Re: Bogus error message? (covariant return types)

[Omar Loggiodice <ologgio at vrainn dot com>]

On Thu, 18 Dec 1997, Joe Buck wrote:

> 
> > Now to the potential bug. If I compile the following code with egcs-1.0
> > (plus H.J. patches for RedHat 5.0) I get an error message saying:
> > 
> > covariant-bug1.cc:26: sorry, not implemented:
> > adjusting pointers for covariant returns   
> 
> Sigh.  Any message beginning "sorry, not implemented" means that while you
> have written correct C++, the compiler cannot handle the code.  The
> compiler is even apologizing for you.  Given how polite the compiler is
> being to you, it seems rather impolite to complain about a "bogus error
> message".
> 

I --obviously-- understand this, that is why I said "potential" bug, and
it is also the reason why I put a question mark after "bogus error
message". I was not trying to be impolite by any measure, I actually was
trying to get more information (which you did supply afterwards, thanks
for that). I wasn't sure this was a bogus error message, again, that is
why I used a question mark in my subject. 

Since I am completely aware of the fact that egcs and previous versions of
gcc support covariant return types (in the single inhertance cases where
I've used it before) I was wondering wether the "sorry, not implemented"
error was some historic code from the times when covariant return types
where not properly supported. I had no idea of the fact that the
implementation for covariant return types is a partial implementation,
that is what led me to believe that *maybe* (hence the question mark and
the word 'potential') this was a bogus error message. Sorry if that meant
impoliteness to you, it was not my intention at all. 

> > However, if I change the order of the base classes the code compiles and
> > runs perfectly. 
> 
> That's because if you change the order of the baseclasses, the offset
> becomes zero and the compiler doesn't need to adjust the pointer returned
> by C::who.  That is, with
> 
> class C: public A, public B
> 
> Then we can convert a C* to an A* with no code, but converting a C* to
> a B* requires adding an offset to the address of the pointer.  This is
> what is not implemnted.
> 

Thanks, this is the information I was looking for. Appreciate it. 

[...]
> > Here is the workaround:
> 
> Sorry, that kind of workaround won't help people who cannot change their
> inheritance structure for other reasons.  But it will work for some
> people.

Agreed, for some projects it might not work at all. It does help many
others though. 

____________________________________________________________________
            /   __  __  __  - __  __ / - _  __  ologgio@vrainn.com
  Omar R.  /__ /_/ /_/ /_/ / /_/ /_/ / /_  /-_  CIS: 74040,1543
                  __/ __/                         
___C++/7_yrs____Virtual Reality/4_yrs____Vorl_____Linux(free)_______
Linux: A Unix OS for _you_ ...ftp://sunsite.unc.edu/pub/Linux