No drop flag check for DST structs

[nrc]

Since drop flags come after all named fields and the last field in a DST struct is unsized, finding the drop flag is a bit tricky. Therefore, it is currently not checked. A possible fix is to find the alignment by taking the size of the struct from its vtable and subtracting the size of the drop flag. However, since the plan is to remove the drop flag anyway (#5016), I have just left a FIXME and issue a warning when we skip the check.

cc @pnkfelix

[steveklabnik]

@nrc is this still relevant today?

[nrc]

Yes, afaik. cc @pnkfelix

[tailhook]

Does this mean if i get:

warning: Ignoring drop flag in destructor for std::sync::mutex::Mutex<Something>because the struct is unsized. See issue#16758

Then I will get a memory leak?

[pnkfelix]

some relevant recent discussion from #rust-internals IRC channel: https://botbot.me/mozilla/rust-internals/2016-04-28/?msg=65078367&page=3

[chriscoomber]

Just hit this when coercing a Box<RwLock< (struct) >> to a Box<RwLock< (trait) >>.

(Aside: it is of course illegal to coerce a RwLock< (struct) > to a RwLock< (trait) > - I don't really understand why the outer box makes this legal.)

Here's a small example that produces this warning:

use std::sync::RwLock;
use std::boxed::Box;

trait Trait {}
struct Thing;
impl Trait for Thing {}

fn main() {
  let _: Box<RwLock<Trait>> = Box::new(RwLock::new(Thing));
}

Will I get a memory leak with this code?

[jonas-schievink]

This should no longer be relevant after #35764 lands since it will remove in-struct drop flags

[Osspial]

Should this warning be removed now? Non-zeroing drops have landed, so there isn't a flag to check.

[arielb1]

Sure enough. There is no such a thing as a drop flag in the MIR (1.13+) universe.