To: djgpp AT sun DOT soe DOT clarkson DOT edu
From: "Josef Spangler"  <JS AT rphnw3 DOT ngate DOT uni-regensburg DOT de>
Date:         22 Jul 93 14:55:48 MET
Subject:      printf format bug fixed; where is debug32v1.10a

Hello DJGPP gcc freaks,

1. Question about go32 und debug32 version:
-------------------------------------------
In the distibution djdev110.zip on omnigate the above programs have
version 1.10. But these don't work with gr or grx graphics. In the pub
subdir there is csdpmit1.zip with go32.exe version 1.10a and with this
the graphics works again. But where can I find a new debug32?


2. Bug in printf %lg format found
---------------------------------

The default precision lengths for the %g and %lg formats are to big.
To clarify, test the following program:

#include <stdio.h>
#include <string.h>
#include <math.h>

double anint(double x)
{
  return ( floor(x+0.5) );
}

/* fight the dust; approx gives the order of magnitude */
double hoover (double dusty, double approx)
{
  double ani;

  ani = anint(dusty/(approx)) * approx;
  /* this gives dust */
  printf("%le\t%le\t%le\t%lg\n",dusty,approx, ani, ani );
  /* this is ok */
  printf("%le\t%le\t%.14le\t%.14lg\n",dusty,approx, ani, ani );
  return( anint(dusty/(approx)) * approx);
}

int main(void)
{
   int i;
   double y = 0.0;
   double dusty,approx;

   dusty = 0.0;
   approx = 1.000000000000001e-4;
   for (i = 0; i < 5; i++)
   {
     y = hoover(dusty,approx);
     dusty += 0.5;
   }
   return 0;
}

The problem is in file libsrc\c\io\doprnt.c line 42, 43, where the
default precision lengths are defined:
#define DEFPREC         7
#define DEFLPREC        16

So without explicit specification more digits as valid are printed.
Look at the following table:

Name        Bits   valid digits     DEFxPREC old  DEFxPREC new
float        32         7-8              7            6
double       64        15-16            16           14
long double  80        19-20            16           14

All other c compilers like gcc2.2.x on sparc, emx-0.8g, bcpp3.1,
watcomc386v9.01, msc6.0 ... use a default len of 6 for %g *and* %lg

Changing these values to 6/14 looks resonable and corrects the problem:
#define DEFPREC         6
#define DEFLPREC        14

Please could this fix applied in the next release?

with greetings
J.Spangler
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Josef Spangler                  Phone: (+49) 941/943-2059
Institute for Applied Physics   Fax:   (+49) 941/943-4223
University of Regensburg        Mail:  js AT rphnw3 DOT ngate DOT uni-regensburg DOT de
Universitaetsstr. 31            or:    c4011 AT rphs1 DOT physik DOT uni-regensburg DOT de
D-8400 Regensburg, Germany      X400:  spangler AT vax1 DOT rz DOT uni-regensburg DOT dbp DOT de