Hi Paul, > > - if (nfd < 0) > > + if (nfd > TYPE_MAXIMUM (nfds_t) / 2) > > This doesn't look right. What does TYPE_MAXIMUM (nfds_t) / 2 have to do with > anything?
TYPE_MAXIMUM (nfds_t) / 2 is good for a quick check against an out-of-bounds nfd. POSIX [1] says that nfd is out-of-bounds if it is > getdtablesize(). > Plus, five lines later there's a comment that starts "Don't check > directly for NFD too large" and this seems to directly contradict the code > now. This comment means that the code is a bit sloppy for nfd > getdtablesize(). IMO, this is OK for nfd < TYPE_MAXIMUM (nfds_t) / 2 because such values rarely occur by accident. Whereas it's easy to pass -1 by mistake, which gets converted to (nfds_t)-1. > Plus, there's another "nfd < 0" in the WINDOWS_NATIVE case. Oops. Thanks. I've corrected it now. > How about the attached patch? I disagree. It no longer catches the case nfd = (nfds_t)-1 quickly. Bruno [1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/poll.html
