Problem with large number precision

[ahungry]

First off, I noticed on the Janet samples for Rosetta Code factorial, any of them with a factorial of 1e3 would give a return value of "inf" (not correct).

As I began to play with large numbers (I assume these are all floats and not actually ints at all) I observe some logical incorrectness:

 (let [bignum 1000000000000000000000000000000000] (= bignum (+ 1 bignum)))                                                        

   true

a large number that isn't actually "inf" can't be equal to itself + 1
@bakpakin ^ any idea?

   (let [bignum 1000000000000000000000000000000000] (= bignum (+ 1e17 bignum)))                                                     

   false
   (let [bignum 1000000000000000000000000000000000] (= bignum (+ 1e16 bignum)))                                                     

   true

it would appear after some precision point anything afterwards is disregarded, resulting in false comparison results.

I think bignum should be handled in a separate module or package, or baked in for non-float based numbers (no decimal points), or else numbers exceeding the precision that can be properly handled should trigger errors.

What do you think?

[bakpakin]

Janet numbers are IEEE754 double precision floating point numbers, and as such follow al of the semantics these numbers obey. Which includes limited precision. This will not change.

Edit: This includes on parsing - there is no plan to parse large numbers or extremely precise numbers into a different number type.