Behavior change for non-whitelisted types referenced by method

[upsuper]

Input C/C++ Header

struct T {};
struct U {
  void test(T a);
};

Bindgen Invocation

$ bindgen input.hpp --generate types --whitelist-type U

Actual Results

#[repr(C)]
#[derive(Debug, Copy)]
pub struct U {
    pub _address: u8,
}
#[test]
fn bindgen_test_layout_U() {
    assert_eq!(::std::mem::size_of::<U>() , 1usize , concat ! (
               "Size of: " , stringify ! ( U ) ));
    assert_eq! (::std::mem::align_of::<U>() , 1usize , concat ! (
                "Alignment of " , stringify ! ( U ) ));
}
impl Clone for U {
    fn clone(&self) -> Self { *self }
}

Expected Results

Not really expected result, but before #827, it generates

#[repr(C)]
#[derive(Debug, Copy)]
pub struct T {
    pub _address: u8,
}
#[test]
fn bindgen_test_layout_T() {
    assert_eq!(::std::mem::size_of::<T>() , 1usize , concat ! (
               "Size of: " , stringify ! ( T ) ));
    assert_eq! (::std::mem::align_of::<T>() , 1usize , concat ! (
                "Alignment of " , stringify ! ( T ) ));
}
impl Clone for T {
    fn clone(&self) -> Self { *self }
}
#[repr(C)]
#[derive(Debug, Copy)]
pub struct U {
    pub _address: u8,
}
#[test]
fn bindgen_test_layout_U() {
    assert_eq!(::std::mem::size_of::<U>() , 1usize , concat ! (
               "Size of: " , stringify ! ( U ) ));
    assert_eq! (::std::mem::align_of::<U>() , 1usize , concat ! (
                "Alignment of " , stringify ! ( U ) ));
}
impl Clone for U {
    fn clone(&self) -> Self { *self }
}

[upsuper]

I guess this is not really a bug, but a breaking behavior change which we should mention somewhere when people upgrade. (Do we have changelog?)

This definitely breaks Stylo, but it is probably because we are relying on something which we shouldn't rely on, and this can be easily fixed by simply whitelisting those types. So I guess this isn't really an issue.

@emilio, @fitzgen, what do you think?

[upsuper]

And I actually think the new behavior is probably more desirable (because it would generate fewer types that we don't need). We should probably add this to the test suite so that we don't regress this behavior.

[emilio]

Note that I'm regressing this in #835

[emilio]

And my idea is to un-regress it in #836.