Re: test-printf-frexp ?

[Bruno Haible]

Simon Josefsson wrote:
> Bruno Haible <address@hidden> writes:
> 
> > According to the autobuild logs [1], the module 'printf-frexp' fails the
> > tests on a Linux/x86 machine of yours. I cannot reproduce this, on a
> > glibc-2.3.6 machine and on a glibc-2.5 machine.
> > ...
> > [1] http://autobuild.josefsson.org/gnulib/log-200703101113121276000.txt
> 
> I can't reproduce this with only -g, -O2 is required to
> trigger this.  ...
> $ gcc --version
> gcc (GCC) 4.1.2 20061115 (prerelease) (Debian 4.1.1-21)

Thanks. With gcc-4.1.2 I could reproduce it as well. I'm applying this fix.


2007-03-12  Bruno Haible  <address@hidden>

        * tests/test-printf-frexp.c (main): Declare x as volatile.

*** tests/test-printf-frexp.c   25 Feb 2007 14:21:42 -0000      1.1
--- tests/test-printf-frexp.c   12 Mar 2007 22:38:41 -0000
***************
*** 40,46 ****
  main ()
  {
    int i;
!   double x;
  
    for (i = 1, x = 1.0; i <= DBL_MAX_EXP; i++, x *= 2.0)
      {
--- 40,51 ----
  main ()
  {
    int i;
!   /* The use of 'volatile' guarantees that excess precision bits are dropped
!      when dealing with denormalized numbers.  It is necessary on x86 systems
!      where double-floats are not IEEE compliant by default, to avoid that the
!      results become platform and compiler option dependent.  'volatile' is a
!      portable alternative to gcc's -ffloat-store option.  */
!   volatile double x;
  
    for (i = 1, x = 1.0; i <= DBL_MAX_EXP; i++, x *= 2.0)
      {