"let _ = foo();" is differently effectful from just "foo();"

[bblum]

See #2734. In that sample code, which segfaults, removing the characters "let _ =" causes it to no longer segfault. No idea why; maybe it causes it to optimise away the destructor call?

[catamorphism]

I'll see if this is still a bug now.

[bblum]

This is still in the air. The following code prints "First" then "Second", but without the let _ =, it prints "Second" then "First".

class defer {
    f: fn@();
    new(f: fn@()) {
        self.f = f;
    }   
    drop { self.f(); }
}

fn main() {
    let _ = do defer {
        #error["Second"];
    };  
    #error["First"];
}

I'm not sure if the answer is "Well, too bad, there are no guarantees on when destructors get run", but I want to add this defer thing to libcore, and can't in good conscience as long as its behaviour is unpredictable.

[catamorphism]

I have a fix for this, which I'm testing. However, for reference, what you want to do to get the behavior that bblum actually wanted was to write:

let _x = do ...

which means the destructor for the RHS won't run until the end of the enclosing scope, since _x is a perfectly cromulent identifier.

[catamorphism]

My opinion is that let _ = e; should behave just like e; Feel free to argue about it, but for now I'm closing this.