Re: objc native exceptions

[Richard Frith-Macdonald]

On 27 Jun 2008, at 17:57, Nicola Pero wrote:

> On 27 Jun 2008, at 16:35, Richard Frith-Macdonald wrote:
>
> > I was wondering if there was any good reason why gnustep-make  
> > doesn't enable native exceptions by default if the compiler  
> > supports it.
>
> There are some overheads in enabling native exceptions.
>
> I'm not too familiar with the issue, I only know that when I added  
> support
> for native exceptions to gnustep-make and tried to make it the  
> default, someone
> else came out and turned it off again by default claiming that the  
> overheads
> are high.
>
> That's fine btw - the main problem is that because you need a  
> special configure flag to turn them on,
> nobody seems to be using native exceptions, so they don't get used /  
> tested / finished. :-(
>
> Which is a pity, because they are a nice feature to have.

Apparently one problem is that they rely on unwinding the stack, and  
if ffcall (and perhaps ffi) has been used to forward a method using an  
invocation (including any distributed objects call) then the stack has  
been messed up and the unwinding fails, so that the exception aborts  
the process rather than being caught by a handler further down the  
stack.

That's really bad news when using distributed objects ... you might  
make a call to another application, an exception is raised in that  
application, passed back and re-raised in your application which then  
aborts your app!

So this would appear to be a rather severe drawback as invocations are  
used quite extensively.  Maybe the stack unwinding works if you used  
libffi for your invocations, but libffi doesn't work on some platforms  
(eg the debian 64bit intel box I develop on).