Leaks when failures occur during unwinding

[nikomatsakis]

If a failure occurs during unwinding or executing a resource, we potentially leak. This is blocking #1078 (which also includes a reference to a testcase).

[catamorphism]

@bblum - is this still an issue after your work on failure?

[bblum]

To be honest, I'm not sure. I closed #1078 with some pretty easy code reorganisation, and the referred test case magically seems to work fine.

[catamorphism]

Then I'm going to close this. If it becomes an issue again, it should be pretty obvious.

[bblum]

Won't complain, but I'm worried we just might not have a test case. @nikomatsakis can you input?

[catamorphism]

Well, I'm of the philosophy that "if there's no test case, there's no bug" :-)

[nikomatsakis]

I am not sure what I meant about that testcase from #1078, it doesn't seem to fail during unwinding. Well, I recall that it could fail sometimes when invoking send() from within the dtor, in which case leaks would occur. Do we have a test with a class with a dtor that just fails when executed? Something like:

class Failer { let x: @int; new(x: @int) { self.x = x; } drop { fail; } }
fn main() {
     let f = Failer(@10);
     fail;
} 

Something like that would seem to be the test for this bug. For me at least, this test passes right now.