Inlining breaks LLDB expression evaluator

[vadimcn]

Compile this code with -Copt-level=0:

#[inline(never)]
fn zzz() {}

#[inline(always)]
fn inlined_fn() {
    let a = 1;
    zzz(); // #break
}

fn main() {
    inlined_fn();
}

In LLDB, set a breakpoint on line 8, then run the program.
After breaking, type print a'. Result:

error: use of undeclared identifier 'a'
error: 1 error parsing expression

Interestingly, the frame variable command shows the expected value: (int) a = 1.

Meta:
LLDB version: lldb-360.1.25 (OSX XCode 8)
Rustc version: rustc 1.10.0 (cfcb716 2016-07-03) x86_64-apple-darwin

[vadimcn]

Note: this is very likely an LLDB bug, because an equivalent C++ code exhibits the same behavior.

__attribute__((noinline))
void zzz() {}

__attribute__((always_inline))
void inlined_fn() { 
    {
    int a = 1;
    zzz(); // #break
    }
}

int main() {
    inlined_fn();
    return 0;
}

The extra lexical scope in inlined_fn is important - without it, evaluation produces the correct result: (int) $0 = 1

GDB works correctly.

[michaelwoerister]

That makes me sad :(

[Mark-Simulacrum]

So if this is an LLDB bug, then it seems like there's no use in us tracking it. As such, I'm going to close.