Derive macro for Format does not set generic bound correctly

[Sympatron]

The currently implementation of the derive macro does set T: defmt::Format bounds for all generic parameters T. This is not allways correct. Consider the following example:

trait Foo {
    type Bar;
}
#[derive(defmt::Format)]
struct Baz<T: Foo> {
    field: T::Bar
}

This will generate the trait bound T: defmt::Format instead of T::Bar: defmt::Format.
I am not quite sure what the general rule is in this case, but the easiest thing to do would be to require : defmt::Format for all field types whether they are generic or not.

I could try to work on this, but I never wrote procedural macros before, so I am not quite sure how to solve this.
Currently this code is responsible for the where clause:

defmt/macros/src/derives/format/codegen.rs

Lines 47 to 55 in 06650b9

	for param in &generics.params {
	if let GenericParam::Type(ty) = param {
	let ident = &ty.ident;
	where_clause
	.predicates
	.push(parse_quote!(#ident: defmt::Format));
	}
	}

I think where_clause should be moved to EncodeData and constructed in encode_struct_data from the fields.

[jonathanpallant]

OK, interesting! We're going to need to think about this one, and we're also just about to head into Christmas break. We'll get back to this in 2024.

[jonathanpallant]

My colleague offers this expansion of some built-in derives:

trait Foo {
    type Bar;
}
#[derive(Clone, Debug)]
struct Baz<T: Foo> {
    field: T::Bar,
}

expands to

#[automatically_derived]
impl<T: ::core::clone::Clone + Foo> ::core::clone::Clone for Baz<T>
where
    T::Bar: ::core::clone::Clone,
{
    #[inline]
    fn clone(&self) -> Baz<T> {
        Baz {
            field: ::core::clone::Clone::clone(&self.field),
        }
    }
}

#[automatically_derived]
impl<T: ::core::fmt::Debug + Foo> ::core::fmt::Debug for Baz<T>
where
    T::Bar: ::core::fmt::Debug,
{
    #[inline]
    fn fmt(&self, f: &mut ::core::fmt::Formatter) -> ::core::fmt::Result {
        ::core::fmt::Formatter::debug_struct_field1_finish(f, "Baz", "field", &&self.field)
    }
}

It probably makes sense to emit something similar. I don't fully understand why T: Debug and T: Clone in this expansion, but if that's what the standard library does, it probably makes sense.

[jonathanpallant]

More details on this behaviour: https://smallcultfollowing.com/babysteps/blog/2022/04/12/implied-bounds-and-perfect-derive/

[ia0]

I see 2 non-exclusive options going forward:

1. Provide a #[defmt(bounds = "")] container-attribute to overwrite the inferred bounds. This is what serde does.
2. Infer the perfect bounds. This is also what serde does.

See jamesmunns/postcard#153 for a similar issue in another crate.

I think the first option is pretty easy to implement and would provide a solid workaround.

[Sympatron]

Didn't PR #800 already fix this? I think this issue can be closed now.

[ia0]

Is #800 part of 0.3.8? If yes, then I'll extract a minimal example from my concrete problem.

[ia0]

While trying to create a minimal example, I found the issue. The defmt:0.3.8 crate depends on defmt-macros:0.3.2 and my Cargo.lock was using 0.3.6 which doesn't seem to have the fix. I've run cargo update --precise=0.3.9 defmt-macros and now it works. It might be useful to bump the minimum version of defmt-macros in defmt as a minor bump since it adds a new feature.

[Urhengulas]

The next release will bump the minimum version.