Re: [RFA] Perform relocations in dwarf2.c

[Michal Ludvig <mludvig at suse dot cz>]

Daniel Berlin wrote:
> >  You shouldn't be casually deleting
> > code like this if you don't even know what platforms it is for.
> >
> > The list archives would be glad to tell you.
> He's also removing the checks for what the standard specifically says is 
> the right thing.
>
> If it doesn't work as is, something else is broken, as the code is correct.
It was related to these lines:

bfd/dwarf2.c (_bfd_dwarf2_find_nearest_line()):
   1924   if (addr_size == 0)
   1925     addr_size = 4;
   1926   BFD_ASSERT (addr_size == 4 || addr_size == 8);

From bfd/elf.c (ie. on all platforms except arm, alpha and mips) this 
function is called with addr_size==0, it's then reinitialised to 4 - 
even on 64b systems. It's a little confusing to have addr_size==4 on 
x86-64, so I've "fixed" itby initialising to 
get_elf_backend_data(abfd)->s->arch_size/8 instead of 4 in case it was 0 
before.

Then the rest ceased to work for me, because offset_size was initialized 
to addr_size, ie. 8 on 64b archs, but should have been 4 to let the 
.debug_info parser work.

Now I've reverted this change and will follow the rule: "don't fix it 
unless it's broken" :-))

Michal Ludvig
--
* SuSE CR, s.r.o     * mludvig@suse.cz
* (+420) 296.545.373 * http://www.suse.cz