[RFC] PR6015, snprintf, glibc2.3 and all that...

[Paolo Carlini]

Ulrich Drepper wrote:

>>Is there a general "philosophical" point about this which you can 
>>possibly explain to me?
>>
>
>I don't know what you mean.  We will have the uselocale() stuff in glibc
>2.3.  Until then the locale model is basically broken by design wrt
>threads and since the locale data must be used by all the variouses
>interfaces of the libc it also means that those interfaces are also
>broken wrt threads.  Everybody who has to do with locales knows this. 
>It's just that nobody took the initiative to fix this.  The uselocale()
>stuff has a good chance of getting wider acceptance.
>

Yes. And of course v3 will use it on platforms which have glibc 
installed. But consider all the others... In order to have MT-safe 
output operations we don't really need a full printf locale sensitive. 
We just need a /basic/ printing facility which only knows about 
"C"-locale. I can only agree with you that locales are broken vs 
MT-threading, but really we don't need them /at all/ for Stage1 of 
number formatting according to the C++ standard. As I wrote elsewhere, 
the scheme proposed by Benjamin:

 template<typename _Tv>
   int
   __convert_from_v(char* __out, const int __size, const char* __fmt,
             _Tv __v, const __c_locale& __cloc, int __prec = -1)
   {
     int __ret;
     __c_locale __old = __uselocale(__cloc);
     if (__prec >= 0)
       __ret = snprintf(__out, __size, __fmt, __prec, __v);
     else
       __ret = snprintf(__out, __size, __fmt, __v);
     __uselocale(__old);

     return __ret;
   }

is cute, nice, MT-safe, and it will work (with glibc2.3) but I'm still 
getting a feeling of "overkilling"... Don't you?

Ciao, P.