This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Fix Ada bootstrap on ppc-darwin & x86_64
On Tue, 2005-07-26 at 09:52 +0200, Jan Hubicka wrote:
> > On Mon, Jul 25, 2005 at 11:59:24PM +0200, Laurent GUERBY wrote:
> > > A slightly updated version per Andrew comments in bugzilla, test results
> > > show no regression amd64-linux (and on x86):
> > >
> > > http://gcc.gnu.org/ml/gcc-testresults/2005-07/msg01359.html
> > >
> > > OK to commit?
> > No, not ok.
> > > > This patch fixes the Ada bootstrap on ppc-darwin by forcing Ada not to
> > > > produce ADDR_EXPR<CONSTRUCTOR> in its trees. Note this really works
> > > > around a cgraph defect for non GENERIC trees but since everyone dealing
> > > > with trees have said that ADDR_EXPR<CONSTRUCTO> should not be used, I
> > > > decided this was the best way forward.
> > Depending on what cgraph is looking for, you can either hack around
> > this in LANG_HOOKS_CALLGRAPH_ANALYZE_EXPR, or we can bite the bullet
> > and change cgraph to lower to gimple before analysis.
> cgraph lowers to gimple before analysis. Problem is that gimplifier
> don't gimplify static var's constructors.
I don't know a thing about this area, but is the gimplifier extension
simple enough to do in stage3 context? Or is the patch using the right
approach at this stage?