Re: Bug#492488: Reproduction of the Iceweasel crashes.
On Mon, Nov 03, 2008 at 07:48:24PM +1300, Michael Cree wrote:
> Mike Hommey wrote:
> >On Sun, Oct 26, 2008 at 08:09:29AM +0100, Mike Hommey wrote:
> >>Note that we *do* use --no-relax on xulrunner-1.9, but in the CFLAGS.
> >>Without it, libxul.so won't build at all ; the same applies to webkit,
> >>BTW.
>
> Interesting. I thought --no-relax is a linker, not a compiler, option.
It *is* a linker option, but you can pass it in CFLAGS as
"-Wl,--no-relax", where the "-Wl" part marks what follows as an argument
for "ld". Note that if a particular option isn't recognized by the
compiler, it's automatically passed to "ld". The "-Wl" simply makes
that behavior explicit (just in case a particular option might be
understood by the compiler and therefore not passed to the linker).
> >I could reproduce the crash on alpha and spotted it to be caused by
> >apparently bad g++ optimization on alpha. Adding:
> >MOZ_OPTIMIZE_FLAGS := -O0
> >in xulrunner's toolkit/components/url-classifier/src/Makefile.in makes
> >iceweasel work properly on albeniz.debian.org.
>
> Yes, Bob (compiling firefox) and I (compiling icecat) found that turning
> off optimisation (for the complete build) resulted in an executable that
> works. I wonder if -O1 might work okay?
Depending on what kinds of optimizations are performed at a given level,
it's possible that -O1 would work. Eventually, someone will probably
try it to narrow down where to look for the broken code. Regardless of
the outcome, it's probably a useful exercise. However, tracking down
optimizer bugs is not my idea of fun...
--
------------------------------------------------------------------------
Bob Tracy | "I was a beta tester for dirt. They never did
rct@frus.com | get all the bugs out." - Steve McGrew on /.
------------------------------------------------------------------------
Reply to: