Re: printf issues

[Morten Welinder <mwelinder@gmail.com>]

> I _would_ like this code to be easily adaptable for use outside libc
> if somebody wants it

FYI, I have been doing just that for Gnumeric in a variant that always
rounds ties away from zero.  Two changes would help with making
the code fit seamlessly into other environments.

1. Make "i" in fmt_fp unsigned.  It's used in connection with
    unsigned values only.

2. Make "char *s" used to hold "NAN" etc. "const char *s".

Neither of these should make any difference in what the function
actually does.

I have run tens of millions random numbers through this function
looking for differences between it and glibc.  The extra 0s from "%g"
is the only problem observed.


It looks like the LDBL_EPSILON version could be used in

    roundl.c
    modfl.c
    ceill.c
    floorl.c

in the definition of TOINT instead of enumerating choices for
LDBL_MANT_DIG.  It's basically the same thing going on
there.

While I was looking for that, I noticed that this modfl fallback looks
problematic.  Even if long double and double are the same thing
under the hood, I don't think you can cast pointers like that and
assume it works.  It needs a temporary.

#if LDBL_MANT_DIG == 53 && LDBL_MAX_EXP == 1024
long double modfl(long double x, long double *iptr)
{
return modf(x, (double *)iptr);
}



Morten





On Fri, Apr 4, 2014 at 11:07 AM, Rich Felker <dalias@aerifal.cx> wrote:
> On Fri, Apr 04, 2014 at 10:35:00AM -0400, Morten Welinder wrote:
>> > before you can mock libc code you need to educate yourself
>>
>> I did and that's why I called the code "cute", not "wrong".  But if you read
>> the porting documentation
>>
>>     http://brightrain.aerifal.cx/~niklata/PORTING
>>     http://www.openwall.com/lists/musl/2012/07/08/1
>>
>> you will notice that nowhere does it warn that defining LDBL_MANT_DIG
>> as anything but a base-10 constant may cause printf-rounding to fail.
>
> Good point.
>
>> > Do you have any ideas for a clean way to avoid this
>> > assumption without having to compute the value at runtime?
>>
>> I don't know if ldexpl will get constant folded by the compiler, but if not,
>> I think (2.0L/LDBL_EPSILON) ought to work as a replacement.  It's not
>> as likely to get prices at the obfuscated C contents, though.
>
> Thanks, I think that's exactly the right solution. FWIW, I _would_
> like this code to be easily adaptable for use outside libc if somebody
> wants it, so eliminating implementation-internal assumptions like this
> is nice.
>
> Rich