The Drop flag is a footgun for #[repr(C)]

[pnkfelix]

(imported from improperly closed bug #18380)

As suggested by @huonw:

the improper_ctypes lint should check for repr(C) on the RHS of a Drop impl. "warning: implementing Drop changes the representation of the type, conflicting with repr(C)".

(It should presumably also check for #[unsafe_no_drop_flag] and not warn if it exists.)

Until #5016 is done, we really should lint for this scenario. People have run into it

[pnkfelix]

triage: P-low

[pnkfelix]

oops.

triage: P-high

(but nonetheless not 1.0)

[geofft]

I feel like this should be stronger than a medium-priority lint. I feel like rustc should reject programs that attempt to impl Drop on a #[repr(C)] struct, for as long as that changes the size of that struct.

In particular, not fixing this by 1.0 would foreclose the ability to change the drop flag after 1.0, since that would change the size / ABI of #[repr(C)] structs, right? Or is ABI stability not a guarantee of #[repr(C)]?

[pnkfelix]

Yeah this label got auto renamed. I'm up'ing the priority

[pnkfelix]

triage: P-high

I'll assign to self too

[pnkfelix]

( @geofft I'm pretty confident we could fix this post 1.0, e.g. adding new warnings is allowed, and the representation issue would be "just a bug to be fixed." Nonetheless I I do I do want I do want to fix.)