Numeric Comparisons

[michaelhkay]

Numeric Comparisons

[ChristianGruen]

A good summary.

Would being bold mean that <x>3.1</x>[. = 3.1] would return no more result in the future? If the answer is yes, could we be even bolder and convert untyped atomics to decimals (instead of doubles)?

[michaelhkay]

(a) Yes, and (b) yes, that would be even bolder - and would no doubt lead to a further batch of compatibility problems.

I have started thinking about whether it might be possible to have an "arithmetic mode" in the dynamic context, set to either "double" or "decimal", used to control all of these things. But it's a big sledgehammer: would people have fewer problems than they had before, or would we just introduce new problems?

[ChristianGruen]

(a) Yes, and (b) yes, that would be even bolder - and would no doubt lead to a further batch of compatibility problems.

In that case, I’d tend to be timid (to be very bold, I know too less about the further batch), and I’d tend to avoid new (explicit) options. As we talk about fairly fundamental features, I assume it would make everyone’s life more difficult.

would people have fewer problems than they had before,

Aside from the (now ancient) confusion over implicit double conversions, I was not aware of the problems myself. Do you think they could be summarized to assess their seriousness?

Looking forward to other opinions as well.

[michaelhkay]

Well, apart from the usability problems caused by the fact that people don't understand floating-point arithmetic, I think the number one issue we are trying to solve is the transitivity problem. That primarily affects grouping and sorting, and we could try to limit the compatibility hit by only making changes to the places that are most directly affected: distinct-values(), group by, order by etc. Alternatively, we could restrict these so that mixed numeric types aren't allowed (or are treated as always distinct).

Looking at the list, I think it's probably deep-equal that has opened the can of worms, because we use it both in places where strict transitivity is needed, and in places where a "more forgiving" equality test is desirable. Perhaps if deep-equal() had an option to control numeric comparison, life would be a bit easier?

[ChristianGruen]

Perhaps if deep-equal() had an option to control numeric comparison, life would be a bit easier?

That could be an “option” indeed (given that we already have so many other options now… And I assume that most people won’t care anyway).