I'm inclined to leave this on gcc/glibc's doorstep
with the fear that in X months, after gcc/glibc is fixed to work correctly,
the workaround may deteriorate to some other incorrect behavior

otherwise ksh/ast could get into the buisiness of maintaining its own
-lm clone to workaround implementation deficiencies on all architectures

-- Glenn Fowler -- AT&T Research, Florham Park NJ --

On Wed, 17 Dec 2008 13:32:33 +0100 Dr. Werner Fink wrote:
> On Tue, Dec 16, 2008 at 01:57:35PM +0100, Dr. Werner Fink wrote:
> > On Tue, Dec 16, 2008 at 01:36:56PM +0100, Roland Mainz wrote:
> > > "Dr. Werner Fink" wrote:
> > > > On Tue, Dec 16, 2008 at 11:22:49AM +0100, I. Szczesniak wrote:
> > > > > On 12/16/08, Dr. Werner Fink <[email protected]> wrote:
> > > [snip]
> > > > > >   test io.sh failed at 2008-12-15+10:37:40 with exit code 1 [ 70 
> > > > > > tests 1 signal ]
> > > > > >
> > > > > >  On IA64:
> > > > > >
> > > > > >   test arith.sh begins at 2008-12-15+10:42:55
> > > > > >          arith.sh[452]: printf '%g\n' $((-Nan)) failed -- expected 
> > > > > > -nan, got nan
> > > > > >          arith.sh[454]: printf '%g\n' -Nan failed -- expected -nan, 
> > > > > > got nan
> > > > > >          arith.sh[456]: print -- $((-Nan)) failed -- expected -nan, 
> > > > > > got nan
> > > > > >   test arith.sh failed at 2008-12-15+10:42:55 with exit code 3 [ 
> > > > > > 127 tests 3 errors ]
> > > > >
> > > > > Did you disable IEEE754 math support? Negative nan is defined in
> > > > > IEEE754-1985 and mandatory required in IEEE 754-2008.
> > > > 
> > > > I use the same spec file for all architectures.  I'm not aware on
> > > > any switch which would enable or disable any math support.  it is
> > > > simply the defaults of the gcc and the defaults of the build for
> > > > the distribution which includes specific architecture switches
> > > > but nothing influence math support.
> > > 
> > > Please use gcc in C99 mode, e.g.
> > > $ CC="gcc -std=gnu99 -fPIC" ./bin/package make CC="gcc -std=gnu99 -fPIC"
> > > #. Note this must be set in the CC variable and _not_ in CCFLAGS.
> > > 
> > > This is the same what we do for the official ksh93 in Solaris (e.g.
> > > force the compiler to operate in C99-conformant mode), otherwise ksh93
> > > misses some C99 math functions... ;-(
> > 
> > Does not make any difference ;(

> The attached patch makes a difference ... it seems to be a problem
> of the glibc in collaboration with the gcc.  That is that at least
> the inlined function __signbitl() from /usr/include/bits/mathinline.h
> and used in the macro signbit() of /usr/include/math.h does not
> work on ia64 ... whereas the gcc builtin for signbitl does.

_______________________________________________
ast-developers mailing list
[email protected]
https://mailman.research.att.com/mailman/listinfo/ast-developers

Reply via email to