svn commit: r283197 - head/usr.bin/soelim

[Bruce Evans]

On Thu, 21 May 2015, Baptiste Daroussin wrote:

> Log:
>  add an include on sys/types.h because we do explicitly use size_t
>  remove unused stdbool.h

sys/types.h may be needed for some other typedef, but not for size_t.
size_t is already declared in at least 3 other of the included headers:
- <stdlib.h> and <string.h> (C90 standard)
- <unistd.h> (POSIX.1-2001 standard)

<sys/types.h> is a prerequisite for <unistd.h> in POSIX.1-1990, and
most section 2 man pages still document it as a prerequiste for their
syscalls.  They place it before the include of <unistd.h> but are too
fuzzy to document if it is a prerequisite for <unistd.h or their
syscalls.  But it have never actually been a prerequisite in any version
of FreeBSD.  In FreeBSD-1 and 4.4BSD-Lite2, <unistd.h> includes
<sys/types.h> and thus gets size_t and lots of pollution.  <unistd.h>
was cleaned up a bit in 2002, but the ifdefs for things like
_SIZE_T_DECLARED in this cleanup are still nonsense, because <unistd.h>
still includes <sys/types.h> before these ifdefs and thus gets size_t
and lots of pollution, much the same as in FreeBSD-1.

Some of the section 2 headers were bogusly changed to say that
sys/types.h is a prerequisite for their syscalls _after_ POSIX.1-2001
standardized it not being a prerequisite for <unistd.h> for the syscalls
declared there.  It is needed for portability to pre-POSIX.1-2001
systems that are not polluted like FreeBSD, if any.  IIRC, even
POSIX.1-1990 permits <unistd.h> (or any POSIX header) to declare any
typedef ending in _t, so <unistd.h> can declare size_t for itself and
might even work without its prerequisite.  Applications just cannot
depend on this.

In the previous change, the use of MAXPATHLEN regressed to use of
PATH_MAX.  MAXPATHLEN was honestly unportable.  It defined in all (?)
BSD systems.  PATH_MAX only defined if {PATH_MAX} is constant.  It
is better to be honestly unportable than to use PATH_MAX incorrectly.

MAXPATHLEN required the include of <sys/param.h>.  That gave <sys/types.h>
and lots of undocumented pollution.  Apparently, none of the pollution
was needed, not even <sys/types.h>.

Bruce