On 01/04/2013 10:38 AM, Vincent St-Amour wrote:
At Thu, 03 Jan 2013 21:09:52 -0700,
Neil Toronto wrote:
I thought it would be helpful to find the most precise types possible
for numeric functions. I wrote a program that infers them using a few
thousand representative argument values, which have
At Thu, 03 Jan 2013 21:09:52 -0700,
Neil Toronto wrote:
>
> I thought it would be helpful to find the most precise types possible
> for numeric functions. I wrote a program that infers them using a few
> thousand representative argument values, which have been chosen to
> exhibit underflow, ove
At Thu, 03 Jan 2013 21:09:52 -0700, Neil Toronto wrote:
> 1. Implementation of `sqrt': most Single-Float-Complex inputs yield
> Float-Complex outputs. The type is sane, but the behavior isn't:
>
> > (sqrt 1.0f0+5.2f0i)
> - : Single-Flonum-Complex
> 1.7741590586312472+1.465482
On Thu, Jan 3, 2013 at 11:09 PM, Neil Toronto wrote:
> (In particular, platform-independent Index and Fixnum types scare me.)
Platform-independent Fixnum is easy -- just if |x| <= 2^30, then x is
a fixnum on all platforms Racket supports. [1]
Index is a little trickier, since it's not documented
I thought it would be helpful to find the most precise types possible
for numeric functions. I wrote a program that infers them using a few
thousand representative argument values, which have been chosen to
exhibit underflow, overflow, exactness preservation (e.g. perfect
squares for `sqrt' and
5 matches
Mail list logo