Re: Test results for m4-1.4.9b

[Gary V. Vaughan]

Hi Bruno,

On 3 Jun 2007, at 02:03, Bruno Haible wrote:
> Gary V. Vaughan wrote:
> > powerpc-apple-darwin8.9.0:      2 failed gnulib tests
>
> Cannot reproduce on powerpc-apple-darwin7.9.0. Which compiler and  
> optimization
> flags did you use?

Gcc from Apple's Xcode package as shipped on my Tiger installation DVD:

  $ gcc --version
  powerpc-apple-darwin8-gcc-4.0.1 (GCC) 4.0.1 (Apple Computer, Inc.  
build 5250)
  Copyright (C) 2005 Free Software Foundation, Inc.
  This is free software; see the source for copying conditions.   
There is NO
  warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR  
PURPOSE.

Standard configure optimisation flags from m4's configure:  '-- 
std=gnu99 -g -O2'

> > Inserting test-frexpl.c:163:
> >       fprintf (stderr, "%lg == 0.505L\n", mantissa);
>
> Still, this doesn't help much. Can you print 40 decimal places of  
> the number
> (or the exact binary contents, via %LA)?

mantissa = 0X2.051EB851EB851EB851EB851EB8P-2

> >   0.505 == 1.06581e-16 (my_ldexp (0.505, - -1135314207))
> >   test-printf-frexpl.c:115: assertion failed
>
> The numbers 1.06581e-16 and -1135314207 don't give any clue.
>
> It smells either like a 'long double' returning function was being  
> called
> without being properly declared, or like dangerous compiler  
> optimizations.

Not the latter in this case.  It looks as though the compiler  
optimised away
the correct values in the output above, so I recompiled with CFLAGS=- 
ggdb3
and ran through gdb.  Here's a (hopefully more useful) typescript  
from what
seems like an appropriate point in the code:

Breakpoint 1, main () at test-printf-frexpl.c:114
114           ASSERT (mantissa == my_ldexp (x, - exp));
(gdb) p exp
$6 = -969
(gdb) s
my_ldexp (x=1.0121050218045312617883299999606451e-292, d=969) at test- 
printf-frexpl.c:55
55        for (; d > 0; d--)
(gdb) n
56          x *= 2.0L;
(gdb) l
51
52      static long double
53      my_ldexp (long double x, int d)
54      {
55        for (; d > 0; d--)
56          x *= 2.0L;
57        for (; d < 0; d++)
58          x *= 0.5L;
59        return x;
60      }
(gdb) fin
Run till exit from #0  my_ldexp  
(x=1.0121050218045312617883299999606451e-292, d=969) at test-printf- 
frexpl.c:56
0x00003048 in main () at test-printf-frexpl.c:114
114           ASSERT (mantissa == my_ldexp (x, - exp));
Value returned is $7 = 0.50500000000000000444089209850062616
(gdb) p mantissa
$8 = 0.50499999999999989341858963598497212
(gdb) n
test-printf-frexpl.c:114: assertion failed

Program received signal SIGABRT, Aborted.
0x9004822c in kill ()

Rounding error?

Cheers,
        Gary
--
  ())_.              Email me: address@hidden
  ( '/           Read my blog: http://blog.azazil.net
  / )=         ...and my book: http://sources.redhat.com/autobook
`(_~)_ Join my AGLOCO Network: http://www.agloco.com/r/BBBS7912

PGP.sig
Description: This is a digitally signed message part