second maximization causes paint issues

[Antoine Martin]

This bug is not reproducible with virtualbox or the 3 machines I have: 2 with nvidia cards (mswindows and Linux) and 1 Intel chipset (Linux only).

• run in qemu, ie:

  qemu-kvm -vnc unix:/tmp/vnc -m 4096 \
      -device qxl-vga \
      -drive file=./CentOS-8.1.1911-x86_64-dvd1.iso,media=cdrom \
      -drive file=./disk-0,cache=unsafe,format=raw,aio=threads,detect-zeroes=on \
      -net nic -net user
  

Steps:

• connect with vncviewer /tmp/vnc
• install centos (default install)
• install xpra, ie:: cd /etc/yum.repos.d/;wget https://xpra.org/repos/CentOS/xpra.repo;yum install xpra
• xpra start --start=xterm --attach=yes reproduces the bug
• maximize the xterm, unmaximize then maximize it again

[Antoine Martin]

xterm looks wrong!

[Antoine Martin]

The whole xterm window should be plain white.
The cursor is a bit off too, but this looks like a different bug: ticket:2618#comment:7.

[Antoine Martin]

Also reproducible with -display sdl and -display gtk instead of using a VNC connection.
Also affects python2 client builds.
Using -vga cirrus instead of -device qxl-vga makes no difference, but -vga vmware does not have this problem, that's both with and without opengl acceleration!
(so likely a graphics driver bug)

[Antoine Martin]

Latest versions of all branches do not have this problem thanks to the workaround in r27438 (and the backport in r27439).

[Antoine Martin]

So... I wanted to see how far back I needed to go to find a version without this painting problem (assuming that a rendering change had caused this issue), and that cost me hours for nothing: the bug comes and goes, making it impossible to bisect.

[migration script]

this ticket has been moved to: ​https://github.com/Xpra-org/xpra/issues/2879