting
> > > > "VDPAU_DRIVER_PATH=/run/current-system/profile/lib/vdpau").
> > > > * https://issues.guix.gnu.org/48868 appears to be the same
> > > > VDPAU_DRIVER_PATH issue as https://issues.guix.gnu.org/62313.
> > 
> > For the VDPAU drivers, I plan to do a little more digging and some
> > experimenting but I suspect defining VDPAU_DRIVER_PATH as a
> > native-search-path is the best way forward. I'll send a patch once
> > I've tested a change locally without having my profile set
> > VDPAU_DRIVER_PATH to /run/current-system/profile/lib/vdpau.
> 
> 
> I checked that both 48868 and 62313 were fixed in the recent updates
> and closed both.
> 
> Thanks for the patches!
> John

You're welcome! And thank you for checking and resolving the bug reports!

Cheers,
Kaelyn