Type of Functions returning an Iterator

[HadrienG]

@haselhorst

Then the error message is kind of misleading saying something about not beeing able to call map on trait objects… But ok, this reasoning is straight forward otherwise.

Yes, that's a product of the fact that the Rust syntax for dynamic dispatch (aka "trait objects") is to use just the name of the trait in a context where a type is expected, such as in &Trait. Again dyn trait will hopefully fix this someday, if people can agree on that breaking change. In the meantime, the compiler warning could be improved in isolation.

@haselhorst

I’m not sure if it is exactly the same. What would the compiler do if it sees an impl Trait in argument position? I suppose that it generates a specialized function for each concrete call as to know the size of the argument at compile time. Or is there some magic going on?

If it’s the case, then impl Trait in argument position would only be syntactic sugar for a generic <I: Trait> and an argument of type I?

This is exactly my understanding of it.

@haselhorst

For an impl Trait in output position however there is no quantification - and only one function with an (internal) concrete output type.

Exactly, impl Trait in argument and result position are not implemented in the same way, which is why impl Trait in argument position was not part of the original proposal. From my understanding, it's just that people thought it would have nicer ergonomics to be able to stick with a single syntax for "I want something that implements a trait" when they want to.