This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Patches to C++ front end have caused numerous regressions
- From: Mark Mitchell <mark at codesourcery dot com>
- To: David Edelsohn <dje at watson dot ibm dot com>
- Cc: jh at suse dot cz, rth at redhat dot com, gcc at gcc dot gnu dot org
- Date: 12 Sep 2003 14:09:36 -0700
- Subject: Re: Patches to C++ front end have caused numerous regressions
- References: <200309122047.h8CKlmg08848@makai.watson.ibm.com>
On Fri, 2003-09-12 at 13:47, David Edelsohn wrote:
> >>>>> Mark Mitchell writes:
>
> Mark> The problem seem to include missing constructors and destructors, and
> Mark> internal errors in cgraph_finalize_function.
>
> Mark> I will revert the patches causing these problems under the 48-hour
> Mark> rule if the problems are not corrected by then.
>
> What patches are you proposing to revert? I think that we need to
> push forward and get this feature working when Honza gets back from his
> trip. I would much rather see the backlog of patches reviewed than
> causing even more instability by reverting patches.
I will revert all of the recent patches that have caused regressions,
until we get back to a pre-regression state. The obvious candidates
are:
2003-09-11 David Edelsohn <edelsohn@gnu.org>
* decl2.c (var_finalized_p): Swap arms of conditional.
2003-09-10 Jan Hubicka <jh@suse.cz>
* decl2.c (var_finalized_p): New.
(maybe_emit_vtables, write_out_vars, finish_file): Use it.
If Honza is on vacation, then this is an additional argument for
reverting these patches.
I will try to review what I can, but I have limited time. Any patches
which will fix these problems should be tested on i686-pc-linux-gnu and
on AIX, if at all possible, since the linkage models on those platforms
are different, and we seem to be pushing problems back and forth between
those two platforms. If there is a patch which claims to fix these
problems, please forward it to me for review; I do not have time to go
through all of the gcc-patches mail looking for patches for this issue
If necessary, I'll revert necessary pieces of these patches as well, and
perhaps patches in the top level. I intend to go through them
one-by-one to see what needs to be reverted.
2003-09-11 Richard Henderson <rth@redhat.com>
* semantics.c (expand_or_defer_fn): Update for new
cgraph_finalize_function argument.
2003-09-10 Richard Henderson <rth@redhat.com>
* decl2.c (cxx_callgraph_analyze_expr): Mark argument unused.
-- 2003-09-10 Richard Henderson <rth@redhat.com>
* decl2.c (cxx_callgraph_analyze_expr): New, from corpse of
mark_member_pointers.
(lower_function): Remove.
* cp-tree.h: Update to match.
* cp-lang.c (LANG_HOOKS_CALLGRAPH_ANALYZE_EXPR): New.
(LANG_HOOKS_CALLGRAPH_LOWER_FUNCTION): Remove.
2003-09-09 Richard Henderson <rth@redhat.com>
* semantics.c (expand_or_defer_fn): Update call to
cgraph_finalize_function.
* semantics.c (expand_or_defer_fn): Use cgraph_finalize_function
always.
* decl2.c (finish_file): Avoid out-of-bounds array reference
during memmove.
2003-09-09 Richard Henderson <rth@redhat.com>
* decl2.c (mark_member_pointers): Rename from
mark_member_pointers_and_eh_handlers and don't check eh handlers.
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com