#5145 (comment)

I'm curious why we want this now and not before.

#5145 (comment)

I'm curious why we want this now and not before.

It's still the same issues as before. I think the biggest problem is serialization. It really only adds noise to the serialized output. And if we make it Reflect, then it becomes easy for library authors to forget to ignore it themselves, which means downstream users will see it if they serialize it.

If we do this, we may want to consider skipping it during serialization (and other operations such as reflect_partial_eq).

do we want really need this? i feel like PhantomData is supposed to act as if its not there at all (its a 👻 after all). is there anything not accomplished by annotating #[reflect(ignore)] or is it simply an ergonomic improvement? i'd really rather not have all the "noise" that @MrGVSV mentioned (not just in serialization mind, also in runtime and binary size) just for the sake of a few less tokens to type out.

tldr; the implementation is correct but i'm not sure the feature is desirable.

Maybe it would be better to instead implement a way to always ignore marker types?

I don't know how to fix the unit test. Someone else can give me a pointer or do it themselves if they want. :)

Could there be a way of saying to ignore a whole type when you register it?

Yeah, that would be my preferred solution here.