On 03.11.2011, at 22:12, commits@source.squeak.org wrote:
Print Float with the minimal number of decimal digits that describe it unambiguously. This way, every two different Float will have a different printed representation. More over, every Float can be reconstructed from its printed representation.
Sounds desirable.
Maybe some won't like Float nan printString = 'Float nan'. But that's a detail we can eventually change back.
How about adding NaN as a global?
- Bert -
2011/11/4 Bert Freudenberg bert@freudenbergs.de:
On 03.11.2011, at 22:12, commits@source.squeak.org wrote:
Print Float with the minimal number of decimal digits that describe it unambiguously. This way, every two different Float will have a different printed representation. More over, every Float can be reconstructed from its printed representation.
Sounds desirable.
I forgot to say that we have to pay a penalty for this accuracy, around 4x... But to me, right > fast. it should be up to the programmer to decide to compromise right for fast, and in this case he can also sacrifice a few more digits and be even faster...
Maybe some won't like Float nan printString = 'Float nan'. But that's a detail we can eventually change back.
How about adding NaN as a global?
Yes, but also Infinity, and we would have to handle -Infinity with some trick... We can delay that decision. In fact I shouldn't have changed the NaN/infinity part, this was just to lazily reuse #storeOn:base: One change at a time, I will commit a version in trunk with unchanged behaviour for those exceptional cases.
Nicolas
- Bert -
squeak-dev@lists.squeakfoundation.org