https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122835
--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Regarding asm goto, I don't see that documented anywhere, anyway, gcc has
behaved that way forever.
Anyway, unlike computed goto/longjmp, I think asm goto is at least in theory
changeable, the question is if we do want that.
In e.g.
struct S { S (); ~S (); };
void
foo (int x)
{
{
S s;
if (x == 1)
goto l1;
asm goto ("" : : : : l2);
}
l1:;
l2:;
}
it is the eh pass which changes following 2 lines in:
// predicted unlikely by goto predictor.
goto l1;
...
}
finally
{
S::~S (&s);
}
}
finally
{
s = {CLOBBER(eos)};
}
l1:
to
// predicted unlikely by goto predictor.
goto <D.3002>;
...
<D.3002>:
S::~S (&s);
goto <D.3005>;
...
<D.3005>:
s = {CLOBBER(eos)};
goto l1;
So, supposedly we could do the same from asm goto, change the label to an
artificial label which handles the needed cleanups on the way out.
