Re: [PATCH] PR17913:[4.0 Regression] ICE jumping into statement expression

[Paolo Bonzini <bonzini at gnu dot org>]

> Can we first determine if this code is valid at all?  Someone
> optimistically added the ice-on-valid keyword, but I'm not so
> sure if this test case you are fixing is valid code at all.
> In fact, I'd argue that if we accepted it in the past, that
> was an accepts-illegal bug.
>
> The test case is the following:
>
> void f(void) { 1 ? 1 : ({ a : 1; 1; }); goto a; }
>
> So we are jumping into a statement expression.  That looks wrong.

Ouch.  Think about an extremely simplistic stack-based code generator, 
and feed it

  2 + ({ a: 3; });
  goto a;

This might be something like:

  push 2
a:
  push 3
  add
  goto a

That's a stack imbalance.

From some experiments, gcc manages to cope with this even with very 
complex nested expressions, but making labels local would make a lot of 
sense.  I second your opinion and suggest marking as ice-on-invalid, 
closing as WONTFIX, and opening another accepts-invalid PR.

BTW, C++ accepts it for 3.3; I don't have 3.4 at hand.

Paolo