Passing self by raw pointer

[mahkoh]

This seems like a useful feature especially for traits. Why isn't this allowed?

[bluss]

It was removed (while entirely buggy) in rust-lang/rust#26225

[mahkoh]

If it was removed without an RFC, then it should be possible to fix the underlying problems and re-enable them without an RFC too.

[bluss]

Not really. It was never an intentional feature. I can't find an issue or rfc for generalized Self in my searching at the moment, but you might look for that.

[mahkoh]

I can't find an issue or rfc for generalized Self

Most of the language wasn't introduced via an RFC.

It was never an intentional feature

From the PR you posted, it seems that it was explicitly allowed before it was removed: https://github.com/rust-lang/rust/pull/26225/files#diff-4ed25c00aceb84666fca639cf8101c7cL2177 How can this be unintentional?

I can't find an issue or rfc for generalized Self in my searching at the moment, but you might look for that.

Yes, generalized Self is one of the reasons that this should be allowed. It's one of the things that forces Box to be special cased.

[mahkoh]

In the absence of of &own, raw pointers are the only way to pass traits by value.

[arielb1]

It didn't actually work before it was removed - it was another syntax for by-value Self.

[quadrupleslap]

I think this might be a lot more useful for adding methods to C ADTs that are represented as enum {}:

&Empty is not okay, but you're working with *mut Empty and *const Empty in FFI.

Related issue for the nomicon.