ldouble math

[Carmelo AMOROSO]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Bernhard Reutner-Fischer wrote:
> On Thu, Mar 05, 2009 at 10:33:16AM -0800, Ned Ludd wrote:
>> On Thu, 2009-03-05 at 18:50 +0100, Bernhard Reutner-Fischer wrote:
> 
>>> Does anybody remember the story behind __NO_LONG_DOUBLE_MATH?
>>> Let's take powerpc32 as an example. Can sombody please show be
>>> $ nm -D /lib/libm.so.* | grep finite
>>> from some recent glibc's libm on a powerpc32?
>>> To be specific, i want to know if that one contains finitel or not.
>>> If it does, please show me the disas of it.
> 
>> luna(ppc32) ~ # nm -D /lib/libm.so.* | grep finite
>> 00028010 T __finite
>> 00034594 T __finitef
>> 00028010 T __finitel
>> 00028010 W finite
>> 00034594 W finitef
>> 00028010 W finitel
>>
>>
>> luna(ppc32) ~ # objdump -D /lib/libm.so.6 > dump.x
>> luna(ppc32) ~ # grep __finitel dump.x 
>> # nothing.. Odd. It's a T 
> 
> 00028010 == 00028010, ok that helps. Thanks!
> 
>> Note my attempt to build/update to uclibc-0.9.30.1 in the
>> ppc32-uclibc-chroot results in the following error (which is what I
>> guess you are trying to solve).
>>
>>
>> libm/ldouble_wrappers.c:561: error: '__EI___finitel' aliased to
>> undefined symbol '__GI___finitel'
>> make: *** [libm/finitel.os] Error 1
>> make: *** Waiting for unfinished jobs....
> 
> heh, yes, that's why i'm asking ;) I'm taking care of this ASAP,
> you could turn off C99 math support to bypass it for now.
> 
> 
In ./libc/sysdeps/linux/common/bits/mathcalls.h
there is a comment

51 /* __MATHCALLX and __MATHCALLI include libm_hidden_def
52  * (for "double" versions only, xxxf and xxxl do not get this treatment).

it seems that we are missing some hidden_proto in function prototype.

cheers,
Carmelo

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org

iEYEARECAAYFAkmw69kACgkQoRq/3BrK1s9S/ACgjSVENm7oQhQ1omW0iwG/D4Zi
WIoAoJHMdi+yFYToIncgIsE2Jk/aacDz
=avVN
-----END PGP SIGNATURE-----