Ah, we've gone full circle now.

It doesn't seem to be bad at all. There are always nitpicks like an unecessary .to_string() before parsing and type hint that would have been inferred, but it looks like pretty standard Rust code.

Though I do wonder, if you wanted to write a Perl solution, why not fork it in that language directly?

Just so you know, the purpose of .expect("text goes here") is to provide context if the unwrap panics. If you don't feel the need to provide context, a simple .unwrap() makes more sense.

I've been wondering, do you think random tests are helpful in kumites? It looks like asserting the equality of two implementations, which only works when the reference is known to be definitely correct (in which case, why not use it in the first place instead of doing testing?).

Or maybe you're writing them just for practice, I don't know.

It's a good implementation, but by nature converting a number into a string is much more complicated than a simple while loop which runs a dozen or so times.

Because sentinal values are a bad idea for many reasons, though you probably already learned this in the 4 years since.

However I'm still bamboozeled as to why it returns Option<Vec<i32>> as opposed to Option<[i32; 3]>. This isn't an arbitrary list, it's exactly 3 elements. I might even expect Option<(i32, i32, i32).

Mind clarifying? Rust implicitly returns the last expression in a function body if that's what you were wondering.

lol. I've checked on godbolt and we've been getting less and less efficient (at least in terms of assembly length). My initial solution was 186 lines, then 247, and yours is 504. It's not an actual speed comparison, but assuming it's a similar effect, that would show an inverse relation to legibility.

Yours is definitely the most legible, whereas my initial minimizes the number of passes.

Is there a reason usize is not used for the return type? It seems more natural if we're going to do conversions anyways, since a negative value is meaningless. Plus now there are situations where we can't give the correct answer. [i16::MIN, i16::MAX] would product an answer of 2^16 - 2, which is way to big to fit in an i16. I guess if we're going for minmalism u16 would be the appropriate type, but would require more conversion unless the input was also &[u16].

I understand that take, but not from a JavaScript developer. I'm intrigued. What do you think makes it worse than JavaScript?

Hmm, the descritions looked identical, but in the editor there was some weird old/===/yours formatting that I removed in the new fork. They should be identical behind the scenes too now.

The Rust translation needs some work.

• Tests should be in a tests module with #[cfg(test)]
• &[_] is more idiomatic and flexible and should always be preferred over &Vec<_>
• -> () return type is moot and discouraged.
• The spacing should be inline with rustfmt, which leaves an empty line between functions, and spaces lists as [a, b, c] instead of [ a,b,c ].
• The todo!() macro is a more idiomatic way to mark unwritten code, rather than a comment that fails compilation.
• The type for a function that takes 2 i64s and returns 1 is fn(i64, i64) -> i64. The user shouldn't have to write their own function signatures.
• The reference solution is needlessly complex.

Updated

Rust tests are run in parallel with nondeterministic ordering, though with only two of them it would be easier enough to submit until such a solution passed. Added random tests.

I've improved the non-sample assertion messages.

Random tests don't do anything if the test coverage is exhaustive. Though I might have missed some characters, so please let me know. I'm assuming only ascii printable characters are included.