Subject: Re: CVS commit: pkgsrc
To: Jason Beegan <jtb@netbsd.org>
From: Johnny C. Lam <lamj@stat.cmu.edu>
List: pkgsrc-changes
Date: 03/31/2001 11:56:06
The situation you are describing where CFLAGS needs to be set to
something specific lest the compilation breaks sounds like the unusual
case which should be documented explictly in those packages that need
it.  If you are aware of packages that need this, please be proactive
and add the necessary comments.

	Thanks,

     -- Johnny C. Lam <lamj@stat.cmu.edu>
        Department of Statistics, Carnegie Mellon University
        http://www.stat.cmu.edu/~lamj/


Jason Beegan <jtb@netbsd.org> writes:
> 
> Sometimes one doesn't want to respect CFLAGS.  For example, the
> bytesex.c file in the devel/hdf package only tries to determine the
> endianness of the host machine and isn't part of the final compiled
> package.  Using optimization might cause the compiler to be too clever
> and give us a wrong result.
> 
> It should be alright in this case, but there are others where it would
> cause the package to malfunction.  This is especially true when trying
> to determine machine parameters, eg. math/slatec contains three files
> which compile with no optimization and these end up in the final
> library.
> 
> Thanks a lot for tidying up those Makefiles, but watch out
> for this kind of thing.
> 
> Jason.