svn commit: r215572 - stable/8/sys/sys

[Jilles Tjoelker]

On Sun, Nov 21, 2010 at 12:42:24PM +0000, Gavin Atkinson wrote:
> On Sat, 20 Nov 2010, Jilles Tjoelker wrote:
> > Author: jilles
> > Date: Sat Nov 20 16:46:16 2010
> > New Revision: 215572
> > URL: http://svn.freebsd.org/changeset/base/215572

> > Log:
> >   MFC r215183: Make POLL_ERR and POLL_HUP different.

> >   The kernel currently does not generate any of the POLL_* constants, but
> >   some applications use them and break if they are not all distinct.

> Is this a good idea to merge to a stable branch?  Presumably some of those 
> same applications will now break due to the constants changing?

The typical usage appears to be
	switch (si->si_code) {
		case POLL_IN: ...; break;
		...
		case POLL_ERR: ...; break;
		case POLL_HUP: ...; break;
	}

This doesn't compile if some of the constants are equal.

Boost has worked around FreeBSD's weirdness by placing a
  #if defined(POLL_ERR) && defined(POLL_HUP) && (POLL_ERR - POLL_HUP)
around the case POLL_HUP. This #if does not hurt systems that have the
POLL_* values all distinct.

Given that the kernel does not generate these anyway, I considered this
safe enough. True SIGIO, which is similar to SIGPOLL, always puts
si_code=SI_KERNEL.

-- 
Jilles Tjoelker