Test runs but does not restore correct monitor setup

[Danie10]

Ran tests on Manjaro Linux, and after doing a force reinstall of PyMonCtl. Reason is I was testing the newest version of OBS Zoom and Follow script, which was giving an error ModuleNotFoundError: No module named 'pymonctl' in OBS. But I see that PyMonCtl is indeed running. So I decided to run the test script for PyMonCtl with python test_pymonctl.py.

I do see it going through the various changes on each monitor. At some point, though, it mirrored my monitor 2 to the display on monitor 3. When it exits the test run, it ended with this output below, but had lost the original config of my monitor order, and monitor 2 was off and no longer recognised by the OS (I imagine it will restore after a reboot):

IS ATTACHED?: True
DETACH
IS ATTACHED?: False
ATTACH
IS ATTACHED?: True

NAME DP-4
HANDLE/ID: 475
IS PRIMARY: False
SIZE: None
POSITION: None
Traceback (most recent call last):
  File "/home/danie/Downloads/OBS Studio/PyMonCtl-master/tests/test_pymonctl.py", line 38, in <module>
    print("FREQUENCY:", monitor.frequency)
  File "/home/danie/.pyenv/versions/3.10.11/lib/python3.10/site-packages/pymonctl/_pymonctl_linux.py", line 366, in frequency
    crtcInfo = randr.get_crtc_info(display, outputInfo.crtc, res.config_timestamp)
  File "/home/danie/.pyenv/versions/3.10.11/lib/python3.10/site-packages/Xlib/ext/randr.py", line 758, in get_crtc_info
    return GetCrtcInfo (
  File "/home/danie/.pyenv/versions/3.10.11/lib/python3.10/site-packages/Xlib/protocol/rq.py", line 1368, in __init__
    self.reply()
  File "/home/danie/.pyenv/versions/3.10.11/lib/python3.10/site-packages/Xlib/protocol/rq.py", line 1388, in reply
    raise self._error
Xlib.error.XError: <class 'Xlib.error.XError'>: code = 148, resource_id = 0, sequence_number = 307, major_opcode = 140, minor_opcode = 20

[Kalmat]

Hi! Thank you for your feedback!

Worry not. The missing monitor will be there again after reboot (or if you just unplug and re-plug it to your system). The config, unfortunately, will not be restored. Running the tests will end up with a totally new config every single time (even though it doesn't crash).... Good point to work on (I mean, restore initial config when tests finisth)!

Besides, the tests are built for a 2-monitor setup (unfortunately I have no access to three monitors), so the weird behavior you have detected (like mirroring) may be caused by this.

I have not tested in Manjaro Linux (I will try to install it on my system. Which desktop manager / window manager are you using?), but perhaps Manjaro is assigning a new monitor handle after attaching / detaching a monitor (Windows works this way), so this is why it doesn't find it anymore and crashes.

In short:

• If you are interested in using PyMonCtl as a separated module, we can work on specific tests/fixes in your case (I will, anyway for future cases)
• If you are interested in OBS-zoom-and-follow only, dont' worry about PyMonCtl tests, just test OBS-zoom-and-follow and check if everything works. Bear in mind that OBS-zoom-and-follow script itself has not been prepared for multi-monitor setups until very recently. See this. If this is the case and you detect any issue, just let me know!

[Danie10]

Thanks @Kalmat yes you are right. It seemed to assume two monitors. Did not work after reboot, but I did get it fixed. The script could maybe test for a monitor count before running, and maybe trying to read the existing config to save.

I'm using Manjaro KDE. Yes, I'm really using it on OBS (although I see it has really impressive monitor control).

Why I ran your tests was that OBS was reporting it could not find PyMonCtl, but it is definitely installed and 'works'. But it is probably an issue then I should report on that project (the not finding).

[Kalmat]

It didn't detect the monitor after reboot??? Well... I definitely have to test that by my own! I also have to test the crash after detaching/attaching a monitor (I guess the monitor handle changes, like in Windows).

I am working in a "more-than-two" tests script, which in addition restores the original config right before finishing (if it crashes... it will not be able, unfortunately).

Thank you for your feedback again!

[Kalmat]

Hi @Danie10! I have uploaded a new version (v0.6) to github (not to PyPi yet since I have to make more tests). This new version should take into account 2 or more monitors, and restore initial monitors config. I have tested it on Kubuntu (which shares the KDE environment with your setup) and it doesn't crash... If you have the time it will help me a lot if you could test this new version on your system to try to diagnose any issue that may arise.

Thank you!

[Danie10]

@Kalmat OK still working as expected. I did the --force-reinstall again as the version had not changed, and downloaded the code to be sure I run the correct test script. It starts off fine on my 3rd monitor, then moves to the 2nd monitor, but ended with only my 1st monitor visible as other two had their resolution set to None. I recovered now by just running Nvidia-Settings and re-enabling the resolutions.

I then tried running the script again but directing output to a file. Problem this time is it next finished the script. It seemed to just stop at some point. I did a CTRL-C to exit, and included that terminal output at the end of the attached file. Maybe this helps shed some light.

I am using a single Nvidia RTX-2060 GPU with the monitors connected to available outputs, so it is a mix of one HDMI and two DP connectors.

