#12 [PATCH] fixed select() call in CEPoll::wait() for sys sockets when OS is not Linux

[André Pacheco Neves]

I've been making some developments under OSX and iOS and was trying to use udt's epoll with both UDT and SYS sockets. After noticing that the epoll_wait wasn't firing events for SYS sockets I delved into the code and found out that when the OS was not Linux the code invokes select() by default for SYS sockets, which was my case.

Then I found out that the problem was that select() was being invoked with the first argument equal to 0, which should be (max_fd + 1), where max_fd is the maximum file descriptor of all descriptors being monitored. You can confirm this in the select man page.

So, I made a small fix to determine the max_fd and invoke select() with (max_fd + 1) as the first argument.

In the future I may try to add kqueue support for SYS sockets when on OSX / iOS.

Thanks for the awesome project!

[André Pacheco Neves]

Hi,
Please ignore the previous file and use this one instead.
SF doesn't let me edit my patch request even though it's open.

Thanks!

[Mikhail T.]

Shouldn't any errors encountered by select() be propagated to the caller too?

[André Pacheco Neves]

Hi Mikhail,

Since error handling isn't being done in neither native epoll() or select(), I didn't include it.

My guess is that the reason for this was to not interrupt normal UDT CEPoll::wait() behavior unless the error was UDT related.

You could add error handling, but to be consistent with the existing code you would have to do something like

(...) // epoll() or select() call
else // error
{
CGuard::leaveCS(m_EPollLock);
throw CUDTException(X, Y); // new exceptions
}
(...)

which would obviously interrupt the CEPoll::wait() call.

One could however create a specific error handling which would print something to stderr based on errno value. That would be better than nothing I guess.

Cheers