From: moshier AT mediaone DOT net ()
Subject: strtod bug in 202 beta
Newsgroups: comp.os.msdos.djgpp
X-Newsreader: TIN [version 1.2 PL2]
Lines: 50
Message-ID: <nMPZ1.49$2U3.521928@lwnws01.ne.mediaone.net>
Date: Thu, 29 Oct 1998 01:44:51 GMT
NNTP-Posting-Host: 24.128.118.69
X-Trace: lwnws01.ne.mediaone.net 909625491 24.128.118.69 (Wed, 28 Oct 1998 20:44:51 EDT)
NNTP-Posting-Date: Wed, 28 Oct 1998 20:44:51 EDT
Organization: Northeast Region--MediaOne
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com



Using the new beta djdev202.zip the following program prints, in error,

  scanf("1.1125369292536017e-308") = 0.0000000000000000e+00

This is a regression from previous releases of djpp libc.a.

-------------------------------------
char *s = "1.1125369292536017e-308";

int
main()
{
  double x;

  sscanf (s, "%lf", &x);
  printf("scanf(\"1.1125369292536017e-308\") = %.16e\n", x);
  exit(0);
}
-------------------------------------

The problem is that the result is compared with DBL_MIN, the
smallest normalized double.  Thus all denormal values get flushed to zero.
The patch below seems to fix the problem.
For additional scanf exercises, please get
  http://www.netlib.org/cephes/ieetst.shar


*** tmp\strtod.c        Sat Jul 25 18:51:38 1998
--- strtod.c    Wed Oct 28 20:24:30 1998
***************
*** 96,102 ****
      for (i = 1; i <= e; i++)
      {
        r *= 0.1L;
!       if (r < DBL_MIN)        /* detect underflow */
        {
        errno = ERANGE;
        r = 0.0;
--- 96,104 ----
      for (i = 1; i <= e; i++)
      {
        r *= 0.1L;
!       /* Detect underflow below 2^-1075, which is half
!        the smallest representable double. */
!       if (r < 2.47032822920623272088e-324L)
        {
        errno = ERANGE;
        r = 0.0;