Martin Bochnig <[EMAIL PROTECTED]> wrote:

> And for the inventor of C++: I have no clue what definition of an ABI he
> has, if he really thinks, he has to declare g++3.x's ABI (which is much
> more universally designed, not at all limited to g++) void. What is it
> then, if not an ABI?

The URL with the ABI proposal still does not include templates.

Sun CC and g++ are not even compatible at source level regarding 
templates. I am not sure if this is a lack of sufficient standardizition
or whether one of both does not follow the standard.


> Isn't it partially based/influenced by the SYSVR4 C_ABI?
> Ahh, well, such a thing doesn't exist either, according to what he had
> stated. Mhhh.
> Among others his statement concerning the "lack of a C_ABI" confuses me.
> And it indicates, that there might have been some other problem, when
> his interview had been written down. _Respect_ respect otherwise for his
> work, but excerpts from that interview make me wonder a bit.

He seems to have problems with the fact that the C ABI was agreed on
while there is no suitable way to agree on a C++ ABI.

While it looks simple to agree on a common way of mangling, it is not
trivial to agree on templates. You would need to replace majorparts of the
compiler in order to use a different template ABI. This may even affect the
optimizer and void millions of Euro/$ spend on the optimizer.....nobody
seems to be interested in this. 

As replacing the calling conventions in order to support a different platform
is simple and only related to the CPU, the related code in the compiler is
inside platform dependent parts. Portable compilers would need to replace
their internal architecture depending on the target architecture.

If you like to use templates, you would need to agree on a world wide
universal standard. If you can't, you would need to remove templates from the
C++ specs.

I don't see this to happen anytime soon.

P.S. let me comment another claim from Stroustrup: "C++ programs are not 
resulting in bigger binaris".

Let us have a look at GNU CSSC (a g++ reimplementation of SCCS).

-       CSSC is less than 70% of the amount of sourcecode compared to
        SCCS.

-       CSSC is resulting in 6x times bigger binaries than SCCS.

Jörg

-- 
 EMail:[EMAIL PROTECTED] (home) Jörg Schilling D-13353 Berlin
       [EMAIL PROTECTED]                (uni)  
       [EMAIL PROTECTED]     (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily
_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to