Type restriction of String#rindex

[straight-shoota]

I have fixed this issue (#10956), then another issue crops up, text.rindex want to return a Int32 but the number given to it was explicitly Int64. you can see the patch https://github.com/iv-org/invidious/pull/2263/files#diff-c7fb42237f52a2684edf3d22d5f0b36f7f0ac8940f71c53886fe4a294ef73357L135

I wonder why isn't these issues showing up before. did something changed in the type inferencer? or it's simply a bug got fixed.

Originally posted by @tleydxdy in #10965 (comment)

[straight-shoota]

The return type restriction to String#rindex was added in #10583 and it appears that this change was not correct. The return type actually depends on the type of the offset parameter and that's unrestricted. So if you passed in Int64 as an argument, the return type would be Int64 as well.

We should revert this return type restriction. It's a breaking change and a regression in 1.1.0.

[straight-shoota]

This is a very common pattern, by the way: offset or similar parameters without a type restriction or with a wide restriction to Int that influence the return type of the method (or input type of a block parameter). Usually, only Int32 makes sense as a type because that's the type of collection sizes and indices. There's no real benefit from allowing a wide variety of types.

I think we should deprecate such unrestricted or broadly restricted parameters and move towards explicit Int32 restrictions.

[asterite]

I think so too.

Ideally, we would have an Int type that's used everywhere, so these issues would be much less common. Unfortunately we didn't go that route and now users have to juggle between many int types... 😢

I wonder if there would still be a way to change that for 1.x...

[straight-shoota]

/cc #8111