[stdlib] Audit uses of StaticIntTuple

[JoeLoser]

Similar in spirit to #2514, it would be nice to audit our use of StaticIntTuple in the standard library which is a mix of InlineArray[Int, size] with some odd APIs that make it represents dimensions or tuple-like things.

Where it makes sense, it would be good to simplify APIs to not be coupled to this unique/non-generic type. For example, this Int._divmod function should just return a Tuple[Int, Int] instead of using this unique type. There's no reason to prefer StaticIntTuple[2] over Tuple[Int, Int] here.

Feel free to ask questions about specific APIs if you're not sure whether to change it or not.

[YichengDWu]

I propose to use 'NTuple[2, Int]' as the type alias for 'Tuple[Int, Int]', and 'NTuple[N, T]' in general, so that it can be easily dispatch on 'N'.

[JoeLoser]

I propose to use 'NTuple[2, Int]' as the type alias for 'Tuple[Int, Int]', and 'NTuple[N, T]' in general, so that it can be easily dispatch on 'N'.

Can you elaborate on the value of the type alias? I'm not sure I understand the full value yet. Are there places you have in mind in the library that would benefit from this?

[YichengDWu]

NTuple[2, Int] = Tuple[Int, Int]
NTuple[3, Int] = Tuple[Int, Int, Int]

Or In general I should be able to write

alias NTuple[N, T] = Tuple[VariadicPack[T, N]]

Then I can dispatch on NTuple[5,Int] instead of Tuple[Int, Int, Int, Int ,Int], it's just shorter.

[artemiogr97]

@JoeLoser while working on this I found a potential issue, StaticIntTuple is imported in builtin/int.mojo which exposes it everywhere as all builtins, is exposing stuff in this way intentional?

[gabrieldemarmiesse]

@JoeLoser while working on this I found a potential issue, StaticIntTuple is imported in builtin/int.mojo which exposes it everywhere as all builtins, is exposing stuff in this way intentional?

Maybe we should open an issue to have a new mechanism in the stdlib to define what exactly does not need import and what needs imports.

This issue has bitten me quite a few times.

EDIT: Created the issue and proposed a fix that must be implemented partly on the compiler side:

• [BUG] Functions, traits, structs and aliases are leaking into builtins and are importable from anywhere #2529

[artemiogr97]

Seems like StaticIntTuple is used a lot in the tests, for those cases it should be possible to replace it once Tuple conforms to the Stringable trait

[msaelices]

Off-topic: Why _divmod instead of the Python __divmod__?

[gabrieldemarmiesse]

@msaelices it's ongoing changes, you can take a look at #2335