search

egallesio / STklos

STklos Scheme
http://stklos.net
GNU General Public License v2.0
67 stars 17 forks source link

Fix `(atan 0.0 0)` => +pi/2 #622

Closed jpellegrini closed 2 months ago

jpellegrini commented 2 months ago

And also (atan -0.0 0) => $-\pi/2$

This is explicitly given as a corner case in R7RS (in the table in the section for trigonometric procedures).

We also make (angle 0) trigger an error, since this would have to be exactly the same as (atan (/ 0 0)), which can't be computed.

Interestingly, only Chez does trigger the same error (for "(angle 0)"), but it is conceptually the right thing to do.

Tests included.

egallesio commented 2 months ago

Merged. Thanks.

I have seen that your return a NaN for (atan 0 0) whereas it is an error for Chez. Don't know what is better (I left your code).

jpellegrini commented 2 months ago

I have seen that your return a NaN for (atan 0 0) whereas it is an error for Chez. Don't know what is better (I left your code).

Hmm, indeed. This is what others do:

System (atan 0 0)
Bigloo error
Chibi 0.0
Chez error
Cyclone PI
Gambit 0
Gauche +nan.0
Guile 0.0
Kawa 0.0
Loko error
MIT 0
Sagittarius 0
Unsyntax 0.0

behavior seems to vary a lot.

My opinion: exact argumetns should not result in ineactness (not even NaN), so it would make more sense to do what Bigloo, Chez and Loko do. What do you think @egallesio ?

egallesio commented 2 months ago

You are right @jpellegrini , giving an error is a better option. I have changed the code to yield an error in this case.