Bug 66 - wcstold functions are incorrect for 128bit long double
Summary: wcstold functions are incorrect for 128bit long double
Status: RESOLVED FIXED
Alias: None
Product: glibc
Classification: Unclassified
Component: libc (show other bugs)
Version: 2.3.3
: P2 normal
Target Milestone: ---
Assignee: GOTO Masanori
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2004-03-09 22:11 UTC by H.J. Lu
Modified: 2019-04-10 12:23 UTC (History)
0 users

See Also:
Host:
Target:
Build:
Last reconfirmed:
fweimer: security-

Attachments
A patch to fix 128bit wcstold/wcstold_l (2.21 KB, patch)
2004-03-11 16:27 UTC, H.J. Lu 		Details | Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description H.J. Lu 2004-03-09 22:11:37 UTC 
[hjl@gnu libc]$ find -name wcstold.c
./wcsmbs/wcstold.c
[hjl@gnu libc]$ find -name strtold.c
./sysdeps/generic/strtold.c
./sysdeps/ieee754/ldbl-128/strtold.c
./sysdeps/ieee754/ldbl-96/strtold.c
./sysdeps/m68k/strtold.c

We are using 96 bit double for 128bit long double in wcstold functions.
wcstold should be treated similar as strtold.
Comment 1 H.J. Lu 2004-03-11 16:27:18 UTC 
Created attachment 30 [details]
A patch to fix 128bit wcstold/wcstold_l

There is only one wcstold.c under wcsmbs, which is used on all
platforms. It has

# define SET_MANTISSA(flt, mant) \
  do { union ieee854_long_double u;					      \

       u.d = (flt);							      \

       if ((mant & 0x7fffffffffffffffULL) == 0) 			      \

	 mant = 0x4000000000000000ULL;					      \

       u.ieee.mantissa0 = (((mant) >> 32) & 0x7fffffff) | 0x80000000;	      \

       u.ieee.mantissa1 = (mant) & 0xffffffff;				      \

       (flt) = u.d;							      \

  } while (0)

But it is for 96bit IEEE long double and won't work for 128bit IEEE
long double. This patch just includes <strtold.c> or <strtold_l.c>,
which support different long doubles.
Comment 2 Ulrich Drepper 2004-03-15 03:50:21 UTC 
Should be handled correctly in CVS.  Not with the attached patch, though, but
with a completely different solution.