[PRL] Programming Languages in the Code of Federal Regulations

Pete Manolios pete at ccs.neu.edu
Mon Apr 26 19:36:26 EDT 2010


Following the IEEE 754 floating-point spec is not easy. For example,
see "The pitfalls of verifying floating-point computations" by David
Monniaux. Even if floating point is handled in a reasonable way, it is
still floating point, so many "obvious" properties, like associativity
of addition, do not hold. That is, there is no way to satisfy
Matthias' criterion number 2 if we use floating point (unless
"intended mathematical meaning" below means IEEE floating point).

2 Programs should have the intended mathematical meaning.



On Mon, Apr 26, 2010 at 6:19 PM, Paul Steckler <steck at stecksoft.com> wrote:
> R6RS suggests, but doesn't require, that Scheme implementations follow
> the IEEE 754 floating-point spec.
> So you don't get behavior guarantees simply by choosing "Scheme" as
> your language.
>
> Years ago, Turbo Pascal had a binary-coded decimal (BCD) type that was
> especially suited for doing
> financial calculations, avoiding some of the representational problems
> of IEEE 754.  Something like
> that might figure in a DSL for the feds.
>
> -- Paul
>
> On Tue, Apr 27, 2010 at 12:31 AM, Matthias Felleisen
> <matthias at ccs.neu.edu> wrote:
>>
>> Following Jay's and Jordan's example I have submitted my own response. A
>> scribbled version is available at:
>>
>>  http://www.ccs.neu.edu/home/matthias/Thoughts/Python_for_Asset-Backed_Securities.html
>>
>> Thanks for all the feedback -- Matthias
>>
>>
>
> _______________________________________________
> PRL mailing list
> PRL at lists.ccs.neu.edu
> https://lists.ccs.neu.edu/bin/listinfo/prl
>



-- 
Pete Manolios
Northeastern University
http://www.ccs.neu.edu/home/pete



More information about the PRL mailing list