Add repr(transparent) to hyper_context

[Manishearth]

hyper/src/ffi/task.rs

Line 504 in 1d4ff35

unsafe { std::mem::transmute::<&mut Context<'_>, &mut hyper_context<'_>>(cx) }

This comment is incorrect, repr(Rust) provides no such guarantees.

(Also we probably should make most of these FFI types repr(C)))

[seanmonstar]

Looks like cbindgen changes the header generation from typedef struct hyper_context hyper_context; to typedef Context hyper_context;. Can we make cbindgen ignore that?

[Manishearth]

No idea!

[Manishearth]

repr(C) is fine too here.

[Manishearth]

Actually I'm not sure of repr(C) on weirder platforms. repr(packed) probably?

[Manishearth]

Someone checked the ANSI C standard for me, repr(C) is fine

[Manishearth]

@seanmonstar You should pick which one of transparent/packed/C you want here, all three are fine but I don't know enough about cbindgen's behavior. Ideally this would be configurable.

[hjr3]

We want two things:

• a guaranteed layout for hyper_context so transmuate is safe
• to treat hyper_context as an opaque type

cbindgen does not provide to force a type with a guaranteed layout to be opaque.

We have a few options here:

1. Use a work-around where we use two different types and renaming. see hjr3@c5cc1e9#diff-fc56f58b6249ca3abcf76fb3641d25425c53e4d9f6c56514fcfe81fef72c8d3aR130-R143
2. Accept that hyper_context will not be opaque and add support for Context. See Opaque type for Mutex mozilla/cbindgen#930 (comment) which suggests a patch to https://github.com/mozilla/cbindgen/blob/ca78140c01518a655355f84da1f3872939123b66/src/bindgen/parser.rs#L470
3. Ask cbindgen to support an override to force an opaque type. See feat(ir) Add ability to ignore specific struct fields mozilla/cbindgen#579 (comment)

1 is the most pragmatic, 2 should probably be done anyways (adding Context to std types) and 3 is probably the best long-term approach.

If we want to go with 1, I can open a PR using my commit.