Comment 56 for bug 194029

I found a couple of problems in the previous patch.
The test for isEmpty had the sense backwards.
I also reduced the number of writes to the wake-up pipe
and increased the amount read from the pipe per call.
The select-based code (which ubuntu doesn't use) also had
a couple of defects that I found and fixed after forcing the
non-poll based code to be used.

I don't see how those problems could lead to the assertion in
XGetXCBBuffer. There may be a latent problem with reentrant
calls to Xlib that still hides on my single cpu system.
A dual core system may help to bring out that bad behavior.
But I haven't put hardy on such a system yet.