Type libraries are currently parsed in host endianness, so byte arrays
that are going to be interpreted as integers need to be written in the
appropriate byte order.

The stop event should already be signaled by the time we get to that
point since we wait until the child processes terminate in the server
process and the stop event is signaled in the context of one of the
child processes.

Don't call RpcMgmtWaitServerListening if the call to
WaitForSingleObject failed since it is likely that s_stop() hasn't
been called and therefore the call to RpcMgmtWaitServerListening won't
ever return.

rpcrt4: Save the previous BufferLength in test_ndr_buffer and use it when testing the expected buffer length after the NdrFreeBuffer call.

This fixes a test failure on Win2003 upwards when the allocated length
appears to be rounded up to a multiple of 4.

This fixes a number of test failures on XP SP3 and Win2003 upwards.

This avoids calling PreLoad needlessly on a thread change during offscreen
rendering and breaks up the endless recursion due to lastTID != newTID.

There is no reason to do that, now that the SetGLTextureDesc bug is
fixed. This avoids an infinite recursion because PreLoad calls
ActivateContext at some point.

Thanks to Mauricio Teixeira <mteixeira@webset.net> for the translation.