Re: [PATCH 3/6] oe-selftest: extend virgl gtk test to also check the SDL option

From: Michael Halstead <mhalstead@linuxfoundation.org>
To: Alexander Kanavin <alex.kanavin@gmail.com>
Cc: OE-core <openembedded-core@lists.openembedded.org>
Subject: Re: [PATCH 3/6] oe-selftest: extend virgl gtk test to also check the SDL option
Date: Fri, 8 Nov 2019 09:20:04 -0800	[thread overview]
Message-ID: <CADfgfoYap0Z83h-mRXOLvX2kkC8xitN1Mx5wQM4uHugRhG15Nw@mail.gmail.com> (raw)
In-Reply-To: <CANNYZj-Cmx_iNn+tJAW4XE=VSi4MT+LeiaPQk-8REhG+Xd_JsA@mail.gmail.com>

[-- Attachment #1: Type: text/plain, Size: 2912 bytes --]

On Mon, Nov 4, 2019 at 1:11 PM Alexander Kanavin <alex.kanavin@gmail.com>
wrote:
>
> On Sat, 2 Nov 2019 at 23:29, Alexander Kanavin <alex.kanavin@gmail.com>
wrote:
>>
>> Same failures on the Debian 10 worker:
>>>>
>>>>
>>>> https://autobuilder.yoctoproject.org/typhoon/#/builders/56/builds/778
>>>
>>>
>>> runqemu - ERROR - Failed to run qemu: Xlib:  extension "RANDR" missing
on display ":1".
>>> qemu-system-x86_64: ../libepoxy-1.5.3/src/dispatch_common.c:863:
epoxy_get_proc_address: Assertion `0 && "Couldn't find current GLX or EGL
context.\n"' failed.
>>>
>>> Thanks - the Gtk part is passing fine, so it's something SDL does that
upsets the X/GL stack on the Debian 10 host.
>>> I am reluctant to disable the SDL part of the virgl tests on Debian 10,
as it is a new distro (unlike centos 7).
>>>
>>> I'll try to see if I can run Debian 10 in a VM here and try to
reproduce. Or is it possible to debug directly on a Debian 10 worker?
>>
>>
>> So I actually went ahead, and installed Debian 10 into a qemu image,
then transferred a pre-populated build directory into it, and ran runqemu
there against tigervnc (hurray for the nested kvm feature!).
>>
>> Both 'runqemu kvm sdl gl' and 'runqemu kvm gtk gl' work fine, including
running kmscube!
>>
>> So I'd like to see what packages are installed on the Debian 10 worker
vs. my Debian 10 installation.
>>
>> Can you issue 'dpkg -l' on the worker, and send me the output, please?
Maybe something is missing?
>
>
> After additional digging I reproduced this. The culprit is the outdated
VNC server implementation that runs on the Debian 10 autobuilder (and maybe
others as well).
>
> Specifically, it's tightvncserver, where all Linux development has ceased
10 years ago (!).
> https://www.tightvnc.com/
>
> If you replace that with tigervnc (a modern, supported fork of tightvnc),
then the tests pass fine.
> https://tigervnc.org/
>

TigerVNC is now installed from the standard repos where available.

>
> As Fedora has already obsoleted tightvnc in favor of tigervnc, I think we
should do the same on all debian machines (debian provides both tightvnc
and tigervnc, but treats them as equal).
> https://src.fedoraproject.org/rpms/tightvnc/blob/master/f/dead.package
> I also checked that tightvnc is not available for opensuse either.

Server components of TigerVNC are in xorg-x11-Xvnc package which is
installed on all the openSUSE workers so I expect the VNC server there to
work. Please let me know if they do not.

On Ubuntu 16.04 and Debian 8 workers I've installed from
https://bintray.com/tigervnc/stable/tigervnc/1.9.0. This should allow tests
to complete. Installing packages outside of the distribution's package
manager might invalidate these as sanity tested distros. I'd appreciate
feedback about this.

--
Michael Halstead
Yocto Project / SysAdmin

>
> Alex

[-- Attachment #2: Type: text/html, Size: 3777 bytes --]