add list<string> -> string to last

[amtoine]

this should

• close external command output type is not string? #11134

Description

this is band-aid...
but it should address the issue in #11134 until we have a better long-term fix for this i/o types bug 😇

User-Facing Changes

the following will now parse and run fine

def get-initial-commit []: nothing -> string {
    ^git rev-list HEAD | lines | last
}

Tests + Formatting

After Submitting

[fdncred]

I think the test error in CI is the same problem but with the get command. I wish there was a good way to test all this input_output_type stuff.

[amtoine]

ugh, what is that? 😢
i'm worried that changing the i/o types of get would ripple onto more commands without end 😅

[fdncred]

I think the problem is just that our input_output_types aren't defined properly, nor do they work properly on all commands.

[WindSoilder]

I've checked input/output type checking logic.

I think the checking logic is good. But we need to follow a rule:

If we have an input type that may generate multiple output type, last as example, we need to annotated it to <input_type> -> any, and we can't add other things about this input_type.

So you can have list<string> -> any for a command.
But you can't have both list<string> -> list<string>, list<string> -> int for a command.

[amtoine]

So you can have list<string> -> any for a command. But you can't have both list<string> -> list<string>, list<string> -> int for a command.

yeah that's unfortunate but makes sense 👍
or we would need to add some kind of anyof thingy i describe in #10869 🤔
so that you could have

[
    list<string> -> anyof<list<string>, string>
    list<int> -> anyof<list<int>, int>
    ...
]

[amtoine]

good catch @WindSoilder 👌

it's unfortunate but at least it works 😌

[sholderbach]

Thanks for tweaking the commit title @WindSoilder !

[WindSoilder]

[
    list<string> -> anyof<list<string>, string>
    list<int> -> anyof<list<int>, int>
    ...
]

@amtoine Unfortunally, I don't think anyof in input_output_type is well fitted into our current design. It'll impact the parsing performance.

If we have the following commands' signature:

1. command a: list -> anyof<list, string>
2. command b: string -> anyof<int, list, string>
3. cmmand c: string -> int

Then if we run ["1"] | a | b | c
We need to check for all possible anyof type.