Attached is the output file and a screenshot of my Nvidia layout.

pymonctl-output.txt

[Kalmat]

Thank you so much!!! This is pure gold!

I think that attaching / detaching a monitor somehow affects the system monitor management (the error actually comes from Xlib itsefl). I will definitely have to install Manjaro/KDE to find a solution.

Thank you again!

[Kalmat]

Hi again! Just in case you're curious, PyMonCtl v0.7 is now working on my Manjaro/KDE system. Turns out that Manjaro/KDE was not managing monitors as automatically as Ubuntu/GNOME does... after some tweaking, it should be working OK now!

Thank you for your feedback and help!

[Danie10]

The test script is showing the monitor count and primary monitor correctly, but fails with error below before it starts to do the tests:

python test_pymonctl.py
MONITORS COUNT: 3
PRIMARY MONITOR: DP-1

Traceback (most recent call last):
  File "/home/danie/Downloads/OBS Studio/PyMonCtl-0.7/tests/test_pymonctl.py", line 24, in <module>
    monDict = pmc.getAllMonitorsDict()
  File "/home/danie/.pyenv/versions/3.10.11/lib/python3.10/site-packages/pymonctl/_main.py", line 75, in getAllMonitorsDict
    return _getAllMonitorsDict()
  File "/home/danie/.pyenv/versions/3.10.11/lib/python3.10/site-packages/pymonctl/_pymonctl_linux.py", line 54, in _getAllMonitorsDict
    wx, wy, wr, wb = wa[0], wa[1], wa[2], wa[3]
TypeError: 'NoneType' object is not subscriptable

[Kalmat]

Thank you for helping out! Really bad news. In my system (an actual Manjaro/KDE with 2 monitors, not a VM), it works smoothly, but in yours it seems to be failing to retrieve WORKAREA property... the question is why! It didn't fail before, right? But I didn't change that part... really weird to me. Totally lost.

[Danie10]

Yes it's odd. Three monitors should not make a difference. I'm using X11 (not Wayland) with Nvidia GPU, on AMD CPU. And yet previously it was actually going through all the tests. I did do a force reinstall and saw it confirmed the version (as it also shows in the errors).

I vaguely remember with something showing the available resolutions, that there was one that maybe said not available.

[Kalmat]

No, three monitors should not cause the problem. What it is really weird is that it worked before, but not now. That part is identical, and basically consists in a very low-level Xlib function. Another unexplainable point is that it works OK in my Manjaro/KDE, Ubuntu/GNOME, KUbuntu (KDE) and Mint/Cinnamon installs (two monitors, on actual installs for the first two OS, and VMs for the last two ones).

Just shooting in the air:

• Did you change the monitors config recently? Can you try with other configs?
• Before testing, uninstall pywinbox, pywinctl and pymonctl; and physically delete ewmhlib package folder within site-packages folder; then install pymonctl only using pip, and run test_pymonctl.py (all those modules share ewmhlib, which is in charge of accessing Xlib primitives... perhaps one of them has a "bad" version of it in your case).
• Are you using private NVidia drivers or the ones provided by Manjaro out-of-the-box (just curioous: do you know if that is nouveau or mesa or similar?). I have not installed NVidia drivers, so I guess I am using the default ones. I can install NVidia drivers but, since it worked before in your system, I think it will not be the problem.
• If you have time, could you please comment a few lines in test_pymonctl.py, and test again to check what happens (it will likely crash somewhere else)?:

monDict = pmc.getAllMonitorsDict()   # The crash is inside this function, retrieving the WORKAREA property
for mon in monDict:
    print(monDict[mon])
print()

Those lines are at the beginning of the script, right after the "PRIMARY MONITOR" print.

Besides, I can prepare a debug wheel, but this can be a back and forth process, so I don't want to ask you too much abusing your time and patience!

Thanks a lot for your help so far!

[Danie10]

OK interesting... I took a little longer as I was also cleaning up my Python - some packages had shown missing dependencies.

1. After deleting and reinstalling the three packages, and removing the ewmhlib package folder, it still crashed at the beginning.
2. I then commented out that code, and it ran for a few minutes working on each monitor, but did hang eventually, and I had to force logout to get back control. It had disabled the third monitor and could not recover at that point.
3. I ran it again but directed output to the output.txt file. Actually it went all the way through and finished the test script. All monitors were actually on, but my mouse was confined to the right most monitor only. I could though open Nvidia-Settings and restored everything.
4. I remembered I've had issues with the compositor, so I disabled that, and then ran it again directing output to the output-nocompositor.txt file. It did not finish though but hung at some point, and I noted the Xorg process was consuming a good 7% CPU. Mouse was again confined to 3rd monitor, but although I could use shortcuts to open a terminal screen, there was no response inside the terminal, so I had to force logout again. 3rd monitor though was still active.

Both files are attached, and the screenshot of the processes at end of 4 above.

I'm using the Nvidia proprietary drivers. No monitor changes done apart from restoring from previous tests. But yes there is one odd thing on my setup, and that is when some functions call the compositor like rectangular area screenshots, or opening the compositor settings, it will hang for 30 secs to a minutes, and then respond. Which is why I ran a test too with the compositor disabled.

Issue seems to somehow be with the 3rd monitor. Although earlier in the tests that switches on and off and adjusts contrast etc.

output-nocompositor.txt
output.txt