Comment # 3 on bug 207 from art.08.09@gmail.com

(In reply to Iain Buclaw from comment #1)
> It should be the same as calling a function which noreturn attribute set.

Hence $summary ;)

I was surprised that this was not already the case, and by
the cost (I was playing with the new arithmetic builtins and
the impact on calling/surrounding code was significant; the
dead post-throw code emitted isn't that much of a problem).

BTW, there's a similar issue w/ D's asserts (_d_assert) in
non-release builds.
But there, I'm not sure if assuming no return is legal (no
spec is fun...). Still, I ended up using:

   { assert (0); import gcc.builtins; __builtin_trap(); }

instead of `assert(0)`. At least for the `assert(0)` case
emitting that trap unconditionally seems OK (it's what
happens in release-mode anyway, so recovery is already
impossible).