[racket-dev] math collection [was: Hyperbolic functions]

From: Robby Findler (robby at eecs.northwestern.edu)
Date: Tue Jun 26 21:29:00 EDT 2012

This sounds like a terrible solution.

There are lots of places in our system where we just declare facts and
don't prove them and then use them for lots of things (often
optimizations). Why should this one be special?


On Tue, Jun 26, 2012 at 8:16 PM, Matthias Felleisen
<matthias at ccs.neu.edu> wrote:
> On Jun 26, 2012, at 9:12 PM, Neil Toronto wrote:
>> I'm addicted to optimizations. If I use Real -> Real, TR can't prove that (log1p 1.0) is Float and... hmm. I'll let Vincent explain why that's bad. :)
>> Another option is to provide both log1p and fllog1p. I just wrote fllog1p anyway.
> Sounds like the right solution -- and now you see how every type system forces you to duplicate code. -- Matthias

Posted on the dev mailing list.