Tarim <s...@mediaplaygrounds.co.uk> wrote: > Thanks Eric, that makes things a lot cleaner, more resilient and, at a > guess, more portable (surely any OS with select will have O_NONBLOCK?) O_NONBLOCK is common, yes, but I have zero knowledge or interest in proprietary OSes. For systems with O_NONBLOCK, the select calls in kbhit are not even necessary anymore. I didn't change kbhit to minimize the patch size.
Did you get a chance to test and confirm my patch fixes things for you? > I agree it's more a performance bug but, I think, if select returns a > spurious wakeup then the read should set EAGAIN (EWOULDBLOCK) - which > the Linux 3.13.0-65.105-generic kernel doesn't. (Not that this would > have helped the unpatched sox). read may only return EAGAIN when O_NONBLOCK is set; otherwise existing applications which depend on blocking read will break. Fwiw, blocking read can be a performance win in some cases: it can avoid extra syscalls, avoid thundering herd, and improve fairness when multiple threads/processes are reading a off the same pipe. ------------------------------------------------------------------------------ _______________________________________________ SoX-devel mailing list SoX-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/sox-devel
