[kaffe] Re: The problem of "Double.c and buggy strtod" is settled

[Ito Kazumitsu]

Hi Dalibor,

In message "Re: [kaffe] Re: The problem of "Double.c and buggy strtod" is settled"
    on 03/06/05, Dalibor Topic <robilad at yahoo.com> writes:

> the netbsd strtod seems to need some includes that I don't have around, and the
> netlib strtod wants me to specify way more about IEEE than I know ;) The one
> from glibc is entangles in the glibc mess of headers, so I took another look at
> the one from libiberty. I hope that the attached patch works, I've tried a
> slightly adapted DoubleTest.c and it worked for me:

I used it on Linux 2.0.38, and the result seemed satisfactory....

But it was not because the new strtod worked but because I had
chaned the system-wide /usr/lib/libc.so so that it used the
strtod of NetBSD origin.
 
I restored the original /usr/lib/libc.so.  Then I got the
bad result:

$ ./a.out 4.9406564584124654418e-324
4.9406564584124654418e-324 0               0.00000000000000000000e+00

Seeing the source,  I found that the libiberty-strtod.c used atof.
The atof of Linux 2.0.38 uses its own __strtod_internal.

As long as __strtod_internal is used,  you cannot expect a good
result.