Euclideanring overflow

[jacereda]

No description provided.

[garyb]

This looks like it's about right to me, @hdgarrood what do you think? (I'm not really sure if saying that overflow = success is a safe assumption)

(Sorry about the delays on these @jacereda, will get these enum things in ASAP now - thanks very much for working on them).

[hdgarrood]

Yeah I agree, this assumption seems fine for Int but it’s not really clear to me that it makes sense for a general Euclidean ring. I think this could easily mask genuine errors for types where multiplication or division is harder to implement correctly, for instance.

I think the answer is really to use a newtype which only generates pairs of values which do not overflow when multiplied. Maybe problems like this are really a sign that the quickcheck approach of using a standard generator for any type is not the best approach here.

[jacereda]

I guess it isn't perfect, but isn't failing the test worse?

[hdgarrood]

No, I think a false negative (tests pass and miss an error they should have caught) is worse than a false positive (tests fail due to overflow which we were expecting anyway), especially since we can address the latter using a newtype.

[JordanMartinez]

Looks like the next action for this PR is to drop the overflow check and instead wrap Int in a newtype that limits how large the initial generated Int can be, so that an overflow error cannot occur. That would fix the false positive.

[JordanMartinez]

Superceded by #58