Fix an UB in layout tests

[Apostoln]

• Replaced dereferencing of null ptr with
  zero bit pattern + transmute + std::ptr::addr_of!
  to avoid UB. It affects only checks of fields offsets
• Overwritten expected with BINDGEN_OVERWRITE_EXPECTED=1
• Overwritten test_multiple_header_calls_in_builder and test_mixed_header_and_header_contents
  manually because Some expected are not overwritten with BINDGEN_OVERWRITE_EXPECTED=1  #2054
• Do not check the layout for repr(C) unions
• Do not call the destructor of tmp struct to avoid other UB

Close #1651

Note: This way is not ideal (see the latest comments in related issue). Zero bit pattern can be replaces with manual initialization of the fields, but I have no time to investigate how to do it when

1. The item is a struct contains other structs.
2. The item is a union.

However, I think this way is good enough, because there no UB anymore. The only one unsafe block with transmute will not cause an UB, because all C types allows zero bit pattern. Anyway, it's better than current implementation with UB and warnings.

Any feedback or commits are appreciated. Thanks.

[highfive]

Warning

• These commits modify unsafe code. Please review it carefully!

[Apostoln]

r? @emilio

[Apostoln]

r? @fitzgen

[emilio]

This is still not guaranteed to be ub-free, think of a rust enum field like:

enum Foo { Bar = 1, Baz, }

Using a zeroed buffer for that seems like itd still be UB.

But anyhow this is better than what we're doing so it seems ok to me with the suggestion below.

[emilio]

This is not guaranteed to be aligned right? I think using std::mem::zeroed should be better.

[Apostoln]

Thanks, fixed

Though, I don't catch why there is a wrong alignment. AFAIU, In case of mem::zeroed the padding bytes are not zeroed but the total behavior is the same.

[emilio]

No, let's say you have a struct with an u32 member. The compiler would make sure that that is aligned appropriately, but if you use a flat array of u8s then it's not.

[Apostoln]

@emilio
Ok, thank you. Replaced it with mem::zeroed

[emilio]

It seems there are still some tests to update, but I'll try to get to this myself if you can't.

[Apostoln]

Yeah, I've missed errors on CI, locally it was OK. I'm going to fix it but a bit later.

[nbdd0121]

This potential UB can be avoided:

Suggested change

	let struct_instance = unsafe {
	std::mem::zeroed::<#canonical_ident>()
	};
	// Get the pointers to the struct and its field
	let struct_ptr = &struct_instance as *const #canonical_ident;
	let struct_instance = core::mem::MaybeUninit::<#canonical_ident>::uninit();
	// Get the pointers to the struct and its field
	let struct_ptr = struct_instance.as_ptr();

[bors-servo]

☔ The latest upstream changes (presumably 310f7f8) made this pull request unmergeable. Please resolve the merge conflicts.

[kulp]

This might be obsoleted since #2203 was completed.

[kulp]

• Do not check the layout for repr(C) unions
• Do not call the destructor of tmp struct to avoid other UB

It appears that this (like #2064) has a little more in it than #2203 did, but since #1651 is now closed, I am inferring that this PR is no longer needed, so I am closing it.

If I missed something, let me know.