[GRASS-dev] Re: [GRASS-user] Compilation failure r.in.xyz on FreeBSD

[Brad Douglas]

On Sat, 2006-11-04 at 21:05 -0700, Tom Russo wrote:
> On Sun, Nov 05, 2006 at 04:23:50PM +1300, we recorded a bogon-computron collision of the <hamish_nospam at yahoo.com> flavor, containing:
> > Tom Russo wrote:
> > > I just tried to build today's CVS checkout, and got a compilation
> > > failure in r.in.xyz on my FreeBSD system.  The problem is the last
> > > commit changed the type of the "filesize" variable to off_t, but on
> > > FreeBSD it is necessary to have an explicit include of <sys/types.h>
> > > to get the off_t typedef.
> > > 
> > > Inserting a 
> > >  #include <sys/types.h>
> > > immediately before the include of <stdio.h> fixed the problem.
> > 
> > 
> > ok, I guess that wasn't as safe as I thought it was.
> > 
> > should the "#include <sys/types.h>" be included in <grass/config.h> or
> > placed by hand in each module that uses "off_t"?
> 
> I dunno, I guess it'd be safe to include sys/types.h in config.h, but 
> perhaps it would be better to probe for off_t in "configure", so
> that the include is really only pulled in on systems that require it?

No.  config.h is only used to define whether specific functions/features
exist.  Defining include files there is not appropriate.  It enables us
to conditionally define sections in source (eg.):

#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif

config.h is a result of the 'configure' script.  There is no need to
"probe" for off_t.  It is part of the POSIX specification and I believe
'configure' has sufficient support to determine the size of off_t.

All POSIX systems that want largefile support must use off_t and use the
appropriate compile macros for that arch.  'int' and 'long' are not of
sufficient length.


-- 
Brad Douglas <rez touchofmadness com>                    KB8UYR/6
Address: 37.493,-121.924 / WGS84    National Map Corps #TNMC-3785