@Guido: agreed.

> there's no SELECT_ERR anymore [...] the error will be reported when the FD will be read/written

I don't think this is a good idea, particularly during this early stage.
This assumption might be valid for select() but not for poll() and epoll() where (POLLERR | POLLHUP | POLLNVAL) is an alias for "connection lost", and that's an event which both Twisted and Tornado take into account and treat differently than a pure "read" event.
The same assumption should also apply for kqueue().

> the exceptions set isn't passed to select() anymore: exception events (like OOB) are already reported in the read or write sets

That's true only if SO_OOBINLINE has been pre-emptively set against the socket.
It might be ok to ignore select()'s exceptional events anyway though.
This is probably another thing you might want to decide later.

> there was something missing/annoying in the API [...]  now, it only accepts only file descriptors, and also an optional opaque 'data' object

I like this change as it avoids maintaining a separate fd->handler map.

Will add line-by-line comments by using the review tool.