Bug 19512 – Exception during scope(exit) of an exception yields undefined behavior

An exception during scope(exit) handler of an already-propagating exception does not get handled correctly. Under some circumstances, it triggers missing destructor bugs too. Example program: // use puts, not writeln, because puts is nothrow. any code that may throw prevents some of the bug from reproducing import core.stdc.stdio: puts; struct C { this(int x) {} ~this() { puts("C.~this()"); } } struct D { auto c() { return C(1); } } D d; void foo() { with(d.c) { puts("This code is reached"); // The exception here seems to long-jump back to the original exception propagation try { throw new Exception("Exc"); } catch(Exception e) { puts("This handler is never reached"); } // Iff ONLY nothrow things are done below here, then d.c's dtor does not run at all! // THIS CODE IS NOT REACHED: puts("This code is NOT reached, and C's destructor never runs!"); } } void main() { scope(exit) { foo(); } throw new Throwable("Bomb"); }

Using the latest version git master, I am now getting: This code is reached This handler is never reached This code is NOT reached, and C's destructor never runs! C.~this() object.Throwable@(0): Bomb This looks like the correct behavior, so it seems this has been fixed.