cvs commit: src/share/man/man3 Makefile src/lib/libc/alpha/gen Makefile.inc fpsetsticky.c src/lib/libc/amd64/gen Makefile.inc fpsetsticky.c src/lib/libc/powerpc/gen Makefile.inc fpsetsticky.c src/lib/libc/softfloat Makefile.inc src/lib/libc/sparc64/gen ...

[Steve Kargl]

On Wed, Mar 16, 2005 at 01:46:52PM +1100, Bruce Evans wrote:
> On Tue, 15 Mar 2005, David Schultz wrote:
> 
> >On Tue, Mar 15, 2005, David O'Brien wrote:
> >>On Tue, Mar 15, 2005 at 10:11:27AM -0800, Steve Kargl wrote:
> >>>On Tue, Mar 15, 2005 at 03:53:39PM +0000, David Schultz wrote:
> >>>>  Log:
> >>>>  Remove fpsetsticky().
> >>>
> >>>Library version bump?  It took several years to live through
> >>>the libm.so.2 fiasco, and I would hate to relive it.  I haven't
> >>>verified it yet, but you may have just broken NAGWare's Fortran
> >>>95 compiler without the version bump.
> >>
> >>We've already bumped libm.so for 6-CURRENT [to .3 from .2].
> >>This change certainly cannot be MFC'ed though.
> >>
> >>How would bumping libm.so not break NAGWare Fortran?
> >
> >It turns out that fpsetsticky() was misplaced in libc, and libc
> >has already been bumped for 6.X.  But as you point out, the bump
> >probably wouldn't fix NAGWare Fortran if this change broke it.
> >However, my change doesn't seem to have broken the demo version
> >available on the NAG website.  But if I did break anything, please
> >let me know.
> 
> It isn't even in libc for i386's (since it is inline and not backed
> by a function).  So removing it can't affect binary compatibility on
> i386's.
> 

Odd, I haven't seen David's or David's replies.

If you re-read what I wrote, you see "I haven't verified it yet,
but you may have just broken NAGWare ...".  Note the speculation
in the quoted passage.  In NAGWare's f95.h, there is a function
with a name similar to _NAG_IEEE_fpsetsticky(), which suggests
that NAG uses fpsetsticky().  I'll also admit I completely
miss that das changed libc not libm.  Das is the only person
who regularly commits to libm source, so I naturally assumed
that fpsetsticky()  was in libm. 

-- 
Steve