Xpra: Ticket #1460: Paint bug with latest trunk (r15230)

Testing with a Fedora 25 client and server using Trunk 2.X r15230:

Navigating to Google Maps using Chrome and moving the map around causes Chrome to stop updating paints properly, and the logs print out this error:

2017-03-07 15:02:05,593 Error during scrolling detection
2017-03-07 15:02:05,593  with image=XShmImageWrapper(BGRX: 5, 81, 1106, 750), options={'scroll': True}
Traceback (most recent call last):
  File "/usr/lib64/python2.7/site-packages/xpra/server/window/window_video_source.py", line 1631, in do_video_encode
    return self.encode_scrolling(scroll_data, image, options)
  File "/usr/lib64/python2.7/site-packages/xpra/server/window/window_video_source.py", line 1502, in encode_scrolling
    sub = image.get_sub_image(0, sy, w, sh)
  File "xpra/x11/bindings/ximage.pyx", line 371, in xpra.x11.bindings.ximage.XImageWrapper.get_sub_image (xpra/x11/bindings/ximage.c:4276)
Exception: invalid sub-image height: 0+836 greater than image height 750

Tue, 07 Mar 2017 23:19:47 GMT - J. Max Mena: attachment set

attachment set to 1460paintbug.png

Tue, 07 Mar 2017 23:29:48 GMT - J. Max Mena:

The screenshot I attached is of what I'm seeing in Chrome (with the --show-paint-rects flag).

Tue, 07 Mar 2017 23:31:04 GMT - J. Max Mena: attachment set

attachment set to 1460info.txt

Xpra info with the stuck window. (Window is the one titled Hanover, 71)

Wed, 08 Mar 2017 12:34:42 GMT - Antoine Martin: owner changed

owner changed from Antoine Martin to J. Max Mena

Running with "-d scroll" quickly uncovered the problem:

encode_scrolling(XShmImageWrapper(BGRX: 7, 105, 2282, 1774), {'scroll': True}) window-dimensions=(2294, 1879)
too many items: 2 scrolls, 38 non-scrolls - sending just one image instead
 will send scroll data={}, non-scroll={0: 1879}
Error during scrolling detection
 with image=XShmImageWrapper(BGRX: 7, 105, 2282, 1774), options={'scroll': True}
Traceback (most recent call last):
  File "/usr/lib64/python2.7/site-packages/xpra/server/window/window_video_source.py", line 1631, in do_video_encode
    return self.encode_scrolling(scroll_data, image, options)
  File "/usr/lib64/python2.7/site-packages/xpra/server/window/window_video_source.py", line 1502, in encode_scrolling
    sub = image.get_sub_image(0, sy, w, sh)
  File "xpra/x11/bindings/ximage.pyx", line 371, in xpra.x11.bindings.ximage.XImageWrapper.get_sub_image (xpra/x11/bindings/ximage.c:4276)
Exception: invalid sub-image height: 0+1879 greater than image height 1774

When we hit "too many items", we were using the wrong dimensions. Fixed in r15231 - which also makes the whole thing more robust.

Fri, 10 Mar 2017 17:39:12 GMT - J. Max Mena:

Upped server and client to r15267:

I no longer see the error but I'm still not getting the paint updates client-side. I know the server is doing the calculations proper and there are client-side paints going on (XPRA_OPENGL_PAINT_BOX=1, but I'm not seeing the paints properly. (Will attach a screenshot)

It seems to only be for the maps tab so I suspect this is an issue with my Chrome install. I'll try and investigate in the next hour while I'm still home until I leave. I'll follow up on Tuesday (California time).

Fri, 10 Mar 2017 17:39:47 GMT - J. Max Mena: attachment set

attachment set to 1460paintrects.png

XPRA_OPENGL_PAINT_BOX=1

Sun, 12 Mar 2017 12:15:31 GMT - Antoine Martin: priority changed; milestone set

priority changed from major to critical
milestone set to 2.0

Is this still an issue? (works fine here on Fedora 25)

Fri, 17 Mar 2017 20:46:36 GMT - J. Max Mena:

Yes I still don't see any paints whatsoever in Chrome, but only on the Maps page. I'm not sure if this is an issue anyone else is seeing or maybe it's just my work machine.

Sat, 18 Mar 2017 05:48:02 GMT - Antoine Martin:

Try to:

downgrade to v0.17.x (or 0.14.x even) - does it still occur?
compare with another system, ie: try it in a VM
try with vglrun in case maps using opengl rendering
try with opengl rendering on / off
try taking a screenshot "xpra screenshot"
try with "sync-xvfb" enabled and take a screenshot of the backing display (using "scrot" or other screencapture commands)

etc..

Tue, 21 Mar 2017 17:24:15 GMT - J. Max Mena: attachment set

attachment set to Chrome Paint Bug Local.png

Tue, 21 Mar 2017 17:25:33 GMT - J. Max Mena:

Okay I'm convinced now that this is an issue with Chrome on Fedora - I've reproduced it on Pandora on my local Chrome running directly on my Workstation.

I'll try what you mentioned as well today.

Tue, 21 Mar 2017 23:28:25 GMT - J. Max Mena:

Okay, I tried downgrading to Xpra 1.X and I still got the same issue, so it's not 2.0

Running with VGLRun doesn't change anything

Opengl rendering on/off doesn't change anything

I'll try doing a screenshot Thursday as I've run out of time today.

Wed, 21 Jun 2017 18:25:14 GMT - J. Max Mena: status changed; resolution set

status changed from new to closed
resolution set to fixed

Okay I've tried this on three other machines now (all Fedora), and the issue is limited to just one single machine. I am entirely convinced at this point that the issue isn't with Xpra, but rather my Chrome install on that machine.

I'm gonna go ahead and close this ticket now.

Sat, 23 Jan 2021 05:24:54 GMT - migration script:

this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/1460