Alexey,
> On Fri, 21 Jul 2006, Rainer Orth wrote:
>
> > I think I asked this before when you announced GCCfss on the main gcc list:
> > how much work would it be to turn GCCfss into a generic (i.e. both sparc
> > and x86) GCC with the Sun backend? It seems that the backend interface is
> > identical between sparc and x86, so it may be not much more than a
> > recompile of GCCfss on x86. We should strive for feature parity between
> > sparc and x86 as much as possible.
>
> That's not what I meant.
> "For sparc and for x86 it can be made to be built from the same source tree.
> It will use Sun Studio backend on sparc and plain gcc backend on x86"
I know. OTOH, I think the claim has been made that Studio 11 produces
better code than GCC for a considerable range of applications. If so, it
would certainly be useful to be able to use the Studio backend on x86/amd64
as well as on sparc.
> So what exactly do you suggest?
> have gcc-4.1-csl branch with Solaris related changes?
> 4.2 is out of the question as I mentioned in the other thread.
> I'd much prefer to have 4.0.3 based one, since I don't see the value of 4.1
> for Solaris.
I think you have to be careful here: GCC is neither only a C compiler nor
only for compiling ON. While the differences between 4.0.x and 4.1.x may
not be that large for C, the story is different for other languages.
For languages other than C, you have to consider the stability of the
support shared libraries. E.g. in the C++ case, libstdc++ has been changed
incompatibly for every minor GCC release (i.e. 4.0, 4.1, ...) so far.
While there is work in progress to fix this, this is not in any GCC release
yet. So if you expose G++ users on Solaris to GCC 4.0 and later to say
4.3, they are guaranteed to have to recompile all their applications and
libraries. So it seems best to avoid too many transitions of that kind if
possible, and perhaps help the GCC community to get done with ABI stability
earlier. Another example is gfortran: it was introduced in 4.0 as a
replacement for g77, but the 4.0 version is considered buggy beyond
repair. Even between 4.1 and 4.2, they had to change the runtime library
incompatibly. Since at least in the SFW GCC 3.4, g77 is include, don't
expose users to too much change here. Especially in the gfortran case,
there has been many improvements since 4.0, which make a later release (4.1
or even 4.2) very worthwhile for users.
That's the main reason for my suggestion to target the latest stable GCC
release at the time of Solaris 11 freeze for inclusion there, not something
as old as say GCC 4.0 (and probably not even 4.1 or 4.2, depending on
schedules).
> > I consider it quite important for Sun/the OpenSolaris community to be seen
> > as an integral part of the GCC community: working there as much as possible
> > gives us much more credibility than doing things in our own private hole.
> > Besides, all efforts there would be for the benefits of both vanilla GCC
> > and it's users on Solaris as well as the OpenSolaris community.
>
> Agree.
> I guess first try to convince gcc developers to test new features on ON.
I think here's an important misunderstanding: this is not how free software
works. If Sun as a vendor or the OpenSolaris community as a whole rely on
GCC in some way (as Sun has done for the initial amd64 port), it is their
responsibility to test GCC on their platform, make sure that it continues
to work seamlessly, and make sure that new features requiring
platform-specific code are supported on Solaris as well. Nobody else will
do this for us. This is how other vendors like IBM, HP, Apple and the
Linux distributors handle this: many of them invest considerable time and
man power to GCC development.
That Sun is different here sheds a bad light: so far, there haven't been
many Sun contributions to GCC, and stuff like GCCfss (which is useful for
Solaris users, but for nobody else, like GCC users on non-Solaris SPARC
systems) is seen by some as a rip-off of GCC: use the front ends so
software depending on GNU features compiles without problems, but don't
contribute back anything which benefits the GCC community as a whole. I
think this attitude has to change to give Sun credibility in this
community.
Rainer
-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University
_______________________________________________
opensolaris-discuss mailing list
[email protected]
