TestBot News

Francois Gouget fgouget at codeweavers.com
Mon Sep 30 15:55:08 CDT 2019



I updated the debian10 and w1064 VMs so let me know if there are 
problems. For now they are not base VMs yet, I'ĺl promote them after 
their first WineTest run.

Here's what's new (tl;dr;-style):

debian10: OSS4, dual-screen, Spice vdagent, ich9 sound card.
          (and likely wrong CPU)

w1064: New locale snapshots, a dual-screen snapshot (QXL), metered 
       network connection.

cw-rx460: New AMD RX 460 GPU
          - Debian: OSS4, Vulkan, win32 trouble.
          - Windows: New graphics driver, partial Windows upgrades, some 
            missing disk images.


The TestBot feels a bit slow today, maybe because it fell behind while I 
was working on the VMs. If it continues being slow I may have to 
rebalance the VMs a bit.



debian10
--------

This VM now has two screens. A side effect is that screenshots will 
sometimes not work. That's only one of many annoyances with multi-head 
support for Unix guests. You can read all about it there: 
https://bugs.winehq.org/show_bug.cgi?id=47844

With Unix guests, multi-head support requires installing and running 
Spice's VDAgent in the guest. The VDAgent servers are meant to integrate 
the guest environment with the host to simplify copy/paste between them, 
resizing the guest desktop by resizing the Spice window, etc. Hopefully 
none of that will interfere with the Wine tests but if it does, let me 
know.

Also if we need to test both single and dual screen configurations then 
we'll need to duplicate that VM. So let me know if that's needed.


In the process upgraded the virtual hardware the VM runs on from QEmu's 
i440FX to the Q35 platform which is the default nowadays. This should 
not have any impact on the tests except that the sound card changed from 
ich6 to ich9.

Last thing, when rebuilding the VM I picked the 'Hypervisor default' for 
the emulated CPU. It's missing support for sse3 which, if I recall 
correctly, was causing bcrypt failures on Wine. And sure enough the 
failures are back. So I probably need to tweak the CPU configuration.


w1064
-----

This VM has a new snapshot with a dual screen configuration. On windows 
this is only possible with the QXL graphics card whereas we normally use 
the VGA one. So this is likely to have an impact on the Direct 3D tests 
in that snapshot.
Details: https://bugs.winehq.org/show_bug.cgi?id=47838

On Windows too the dual-screen configuration may cause the TestBot 
to 'fail' to produce a useful screenshot from time to time:
https://testbot.winehq.org/JobDetails.pl?Key=57142&s205=1#k205

Also one needs to have one graphics cards per screen. So this is 
slightly different from the debian10 setup where there is a single 
graphics card with two screens.


I also added five new locales: Arabic, French, Hebrew, Japanese and 
Simplified Chinese.

Recreating the live snapshots was almost a breeze thanks to the recent 
TestBot work. The snag is that it turns out setting the location to a 
European country causes Windows 10 to through a 'Privacy wizard' when 
rebooting to have the new locale take effect. Of course that impacted 
the first locale I tested, French, and broke the automatically created 
live snapshot. Fortunately now I know what's causing the problem and 
only one snapshot needs to be created manually.

Eventually I will probably demote the wvistau64 locale configurations to 
only run the daily WineTests or even just deal with jobs that developer 
manually pick for them.

Another change I made is to put the network connection in metered mode. 
Before this I was blocking the Windows Updates simply by disabling the 
corresponding service. That worked fine in version 1507 but did not work 
so well with more recent versions. In particular a reboot would reenable 
the Windows Update service leading to upgrades takign place.

Putting the network connection in metered mode seems to reliably stop 
Windows update in its tracks and is now easy in 1809 (and a bit more 
hidden in 1703). However I probably went overboard by setting a 1GB 
limit (a 100GB limit seems to work just as well) so there are 
notifications that the quota is exceeded. Let me know if this change is 
causing problems.
Details: https://bugs.winehq.org/show_bug.cgi?id=47854

By the way only the new dual-screen and locale snapshots have the metere 
network connection. So comparing the results there with those of the old 
snapshots can help detect issues. If this is not causing trouble I'll 
eventually go back and do the same change in the other snapshots.


cw-rx-460
---------

Not quite part of the TestBot but provides daily real hardware WineTest 
results. The AMD HD6800 graphics card died so it got replaced by an AMD 
RX 460 graphics card.

On the Debian side I added OSS4 (Faudio is present too) and the new 
graphics card gains us Vulkan support. Yay!

The WineTest results are not great. In particular the wow32 results look 
ok, but half the time WineTest / the system seems to crash in the middle 
of the win32 ones. It may just be that it reboots to Windows at the 
wrong time. The rest of the time the win32 results are on par with the 
wow32/wow64 ones. I did not have time to dig into this yet.


On the Windows side the graphics driver has to be updated for each test 
configuration of course :-( And while that happens Windows Update starts 
applying updates :-((

But if the metered network connection trick works out I should be 
able to make clean updates to the windows images. In the meantime only 
the Windows 8.1 and Windows 10 1809 configurations are operational.


-- 
Francois Gouget <fgouget at codeweavers.com>


More information about the wine-devel mailing list