Re: Common C++ ABI

Re: Common C++ ABI

[Bob Jamison]

Jim Wilson wrote:
----->

Some interoperability may be possible in some special circumstances.  We may
be close enough that it would be possible for one vendor to release a compiler
that is bug-compatible with another vendor.  The FSF is unlikely to do this,
but we might see commercial compilers for GNU/Linux that are bug-compatible
with gcc.

<----

It remains pure unobtainium!  ;-)


Maybe I was just thinking wishfully, but isn't this
the whole point to standardizing binary interfaces?
g++ has always worked with g++ of the same version.
Is cross-version compatibility the only benefit of the
new ABI?



By the way, 3.2 compiles on Irix6.5 and Solaris2.6 here
with tweaking,  and cleanly on RH Limbo.   Nice work,
guys!




Bob Jamison
LinCom Corp

Re: Common C++ ABI

[Jim Wilson]

>>Some interoperability may be possible in some special circumstances.
>It remains pure unobtainium!  ;-)

For now, yes, but a gcc release is just one step of a long process.  It is
possible that 6 months from now you will be able to purchase a proprietary
C++ compiler that is interoperable with a GNU/Linux distribution built using
gcc 3.2.  It is the possibility of this that has gotten people excited.  We
just need to remember that we haven't gotten there yet.

>Maybe I was just thinking wishfully, but isn't this
>the whole point to standardizing binary interfaces?

Yes, the intent is interoperability between compilers from different vendors.
However, not all of the work is done yet.  Much has been accomplished already
though, enough to make this newsworthy.

>Is cross-version compatibility the only benefit of the
>new ABI?

The fact that we have a published C++ ABI means that we can now tell when gcc
is right or wrong, and this should make it possible to have C++ compatibility
from one gcc release to the next.  This is as yet an unproven benefit.  There
is a reasonable chance that we can achieve it for the gcc 3.3 release though.

Jim

Re: Common C++ ABI

[Jim Wilson]

>Is there perhaps a list of third party C++ vendors which gcc 3.2's
>interoperability has been tested against?

There has been no serious interoperability testing as yet.

There has been some limited testing against the Intel C++ compiler.  We
have been finding C++ ABI bugs in gcc and/or the Intel C++ compiler roughly
once a week.  There is a good chance that we will find another bug after the
gcc 3.2 release.

> For example, what are my
>chances of getting sun cc and g++ to play nicely? 

I'd say there is only a small chance of getting this to work.  Most of the
work so far has concentrated on C++ front end issues, such as getting a
common vtable layout and common name mangling schemes.  Full interoperability
requires more than that though.  We need the GNU linker to correctly handle
Sun compiled libraries, and the Sun linker to correctly handle GNU compiled
libraries.  We need GNU compiled libraries to work if linked with Sun startup
files.  We need Sun compiled libraries to work if linked with GNU startup
files.  There are known problems in these areas that have not been addressed
yet.  Also, given the lack of interoperability testing so far, it is likely
that there are still bugs in the compilers and in the ABI itself that still
need to be found and fixed.  There is also the issue of whether Sun has
implemented the common C++ ABI yet.  I don't know if they have done so yet.

Some interoperability may be possible in some special circumstances.  We may
be close enough that it would be possible for one vendor to release a compiler
that is bug-compatible with another vendor.  The FSF is unlikely to do this,
but we might see commercial compilers for GNU/Linux that are bug-compatible
with gcc.

Jim

Common C++ ABI

[Trent WADDINGTON]

Is there perhaps a list of third party C++ vendors which gcc 3.2's
interoperability has been tested against?  For example, what are my
chances of getting sun cc and g++ to play nicely?  Maybe a table on the
web site is the way to go.

Trent Waddington