This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
Re: I believe the current binutils is broken
On Sun, Nov 24, 2002 at 12:57:42PM +0100, Jakub Jelinek wrote:
> On Sat, Nov 23, 2002 at 03:41:30PM -0800, H. J. Lu wrote:
> > Hi Alan,
> >
> > I have many extra gcc 3.3 (cvs) libjava test failures on Linux/x86
> > with the current binutils. It looks like java exception doesn't work
> > anymore. After reverting back to 20021012, everyting is fine. I don't
> > know exactly which changes are the causes. But your eh frame change
> > looks very suspicious. Could you please take a look?
OK, I get the libjava failures too. I'll see what needs to be done to
fix them.
> Most of the elf-eh-frame.c code relies on info->relocateable being false,
Well, yeah, it was written without considering that some idiot would
try to turn it on for ET_REL. ;-)
> for ET_RET I really think it would be way better to just mark the FDEs for
> removal during final link. Flag of FDE to be removed could be
> e.g. changing the FDE into CIE (my preference - it will be wiped out as
> unreferenced CIE and even if not removed will never be put
> into any FDE tables) or by setting FDE address range to 0 and special casing
> this as removed FDE (FDEs with address range 0 don't apply
> to any instructions at all, so aren't containing any useful information).
> The former is harder at ld -r time, since it needs changing of relocations
> against that FDE into R_*_NONE.
We need to do something about relocs anyway. eg. The hack I installed
http://sources.redhat.com/ml/bug-binutils/2002-q3/msg00520.html, since
removed due to turning on the eh_frame edit on ld -r.
--
Alan Modra
IBM OzLabs - Linux Technology Centre