On Wed, 28 Jan 2009, Xavi wrote:
Hi Xavi,
> The return value does not include the terminating null byte in the count.
> ...
> if( c != 0 ) ++size;
> }
> while( c != 0 );
> ...
I cannot find such code in hbprintf.c.
hb_printf_c() for sure sets trailing \0 byte in all cases even if buffer
is too small.
> I made a few tests with BCC and Spd add #define snprintf hb_snprintf_c
> and the basics are working but with differences in accuracy.
> ? Sql_sprintf( "Phi = %.15f", (1 + 5**0.5) / 2 ) )
> Phi = 1.618033988749895 snprintf() it's Ok
> Phi = 1.618033981323242 hb_snprintf_c()
Welcome to the fantastic world of floating point arithmetic by Borland.
Try this code with this "incredible" compiler:
#include <stdio.h>
#include <math.h>
#include <float.h>
int main( void )
{
printf( "[%.15f]\n", ((1 + pow(5.0,0.5)) / 2 ) );
_fpclass( 0 );
printf( "[%.15f]\n", ((1 + pow(5.0,0.5)) / 2 ) );
return 0;
}
It's good to know that calling _fpclass() reduce floating point
precision to 'float' type even if you use 'double' or 'long double'.
It will create problems for us also in HVM calculations if 3-r party
code execute this function, f.e. try this:
proc main()
? str( (1 + 5**0.5) / 2, 20, 15 )
_fpclass()
? str( (1 + 5**0.5) / 2, 20, 15 )
return
#pragma begindump
#include <math.h>
#include <float.h>
HB_FUNC( _FPCLASS )
{
_fpclass( 0 );
}
#pragma enddump
I'll disable _fplcass() and checking for +-inf in BCC builds.
It will fix the results and I think we should document it somewhere
to avoid possible problems which can be created by 3-r party code
which will try to use it.
_fplcass() is forbidden in BCC builds !!!! ;-)
==============================================
best regards,
Przemek
_______________________________________________
Harbour mailing list
Harbour@harbour-project.org
http://lists.harbour-project.org/mailman/listinfo/harbour
