Date: Mon, 23 Nov 1998 08:30:06 +0200 (IST) From: Eli Zaretskii X-Sender: eliz AT is To: Kbwms AT aol DOT com cc: ams AT ludd DOT luth DOT se, djgpp-workers AT delorie DOT com Subject: Re: More rand() In-Reply-To: <9754ad9c.365867e3@aol.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Reply-To: djgpp-workers AT delorie DOT com On Sun, 22 Nov 1998 Kbwms AT aol DOT com wrote: > Whose computer works in 48-bit arithmetic nowadays? And what > advantages accrue when using 48-bit generators? One advantage is compatibility with other platforms. *rand48 are used quite a bit, and their algorithm seems to be well defined. So I think it's a good idea to have an implementation in the library. > Spoken like a man who has no intention of ever using the output of > a random number generator. No serious investigator uses the output > of a random number generator that cannot be identified and verified. Not all DJGPP users are ``serious investigators'' whose concern is to produce random numbers with good properties. Apart of these, there are also people who have a program that works on another platform which they want to compile and run on DOS/Windows. If the program has a good reputation on the other platform, and was tested and debugged by many people, I can easily imagine somebody who couldn't care less about the exact properties of the random generator that program employs; all they'd care is how to get an otherwise working program to run as quickly as possible. This may not be good enough for serious investigators, but that's what having compatibility functions is all about. If you have ever ported a large and complicated program to another OS/compiler, you know how to appreciate this. For example, if I would attempt to identify and verify every algorithm in even in small GNU packages like Make or Sed (to say nothing about TeX and Emacs), I would be investigating forever instead of producing a working port in a few days or a few weeks.