Re: open watcom compiling gcc on win32

["Edward S. Peschko" <esp5 at pge dot com>]

On Fri, Jan 30, 2004 at 04:49:03PM -0500, Ian Lance Taylor wrote:
> "Edward S. Peschko" <esp5@pge.com> writes:
> 
> > > BTW, FYI, Borland is not C++ ABI compatible with MSVC++.
> > 
> > reference? I see different.
> > 
> > http://shop.store.yahoo.com/1-software-group/borcbuil50en.html
>
> I don't see anything there which says that the C++ ABI is the same.
> But I really have no information on the topic.

It comes under the heading of "Microsoft Visual C++ project 6 support". They don't
specifically mention ABI compatibility, but they mention that you can import 
Microsoft visual C++ projects into Borland, and to link to VC++ dlls through an
import mechanism.

In any case, operationally you've been able to link Microsoft dll's to Borland
apps for years. Eg, the following discussion on google (1998):

http://makeashorterlink.com/?X1B225047

> 
> On the other hand, I've heard several times that Microsoft has
> patented their vtable layout.  That would make it impossible for gcc
> to emulate it, unless Microsoft gave explicit permission.
> 
> Ian

yeah. patents. you just have to look for chinks in their armor.

Right now, I'm looking at the patent 5,481,708, which is a 
'system and methods for optimizing object-oriented compilations', and
I assume is the patent that you are talking about.

http://makeashorterlink.com/?J28511047

The most pertinent claim is #1:

1. In a system for generating code executable by a computer from object-oriented
source listings, said source listings including definition of classes of objects 
having data members and functions operative on said data members, the computer
including a memory stack for creating automatic variables, a method for passing
of objects by value to a function, the method comprising:

    a) identifying by the system a class argument being passed by value as a 
       parameter to a particular function, said class argument comprising an 
       object to be instantiated from a particular class;

    b) creating a surrogate variable for the class argument by allocating 
       space on the memory stack for storing said value; and

    c) determining an address of said surrogate variable on the memory stack when code
       referencing said class argument is generated.

Hmm. My attention is immediatly drawn to the phrase:

'In a system for generating code executable by a computer from object-oriented
source listings'


well, what I'm not talking about a system for genrating code executable by a computer
from object oriented source listings, I'm talking about a system that *reads*
an executable format. No source or compilation involved.

In other words, from what I read it would be verboten for g++ to *produce* a 
format that is compatible with vc++. Which is fine - I really couldn't care less
if g++ produces dlls that vc++ can read - all I care is that g++ can read 
dlls that vc++ produces.

Quite ironic really - by patenting the process they insure that other compiler
vendors can be compatible *with* them, but that other compiler vendors can't 
produce compatible dlls; probably much more of a benefit to them than 
whatever small speed increase this optimisation gives them.

Anyways, that's the way I read it. Any other interpretations or patents?

Ed