[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: straw-man [was Re: arithmetic issues]
On Tue, 24 Jan 2006, Bradley Lucier wrote:
>On Jan 23, 2006, at 4:16 PM, Alan Watson wrote:
>> I applaud the authors of SRFI-77 for providing the tools to solve
>> an important problem (implementing the full tower). However, once
>> they've solved the problem, I'd like them to put their tools away
> ... let me just say---these tools are useful in many other application
>domains, not just in building the full numeric tower, and I want them
>available for use in these other application domains.
Hmmm. If there are domains in which they are useful, then
should they be available as a library (ie, a set of bindings to
routines that are normally excluded from the environment, but can
be included explicitly given some directive)?
I will say this; I do a fair amount of what would be considered
by most to be hardcore math; fifth-order regressions, neural
network convergence tests, heuristics that attempt to predict
the movements of financial instruments and build economic models,
natural language frequency tests and correlation tests on corpora,
etc... and *every* time I become aware that I have fixed-width
hardware math, it is because fixed-width hardware math has caused
a bug. Underflows in frequency ratios and rollover errors in
integer counts are *ALWAYS* an error rather than a desired behavior
in my work.