`Option<*T>` should be represented as a pointer

[nikomatsakis]

Just like Option<~T> and Option<&T>.

cc @jld

[nikomatsakis]

Put another way, this test should pass:

assert_eq!(sizeof::<Option<*int>>(), sizeof::<uint>())

[jld]

As mentioned in #10571, this depends on *T being non-nullable and vice versa.

[brendanzab]

Not sure if this is related, but we are also having difficulties with Option<extern fn()> on i386. See PistonDevelopers/glfw-rs#67:

On i386, rustc expects Option instances to be returned on the stack. The caller
passes a pointer to allocated stack space and the callee fills it in. This is
not the calling convention expected by C functions like glfwSetErrorCallback
that return a pointer in EAX.

Somehow this led to a segfault on i386 because the stack pointer was incorrect
when returning from the caller. This didn't happen on x86-64.

I couldn't find any documentation on using Option types with the FFI. If this is supposed
to work then I'll file an issue against rustc instead.

— @rlane

[jld]

Function pointers should already be non-nullable (and thus an Option of one will be optimized as expected), but... the Option will still wrap a layer of LLVM struct around them, which I think changes the calling convention on i386?

That could be changed without too much trouble, I think, but it probably ought to be a separate issue.

[alexcrichton]

Closing because #10571 was closed as well.