Re: Why does casting a function generate a run-time abort?

[Ziemowit Laski <zlaski at apple dot com>]

On 21 Jun, 2004, at 21.42, Mark Mitchell wrote:

> Ziemowit Laski wrote:
>
> > Hm... I can't really claim that I never consider aesthetics in my
> > hac^H^H^Hengineering activities, although technical considerations
> > always trump them, as they should.  I guess I'll just wait for your
> > METHOD_EXPR work, and then will decide which approach is more
> > disgusting. :-)
>
> I'm going to jump in here and say that I strongly agree with Richard.
>
> In over a decade of working on front ends, I've long since concluded 
> that you want to express special things as special nodes.  A good 
> example is exception-handling; a C++ front end could just transform it 
> all into setjmp/longjmp (as Cfront did), but then you get bad 
> performance.

Special nodes are good if you need to propagate semantics to the back 
end, as your C++ EH example shows.  In ObjC, all we need to do is cast 
a function to a different signature and then call it.  While you 
certainly _may_ create a special node for this, I don't see what this 
buys you...

>  If you want good performance, you must send it to the back end as 
> try/catch.  As Richard says, virtual function calls really ought to be 
> the same; there's no reason to devolve into the vtable-indexing C-ish 
> stuff until you have optimized as much as you can.

...except that ObjC message dispatches have nothing whatsoever to do 
with C++ or Java virtual dispatches.  As I said, I'll wait for 
Richard's METHOD_EXPR to see if his stuff is applicable.

--Zem
--------------------------------------------------------------
Ziemowit Laski                 1 Infinite Loop, MS 301-2K
Mac OS X Compiler Group        Cupertino, CA USA  95014-2083
Apple Computer, Inc.           +1.408.974.6229  Fax .5477