Why does Pod not work with 2 generic-fields?

[DasLixou]

I have a struct

struct MyStruct<T> {
    a: T,
    b: T,
}

and deriving Pod on it errors because multiple generics don't seem to be allowed, but I can't find any documentation of the reason.
So, stupidly asked, why?

[Lokathor]

at first glance it looks like that code should work? did you also use repr(C) in the actual code? if repr(C) is in the picture, then your type is basically the same as [T; 2].

If that doesn't work then probably it's just that the proc-macro derive isn't sufficiently advanced, and someone could probably do a PR to support that. I'm going to make this into an issue so that this can be checked on longer term.

[DasLixou]

So yeah, the error (with repr(C)) is

Pod requires cannot be derived for non-packed types containing generic parameters because the padding requirements can't be verified for generic non-packed structs

[DasLixou]

So @Lokathor I did some digging, the important lines seem to be
https://github.com/Lokathor/bytemuck/blob/main/derive/src/traits.rs#L69-L78
The problem I have is that

• A - repr(C) triggers "isn't packed" and repr(packed) triggers "must be either C or transparent"
• B - repr(packed) seems to be kinda bad, the docs say it is not to be used lightly as it could cause UB.
  So my question is, what is the correct/allowed repr a struct like mine could have so I can change the macro accordingly?
  Thanks

[Lokathor]

It should be possible when T: Pod, because Pod doesn't have padding. Perhaps I'm forgetting something there. I think the derive is just being too conservative.

[Lokathor]

http://github.com/Lokathor/bytemuck/issues/278 let's move further discussion to the issue instead

[DasLixou]

Oh yeah, didn't know you made an issue, sorry