Bug 31456

Review of attachment 40112 [details] [review]:

Looks sane enough.  Should we even prefer <poll.h> over <sys/poll.h> if both are there?

(In reply to comment #1)
> Review of attachment 40112 [details] [review]:
> 
> Looks sane enough.  Should we even prefer <poll.h> over <sys/poll.h> if both
> are there?

hmm.. on my gentoo linux box, poll.h includes sys/poll.h, so that'd be ok. some other boxes i looked at:

Gentoo Linux (current):         poll.h includes sys/poll.h
HPPA HP-UX B.11.11:             likewise
HPPA + IA64 HP-UX B.11.31:      likewise
AIX 5.3:                        likewise
SunOS 5.10 i386 + sparc:        likewise
Interix (SFU 3.5, SUA 5.2-6.1): no sys/poll.h, poll.h is the one and only.

also, according to opengroup at [1], poll.h is known, but sys/poll.h is not in the specification. i wonder why sys/poll.h is so popular (seeing the same issue over and over again ;))

[1] http://www.opengroup.org/onlinepubs/009695399/basedefs/poll.h.html

Whomever takes this : You might want to have a look at #9537, which has a patch for autodetecting sys/select.h.

(In reply to comment #3)
> Whomever takes this : You might want to have a look at #9537, which has a patch
> for autodetecting sys/select.h.

is such a check really applicable? i intentionally did _not_ do such a check, as i don't want to force each and every package using xproto to do the check too (to have HAVE_POLL_H/HAVE_SYS_POLL_H defined). i rather chose to AC_SUBST the correct header, as is this will not force any changes on other packages.

Why were we using sys/poll.h in the first place?  POSIX requires it to be in 
<poll.h>

I think the correct fix is just:

#else /* USE_POLL */
-#include <sys/poll.h>
+#include <poll.h>
 #endif /* USE_POLL */

Am I missing something?

Mass closure: This bug has been untouched for more than six years, and is not
obviously still valid. Please reopen this bug or file a new report if you continue to experience issues with current releases.