- Increment reference count of bufferevents before initiating overlapped
operations to prevent the destructor from being called while operations
are pending. The only portable way of canceling overlapped ops is to
close the socket.
- Translate error codes to WSA* codes.
- Better handling of errors.
- Add an interface to add and del "virtual" events. Because IOCP
bufferevents don't register any events with the base, the event loop
has no way of knowing they exist. This causes the loop to terminate
prematurely. event_base_{add,del}_virtual increment/decrement base's
event count so the loop runs while there are any enabled IOCP
bufferevents.
Previously we were using InitializeCriticalSection, which creates a
lock that blocks immediately on contention and waits to be
rescheduled. This is inefficient; it's better to wait for a little
while before telling the US to reschedule us, in case the lock becomes
available again really soon (since most locks mostly do).
Good pthreads implementations do this automatically. On Windows,
though, we need to call this magic function, and we need to pick the
spin count ourselves.
Chris Davis points out that GetQueuedCompletionStatus
sometimes returns false not to report "No events for
you!" but instead to report "An overlapped operation
failed." Add a way to tell an event_overlapped that
its operation failed.
svn:r1490
Either I need to make the callbacks get deferred in a base with no events (doable), or I need to make it okay to call launch_read from inside the callback for read (tricky).
svn:r1277
