Date: Tue, 30 Apr 2013 20:23:34 +0200 From: Dimitry Andric <dim@freebsd.org> To: Jan Beich <jbeich@tormail.org> Cc: Brandon Gooch <jamesbrandongooch@gmail.com>, kit <ktsin@acm.org>, Benjamin Kramer <benny.kra@gmail.com>, "freebsd-ports@freebsd.org ports" <freebsd-ports@freebsd.org>, =?iso-8859-1?Q?Rafael_Esp=EDndola?= <rafael.espindola@gmail.com> Subject: Re: firefox build broken under clang 3.3 Message-ID: <ED3C5033-F932-4F56-895F-906873667C64@freebsd.org> In-Reply-To: <1UXDVA-0001b7-7U@internal.tormail.org> References: <20130419020021.GA16918@test.yahoo.com> <51716917.90101@smeets.im> <F82EAEBA-C47D-4AC6-8FAE-AC3541B131C7@FreeBSD.org> <517187B9.40106@smeets.im> <CALBk6yLOV8fzdAY9y0_7SZvR9N8J-19FrwZjFv4t14ub8wmGiA@mail.gmail.com> <1UXDVA-0001b7-7U@internal.tormail.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Apr 30, 2013, at 18:34, Jan Beich <jbeich@tormail.org> wrote:
...
> The faulting function is lost within crash handler. If you build
> firefox with
>=20
> # use DEBUG_FLAGS or set STRIP to empty explicitly
> CFLAGS +=3D ${DEBUG_FLAGS}
> DEBUG_FLAGS +=3D -O0 -g
>=20
> jaeger jit crash would look like
>=20
> =
http://lists.freebsd.org/pipermail/freebsd-current/2013-April/041165.html
The firefox crash in that post was with -O2, so it is most likely not =
due to the vectorizer then.
In any case, I have not been able to reproduce that either. I have =
browsed stackoverflow.com for about 5 minutes, clicking a whole bunch of =
random articles and links, and no problems whatsoever occurred. That is =
with the default www/firefox port, which seems to be the released =
Firefox 20.0, at its default settings, on amd64.
> It doesn't happen on firefox23 with baseline jit[1] disabled via pref.
> A big change like zones (bug 759585) may have refactored code enough
> to not hit the clang bug. So, try either clang trunk or
>=20
> =
http://trillian.chruetertee.ch/freebsd-gecko/changeset/1256/trunk/www/fire=
fox/files/patch-clang33
>=20
> The latter pessimizes inlining for clang 3.2 as well.
>=20
> [1] baseline crashes in a different way
> https://bugzilla.mozilla.org/show_bug.cgi?id=3D860867
I do not like the workaround in that Mozilla bug, since it simply =
disables the vectorizer. I can understand that software must always =
ship tomorrow, but it would have been nicer to attempt to figure out =
what is incorrectly optimized, and why...
>> Rebuilding with debugging symbols provides no further insight, as =
that
>> seems to provide a work-around for whatever the root cause may be
>> (i.e. no more segfaults).
>=20
> DEBUG enables compile-time diagnostics and strips any -O* from CFLAGS.
Indeed. To work around this, use the following diff:
Index: Mk/bsd.port.mk
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
--- Mk/bsd.port.mk (revision 316903)
+++ Mk/bsd.port.mk (working copy)
@@ -1580,7 +1580,7 @@
.if defined(WITH_DEBUG) && !defined(WITHOUT_DEBUG)
STRIP_CMD=3D ${TRUE}
DEBUG_FLAGS?=3D -g
-CFLAGS:=3D ${CFLAGS:N-O*:N-fno-strict*} ${DEBUG_FLAGS}
+CFLAGS:=3D ${CFLAGS} ${DEBUG_FLAGS}
.endif
=20
.if defined(NOPORTDOCS)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?ED3C5033-F932-4F56-895F-906873667C64>