Q119504: PRB: Result Differs Between 16-bit and 32-bit _gcvt()

Article: Q119504
Product(s): Microsoft C Compiler
Version(s): 
Operating System(s): 
Keyword(s): kbCRT kbVC
Last Modified: 19-FEB-2002

-------------------------------------------------------------------------------
The information in this article applies to:

- The C Run-Time (CRT), used with:
   - Microsoft Visual C++, 32-bit Enterprise Edition 
   - Microsoft Visual C++, 32-bit Professional Edition 
   - Microsoft Visual C++, 32-bit Learning Edition 
-------------------------------------------------------------------------------

SYMPTOMS
========

Using the following test case, 32-bit _gcvt() returns a different value than
16-bit _gcvt():

     double r1 = 1.797693134862315E+308;
     char buf[40];

     _gcvt(r1, 15, buf);

The 16-bit _gcvt() rounds the value to 1.79769313486232E+308, but the 32-bit
_gcvt() truncates the value to 1.79769313486231E+308.

CAUSE
=====

The low-level floating-point-to-string conversion routines were completely
rewritten for Win32. The 32-bit output routines use an internal 12-byte
floating-point format for the intermediate values and a new conversion
algorithm. The specific value used in the test case mentioned in the "SYMPTOMS"
section is at the limit of type "double," and the new conversion routine
exhibits this problem when converting the value to a 12-byte value. The
conversion routine yields a mantissa string of 79769313486231490 instead of
797693134862315, which causes the rounding problems in this particular case.

Additional query words:

======================================================================
Keywords          : kbCRT kbVC 
Technology        : kbVCsearch kbAudDeveloper kbCRT
Version           : :
Issue type        : kbprb

=============================================================================