Improve the long explanation of E0207.

[nham]

The previous explanation does not seem to explain what it means for an
implementation parameter to be used or unused. The new explanation lists
the three ways specific ways by which an impl parameter becomes constrained
(taken from RFC 447).

This also adds a link to RFC 447.

The explanation has two different examples. The first is adapted from RFC 447,
and shows an instance of E0207 on a impl for a type. The second one is a trait
impl example adapted from issue #22834.

Closes #33650

cc #32777

r? @GuillaumeGomez

[birkenfeld]

nit: space after colon

[nham]

Addressed, thanks.

[GuillaumeGomez]

What generally bugs me in here is that it doesn't follow the waiting for getting merged RFC about error explanations.

[nham]

@GuillaumeGomez can you be more specific about how this explanation doesn't match? I just added a little bit more description to my first example, but other than that I'm not sure I understand what you would like to see changed.

[GuillaumeGomez]

Your changes are great, the point is not here. But currently, we're trying to uniformize the error explanations "template" by forcing people to follow the RFC I linked. This error explanation doesn't, and since you updated it, could you make fit to it please? 😺

[nham]

The problem is I've read the RFC multiple times and I don't understand how my explanation fails to adhere to the template. The only thing I can see is that my explanation does not literally contain the phrase "Example of erroneous code:", but I don't see how inserting this exact phrase improves the explanation.

---

Here's an attempt at trying to match my explanation to the template:

Error description

Any type parameter or lifetime parameter of an impl must meet at least one of
the following criteria:

• it appears in the self type of the impl
• for a trait impl, it appears in the trait reference
• it is bound as an associated type

Minimal example 1

For example, when implementing a method for a struct Foo, the following code
leads to an error:

struct Foo;

impl<T: Default> Foo {
    fn get(&self) -> T {
        <T as Default>::default()
    }
}

Error explanation 1

The problem is that the parameter T does not appear in the self type (Foo)
of the impl. In this case, we can fix the error by moving the type parameter
from the impl to the method get:

How to fix the problem

struct Foo;

impl Foo {
    fn get<T: Default>(&self) -> T {
        <T as Default>::default()
    }
}

Minimal example 2

As another example, suppose we have a Maker trait and want to establish a
type FooMaker that makes Foos:

trait Maker {
    type Item;
    fn make(&mut self) -> Self::Item;
}

struct Foo<T> {
    foo: T
}

struct FooMaker;

impl<T: Default> Maker for FooMaker {
    type Item = Foo<T>;

    fn make(&mut self) -> Foo<T> {
        Foo { foo: <T as Default>::default() }
    }
}

Error explanation 2

This fails to compile because T does not appear in the trait or in the
implementing type.

How to fix the problem, option 1

One way to work around this is to introduce a phantom type parameter into
FooMaker, like so:

use std::marker::PhantomData;

trait Maker {
    type Item;
    fn make(&mut self) -> Self::Item;
}

struct Foo<T> {
    foo: T
}

struct FooMaker<T> {
    phantom: PhantomData<T>,
}

impl<T: Default> Maker for FooMaker<T> {
    type Item = Foo<T>;

    fn make(&mut self) -> Foo<T> {
        Foo {
            foo: <T as Default>::default(),
        }
    }
}

How to fix the problem, option 2

Another way is to do away with the associated type in Maker and use an input
type parameter instead:

trait Maker<Item> {
    fn make(&mut self) -> Item;
}

struct Foo<T> {
    foo: T
}

struct FooMaker;

impl<T: Default> Maker<Foo<T>> for FooMaker {
    fn make(&mut self) -> Foo<T> {
        Foo { foo: <T as Default>::default() }
    }
}

Additional information

For more information, please see RFC 447.

---

Could you explain specifically what changes you think need to be made?

[arielb1]

Having the error message in a comment would be nice.

[nham]

@arielb1 Thanks, added.

[GuillaumeGomez]

Could you explain specifically what changes you think need to be made?

Sure, you just forgot a few little things. For example, before the very first you need to add "Example of erroneous code:".

In erroneous codes, you need to add the compiler error in a comment.

In other code examples, you need to add little comments to explain what you did.

[nham]

I've added Markdown headers so that the explanation is structured like this:

[Error overview]

### Error example 1

[Error example code along with a discussion of the problem and how to fix it]

### Error example 2

[Error example code along with a discussion of the problem and how to fix it]

### Additional information

[See RFC 447 blah blah]

I think the headers nicely organize this explanation. You can see the HTML version here Without the headers, the explanation is 120 lines long, which is lengthy enough to make headers work here, IMO.

[nham]

@GuillaumeGomez

Sure, you just forgot a few little things. For example, before the very first you need to add "Example of erroneous code:".

I am not sold on doing this yet. I do see your point that more structure is needed in the explanation. I have tried to explore an alternative above by using headings.

In erroneous codes, you need to add the compiler error in a comment.

Good point! Sorry I missed this in the RFC. This was already addressed above.

In other code examples, you need to add little comments to explain what you did.

Thanks, another good suggestion. I've added these in my last commit. Let me know if you think that's enough explanation.

[GuillaumeGomez]

It's really good! The "erroneous code example" sentence is still missing but considering the quality of the rest of the error explanation, I think we can exceptionally get over it. Thanks a lot for your work @nham!

Now it just remains one last thing: can you squash your commits please? 😃

[nham]

Sure! Squashed

[GuillaumeGomez]

Thanks!

@bors: r+ rollup

[bors]

📌 Commit 7d78436 has been approved by GuillaumeGomez