George Vasick <George.Vasick at sun.com> writes:
> Nicolas Williams wrote:
> > On Thu, Jan 08, 2009 at 08:24:13AM -0800, George Vasick wrote:
> >> There is testing and potentially support.
> >
> > Support means the same thing for GCJ and GNAT as for GCC: we'll keep the
> > thing up to date as new versions come out. Support need not mean "we'll
> > fix any bugs in GCJ and/or GNAT before the GCC community does."
> >
> > If it builds and passes its own tests (this shouldn't take long to
> > verify, and we have an attestation that it does) then including it costs
> > *nothing* on the margin.
>
> I guess I disagree with this. It costs nothing right up to the point
> where you are the one doing it and you run into a problem.
But this is true for the current GCC 3.4.3 as well (and to a lesser degree
for all non-Sun software in OpenSolaris), and also for the set of languages
already included. I think there can be several solutions here:
* Just report the problems to the GCC community and wait until a fix comes
up.
* Involve a third party specializing in GCC support, in the same way
CodeSourcery was tasked to complete the amd64 GCC 3.4.3 port.
* Become more active in the GCC community itself to a point where you can
fix everything in the compiler yourself.
> >> Also, we are adding Sparc support for the first time in this release.
> >> Sparc is more complicated than x86 since the stock GCC backend for Sparc
> >> is not especially competitive in performance compared to the Sparc
> >> Studio backend.
> >
> > So... what?
> >
> > Are you actually improving GCC performance on Solaris by modifying GCC?
>
> Yes, by hooking up the GCC frontends with the Studio Sparc backends.
> The plain GCC backend will be available under flag control.
I think there's something wrong here: you have stated several times that
you add SPARC support for the first time. This is simply not true: GCC
3.4.3 has been available on SPARC since Solaris 10. You just add the
gccfss backend and presumably make it the default. I'm not completely sure
this is wise, though: does it pass the GCC testsuite at least as well as
the default SPARC backend?
Rainer
--
-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University
