Michael Kerrisk wrote: > > Von: "[EMAIL PROTECTED]" <[EMAIL PROTECTED]> > > /usr/share/man/man2/open.2.gz has (under O_NONBLOCK): > > "This mode need not have any effect on files other than FIFOs." > > > > I'm almost sure that statement is false. > > O_NONBLOCK prevents blocking on any filehandle (files, sockets, fifos, > > devices, ...). > > Justin is on the right track in mentioning the "standard" > behavior of O_NONBLOCK. And you haven't got things quite > right -- for example, O_NONBLOCK can't be used for open() with > sockets, because we can't open() a socket, and O_NONBLOCK has > no meaning for regular files (it is just silently ignored).
Yes, the de jure standard as susv3 may say is that, but the de facto standard as _any_ modern UNIX does is the same for every type of file. > But I agree that the man page is a bit misleading. One might > get the impression that, for example, O_NONBLOCK can't be used > for sockets, although we can enable O_NONBLOCK on a socket using > fcntl(). And O_NONBLOCK does have meaning for some devices, > and sometimes has an effect for regular files. O_NONBLOCK _always_ has an effect for regular files, it prevents blocking on read or write operations (as an example think in NFS or slow disks) > I have made the following changes: > > * Removed that sentence from open.2 ;-). > > * Added a sentence noting that the effect of O_NONBLOCK > for file locks and leases is described in fcntl.2. > > * Somewhat expanded the discussion of mandatory locking in > fcntl.2. > > The changes will appear in upstream 2.29. You may always want to attach the patch or an URL for review. Thanks for these changes. I already knew they were misleading to many people. Prueba el Nuevo Correo Terra; Seguro, Rápido, Fiable.
