[RFC] Providing init_fini_syms earlier?

[H. J. Lu]

On Tue, Jul 05, 2005 at 09:24:03AM -0700, H. J. Lu wrote:
> On Tue, Jul 05, 2005 at 10:13:43AM -0400, Carlos O'Donell wrote:
> > 
> > binutils,
> > 
> > If I provide the __init_array_start, __init_array_end,
> > __fini_array_start and __fini_array_end symbols earlier, in particular
> > before sizing of the dynamic sections, the code has a chance at removing
> > a possibly hidden local symbol.
> > 
> > A quick test with the following patch is successfull at removing the
> > extra R_PARISC_NONE relocs which are allocated to these symbosls but
> > never used. They are allocated because the provides comes after the
> > sizing and the symbol is assumed to be undefined dynamic. Once the
> > sizing is done the symbol is added and it's too late to remove those
> > entries.
> > 
> > I don't believe this patch is correct, a gut feeling says these
> > symbols should be provided at the last possible instance.
> > 
> > Comments and insight appreciated?
> > 
> > This patch *does* fix the linker, and R_PARISC_NONE extra relocs are no
> > longer emitted. I don't suggest applying this patch, it's just to start
> > a discussion.
> > 
> 
> Can you try
> 
> http://sourceware.org/ml/binutils/2005-05/msg00767.html
> 
> You may have to adjust it a little bit by hand.

Please also apply

http://sourceware.org/ml/binutils/2005-05/msg00726.html


H.J.