Patch: Zero out the allocated dynamic content space.
Alan Modra
alan@linuxcare.com.au
Fri Aug 18 20:13:00 GMT 2000
On Fri, 18 Aug 2000, H . J . Lu wrote:
> On Fri, Aug 18, 2000 at 12:13:40PM -0700, Richard Henderson wrote:
> > Actually, it isn't that bad. Both the alpha and ia64 ports do this
> > correctly: record how many relocations of what class against which
> > symbols, and then compute the size only in size_dynamic_sections.
> > Which happens after all the symbol visibility changes have happened.
>
> It doesn't mean anything to me since both alpha and ia64 ports use
> bfd_zalloc in that particular place. Just search for
Belts and braces, eh? Might be interesting to see what the ia64 port does
if the bfd_zalloc is changed to bfd_alloc, memset(..,-1,..) :-) Do you
ever get R_IA64_NONE relocs?
To change x86, m68k, ppc, etc. to use the ia64 scheme would require quite
a bit of work, and I wonder how much extra time is taken by all those hash
table traversals. 'tis a cleaner design though.
Meanwhile, we have a problem, and it looks like no one has time to fix it
properly. Why not apply H.J.'s one character fix, and maybe put a FIXME
in adjust_dynamic_symbols? It won't be the first FIXME in the source,
after all.
Regards, Alan Modra
--
Linuxcare. Support for the Revolution.
More information about the Binutils
mailing list