On Friday, March 29, 2002, at 06:00 PM, David Vaughan wrote:
> If, however, you are thinking that in future we must deliberately > set a long precision to retain "best available" mathematical > accuracy in sequential calculations, rather than that being the > default, then I would not like it at all. I considered this: 6. If allowBigNumbers and numberFormat are empty (the default), they inherit from global properties that have the afore mentioned default values. In this case you would set the values once, and need not mention the properties in any handler. Unless option C (limitations on precision) is invoked by the implementer, there is no "best available"; that is limited only by available memory. You would chose a value for numberFormat that is consistent with your needs in precision and execution speed. The default will work in most cases. If you are doing least squares fit calculations, you might want more. There are no secret numbers passing along information in sequential calculations; all is above board. There is a potential problem in backward compatibility. The default numberFormat is 0.###### will not cover calculations on small numbers. One possibility is to increase the precision, but that would break current display code. It might be better to warn number crunchers. This came to mind (not exactly related to your comment): 7. The function pi() is added. The number of digits depends on numberFormat. If possible, the constant pi is kept for backward compatibility. _______________________________________________ improve-revolution mailing list [EMAIL PROTECTED] http://lists.runrev.com/mailman/listinfo/improve-revolution
