Re: turning throw into a jump

[Sylvain Pion <Sylvain dot Pion at sophia dot inria dot fr>]

On Tue, Jul 24, 2001 at 03:51:00PM -0700, Richard Henderson wrote:
> On Tue, Jul 24, 2001 at 09:48:14AM +0200, Sylvain Pion wrote:
> > void f()
> > {
> >   try
> >   {
> >     throw int();
> >   }
> >   catch (int i)
> >   {
> >   }
> > }
> 
> Oh, that.  No, we can't do that -- there's more going on behind
> the scenes that it looks.

Could you explain a bit more ?  Is it something that cannot be done because of
some standard requirement ?  Maybe it would be nice to have an option to
optimize this case, even if not enabled by default ?

> What we do now that we didn't before was handle cleanups and
> nested regions with branches.  E.g.
> 
> ...

I see.  That's nice.
BTW, I tried to compile (with gcc 3.1, -O3 -fomit-frame-pointer) :

void f()
{
  try {}
  catch (int i) {}
}

It produced :

_Z1fv:
.LFB1:
        subl    $12, %esp
.LCFI0:
        addl    $12, %esp
        ret

which is not exactly optimal, and not what a really empty function gets
compiled into.  Is there a way to fix this ?

-- 
Sylvain