Zero-sized structs should be FFI-safe

[mzabaluev]

If I get a struct by pointer from a foreign library, and I know the structure cannot safely be copied because the actual object is larger than the struct, I want to declare it as something like this:

#[repr(C)]
struct A {
    // ...
    no_copy: std::kinds::marker::NoCopy;
}

Now the compiler complains at use site that the struct is not foreign-function-safe.
As the markers are zero-sized, they have no bearing on the layout and no consequences for the structure data, so they, as long as any other zero-sized structures, should not waive FFS.

@thestinger suggests the same should apply to structures with one field as well, but I'm not so certain about that: there could be optimizations regarding alignment which would not be compatible with the (default) C ABI.

[zwarich]

Strictly speaking, the grammar for the C language does not allow you to define structs with no members, and the specification confusingly claims separately that they have undefined behavior. The C++ language allows them, but requires that they have nonzero size.

[mzabaluev]

@zwarich: This is about struct definitions in Rust which don't have to follow C word by word.

A tool for converting Rust struct definitions into C or C++ ought to be able to detect and remove zero-size members.

[thestinger]

GNU C supports zero-size structs, and they are zero-size (unlike C++ / GNU C++). It's similar to the situation with packed structs which aren't defined by standard C either.

[mzabaluev]

There may be a special case for zero-size structs by themselves being FFI-unsafe when used by value.

[kornelski]

Related #20660

I'm using 0-sized structures to represent unknown C types (C's struct foo; without definition), so I'm hoping that pointers to such structs are OK.

[thestinger]

@pornel: Rust has truly opaque types (enum foo {}) so there's no need to abuse zero-size types for that.

[thestinger]

Raw pointers to anything are FFI safe, since they have a uniform representation.

[kornelski]

(edit: nevermind the compile problem)

I've been assuming struct foo; is more appropriate, since it supports neater syntax without {}.

[steveklabnik]

I'm pulling a massive triage effort to get us ready for 1.0. As part of this, I'm moving stuff that's wishlist-like to the RFCs repo, as that's where major new things should get discussed/prioritized.

This issue has been moved to the RFCs repo: rust-lang/rfcs#760