Re: Another chapter in the NeXT Tanh bug saga!
- To: mathgroup at yoda.ncsa.uiuc.edu
- Subject: Re: Another chapter in the NeXT Tanh bug saga!
- From: madler at pooh.caltech.edu (Mark Adler)
- Date: Wed, 6 Mar 91 21:24:22 GMT
In an attempt to clear things up a little bit here, this is the status of the tanh() bug on the NeXT: 1) There is a real bug, only on the 68040, in FTANH which is done in software on the 68040 (it was in hardware in the 68030/68882). Wolfram's code snippet shows this quite clearly. With the necessary #include <math.h>, of course. 2) The so called discontinuity caused by not using the #include for that same statement, is not a bug--it's a mistake in coding. And the result is different than the real discontinuity on the 68040. 3) NeXT has new floating point code from Motorola that fixes this and several other problems with the floating point operations on the 68040, and this will be distributed with version 2.1 of the NeXT operating system. 2.1 should be out Real Soon Now, and will be distributed as about four or so floppies to upgrade 2.0. It is an optional upgrade (which I disagree with, but who listens to me anyway?). Mark Adler madler at pooh.caltech.edu