http://bugs.winehq.org/show_bug.cgi?id=9787
--- Comment #234 from Andrey Turkin andrey.turkin@gmail.com 2008-07-28 12:55:42 --- (In reply to comment #229)
Need other people to test it.
So, other people, give it a shot!
I found one problem so far: this code needs at least one async to be registered during connect. Would this be false, the connect would be noticed, pmask would be updated and new poll events would be suppressed. Easiest way to reproduce is: 1) create new socket, bind() it and listen() to it 2) connect to this socket 3) call AcceptEx On Windows AcceptEx accepts the connection and sends completion message; with your patches connection never gets accepted and completion is never sent. With my patches this one works (because of first accept call you removed) but I think there are some similar bug. Existing asynchronous socket implementation (pmask/hmask/custom socket object functions) make things pretty complicated...