Re: vasnprintf fix

[Eric Blake]

Bruno Haible <bruno <at> clisp.org> writes:

> 
> The floating-point output code crashed due to an abort() for values with
> large exponents (> 1e34 for 'double'). This fixes it.
> 
> 2007-11-04  Bruno Haible  <bruno <at> clisp.org>
> 
>       * lib/vasnprintf.c (scale10_round_decimal_decoded): Fix shift loop.

Something's still fishy.  Now I'm getting aborts for test-vasprintf-posix.c on 
cygwin, with the following sequence of instructions in vasnprintf.c:

Breakpoint 1, test_function (my_asprintf=0x413052 <my_asprintf>)
    at ../../tests/test-vasprintf-posix.c:173
173         int retval =
      my_asprintf (&result, "%a %d", 3.1416015625, 33, 44, 55);


2690                    if (type == TYPE_LONGDOUBLE)
(gdb) 
3055                        double arg = a.arg[dp->arg_index].a.a_double;
(gdb) 
3057                        if (isnan (arg))
(gdb) 
3070                            int sign = 0;
(gdb) 
3072                            if (signbit (arg)) /* arg < 0.0 or negative 
zero */
(gdb) 
3078                            if (sign < 0)
(gdb) 
3080                            else if (flags & FLAG_SHOWSIGN)
(gdb) 
3082                            else if (flags & FLAG_SPACE)
(gdb) 
3085                            if (arg > 0.0 && arg + arg == arg)
(gdb) 
3099                                pad_ptr = p;
(gdb) 
3101                                if (dp->conversion == 'f' || dp->conversion 
== 'F')
(gdb) 
3136                                else if (dp->conversion == 'e' || dp-
>conversion == 'E')
(gdb) 
3246                                else if (dp->conversion == 'g' || dp-
>conversion == 'G')
(gdb) 
3421                                  abort ();
(gdb) 

-- 
Eric Blake