Re: sprintf and SFIO

On Fri, Apr 27, 2001 at 11:34:38AM -0400, Chris Nandor wrote:

> In op/sprintf.t, there are four tests that have problems for me (with
> MacPerl 5.6.1a1).
>
> >%.0f<      >0.6<         >1<              >Known to fail with sfio and
> >(irix|nonstop-ux|powerux)<
> >%.0f<      >-0.6<        >-1<             >Known to fail with sfio and
> >(irix|nonstop-ux|powerux)<
> >%012.9g<   >12345.6789<  >0012345.6789<
> >%013g<      >1234567.89< >001.23457e+06<
>
> The first two, as stated above, are known to fail with SFIO (1999).  The
> other two have no such comments.
>
> The results are as follows:
>
> 0
> -0
>   12345.6789
>   1.23457e+06
>
> So questions:
>
> * Does anyone else see the problems with %g and SFIO?  I could find no
> mention of them as problems elsewhere.
> * Do these problems still show up in SFIO 2000?

on linux (ARM) with this:

#include <sfio.h>

int main () {
  char buffer[256];
  sfsprintf (buffer, 256, ">%.0f<\n", 0.6);
  write (1, buffer, strlen (buffer));
  sfsprintf (buffer, 256, ">%.0f<\n", -0.6);
  write (1, buffer, strlen (buffer));
  sfsprintf (buffer, 256, ">%012.9g<\n", 12345.6789);
  write (1, buffer, strlen (buffer));
  sfsprintf (buffer, 256, ">%013g<\n", 1234567.89);
  write (1, buffer, strlen (buffer));
  return 0;
}


I see:

sfio97:             Debian sfio1999:    sfio2000:
>1<                 >0<                 >0<
>-1<                >-0<                >-0<
>  12345.6789<      >0012345.6789<      >0012345.6789<
>  1.23457e+06<     >001.23457e+06<     >001.23457e+06<

[sorry, no sfio98 at home, and my sfio1999 is the Debian distribution, so may
well have patches applied by the Debian packager. It says
#define SFIO_VERSION    19990805L
but it may well differ from a vanilla 19990805]

with -Dsfsprintf=snprintf glibc's stdio gives me

>1<
>-1<
>0012345.6789<
>001.23457e+06<


I hope that this is helpful. Sorry that I don't have a 98 or vanilla 1999
here. [I got sfio95 to build on Debian x86, but it fails a lot of tests]

Do we have an ANSI floating point printf lawyer to say whether perl's tests
are 100% correct? Or if sfio and irix... are acceptable in whatever they do?

Nicholas Clark

• sprintf and SFIO by Chris Nandor
• Re: sprintf and SFIO by Nicholas Clark
• Re: sprintf and SFIO by Nicholas Clark