Re: [PATCH]: Fix PR tree-optimization/21407

[Mark Mitchell <mark at codesourcery dot com>]

Daniel Berlin wrote:
> On Mon, 2005-05-09 at 20:29 -0700, Mark Mitchell wrote:
>
> > Daniel Berlin wrote:
> >
> > > On Mon, 2005-05-09 at 22:15 -0400, Daniel Berlin wrote:
> > >
> > >
> > > > Whee.
> > > > We pass the address of a derived class to the a call, then upcast it
> > > > back to the base class, effectively giving you the structure which
> > > > wasn't passed.  You can't do something like this in C, luckily :).
> > > >
> > > > Thus, we can only assume things about address of a structure field when
> > > > their is not inheritance involved in the thing we are taking the address
> > > > of.
> > > > Thus, a TYPE_BINFO check is necessary on the thing whose address if
> > > > being taken.
> >
> > Every class type has TYPE_BINFO.  So, you're not really checking for 
> > inheritance; you're just checking that it's a class.
>
>
> So i guess i'm looking for BINFO_N_BASE_BINFOS != 0?

I think that the type itself is considered a base class, so that this 
will still be 1.  But, maybe I'm wrong.  I'm also not sure this is 
consistent across languages.  At one point some languages had BINFOs 
pointing the other way, from base to derived, instead of from derived to 
base, or at least the documentation suggested that.

(I'm not actually sure what's getting confused here; given that 
dynamic_cast turns into a function call, I'd think the optimizers would 
see that like any other call.  And if they can eliminate the call, then 
they must be able to see that the value passed in is a derived class, so 
I'm not sure why you need to check to see if the class has any base 
classes.)

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304