fix: interpret empty true-expr of conditional as error

[gwhitney]

Resolves #3578.

[josdejong]

Thanks Glen!

I didn't remember this support for an implict start of a range :3. Maybe we should remove that too according to the same logic of not allowing missing parts 🤔. That would be a breaking change though, and unlike missing a part in a conditional operator, this is actually a useful feature and has a logical default value for the start. I guess it's best to keep supporting it. I'll add a few unit tests for it.

[josdejong]

Ah, I get it again: when ranges are used inside an index, the start and/or end is optional, so you can enter something like A[2, :] or A[:3]. In that case, a missing range start is interpreted as the start of the matrix, and a missing end range end as the end of the matrix. So it has context.

I think it was never intended to allow ranges used outside of an index to omit the start or end. Right now something like :3 works, but 3: not because there is no end available in the context. In short: it may make sense to make start and end of a range required when not inside of a matrix index. That will be a breaking change though.

[gwhitney]

I think it was never intended to allow ranges used outside of an index to omit the start or end. Right now something like :3 works, but 3: not because there is no end available in the context. In short: it may make sense to make start and end of a range required when not inside of a matrix index. That will be a breaking change though.

I would just leave it. :n.forEach(i => ...do stuff with i) is a nice idiom for the ubiquitous loop from 1 to n, and if the Range Matrix PR is eventually worked into shape and merged, it really will be just a loop. Also I could swear that while messing around with that PR i encountered some other example where it would make perfect sense to interpret the end marker outside of indexing a matrix. But I didn't write it down, sadly, because I was focused on other things. So I don't think it is doing any harm and if we leave it, hopefully such examples may come up again.

[gwhitney]

In fact, here's one: If you go to the mathjs REPL right now, you have:

A = [9, 8, 7, 6]
    [9, 8, 7, 6]

A[2:4]
    [8, 7, 6]

A[2:]
    [8, 7, 6]

subset(A, index(2:4))
    [8, 7, 6]

subset(A, index(2:))
    Error: Undefined symbol end

I do not see any reason the last one should not work, completing the analogy between the first and third. Should I file an issue for that, or just leave it as a little under-the-covers foible of mathjs?

[josdejong]

I do not see any reason the last one should not work, completing the analogy between the first and third. Should I file an issue for that, or just leave it as a little under-the-covers foible of mathjs?

I think the logic here is that a range like 2: is only supported inside the square brackets of a matrix index, like A[2:]. Not in the root of an expression like just 2: , nor as a function argument like index(2:). I think it's fine to leave this is it is right now, unless you strongly like to change this behavior.

[gwhitney]

a range like 2: is only supported inside the square brackets of a matrix index,

I agree. That's the only place that it's supported right now. But I don't see any logic for not supporting it anywhere else it makes perfect sense, such as subset(M, index(2:)). Just that we haven't done the work to support it there. I will see if I can incorporate it into the Range-is-Matrix PR, not sure.