Should math.divide(math.unit("1m"), math.unit("1cm")) return a number, or just a "unitless" Unit?

[mrft]

math.divide(Unit, somethingElse) returns a Unit in most cases, but a number when the 2 units are compatible.

This makes it hard for the typing system to know which type will be returned, which can lead to annoying type errors.

But apart from the type errors, I am wondering if math.divide shouldn't simply always return a (sometimes unitless) unit (except for when the input were 2 numbers to begin with)?

[josdejong]

That is an interesting idea, thanks for bringing this up. For these kind of consistency cases we have a configuration option predictable. We can change math.divide(unit, something) such that it always returns a unit when predictable: true.

This will be a breaking change so we need to schedule it for the first next major release.

[gwhitney]

I am confused. At current mathjs 15, it is the case that math.evaluate('10m/5m') returns the number 2 when config.predictable is false, and the unitless Unit 2 when config.predictable is true. So should this discussion be closed?

[mrft]

True: I just tested version 14, which returned a number in all cases, but version 15 returns an object when predictable = true.

Another question that might arise now is: should predictable = true be the default?

[gwhitney]

Predictable = true has wide ranging consequences, many of which seems unlikely to be desirable in a general-purpose calculation context. For example, sqrt(-4) is 2i with predictable = false, but it is NaN with predictable = true. The configuration option is there because for some uses, type consistency may be more important than mathematical fidelity. But I doubt there would be much support for the default configuration to be predictable = true.