Discovered by re-arranging the dual monitor layout of a win7 VM and moving the primary monitor to the right. (the left hand side one must be at least 1080p - or whatever the initial display size is set to) When new windows are created without requesting a specific location (ie: xterm), they are mapped onto the primary monitor by MS Windows. The window then fails to respond to any click events until it is moved at least one pixel into the original display size. (ie: 1080p)
This is reminiscent of the problems fixed by 0003-fix-pointer-limits.patch which was added in r8060.
What is strange is that starting an application after the server's dummy screen has been resized works fine. The problem only occurs if the application is started on a display which is smaller than the eventual client display dimensions. As if the application (or the X11 server's internal data structures for this application) were using out of date monitor geometry data.
Lazy temporary workaround: r15309 starts the vfb at 8192x4096 (the highest resolution we pre-define). We can do this relatively safely with regards to DPI because every distribution we support has a patched dummy driver (including Debian / Ubuntu since #1215)
Hopefully this can be fixed by the pending patches for #56, if not before that. I will try to cook up a plain C example application to report as an upstream X11 server bug. (that's what it looks like)
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/1467