Linux/ppc failure

[Alan Modra]

On Thu, Oct 03, 2002 at 08:27:24AM +0200, Jakub Jelinek wrote:
> But I don't think the .rela.debug_info section can be SHF_ALLOC, thus
> gdb will have to duplicate symbol resolution from ld.so anyway.

Yes.

> And when it does, I think it is preferable to always use debug info
> from the module defining the symbol rather than the one refering to it.

dynbss means the module defining the sym needs the reloc.

> Plus, on REL arches this is incompatible change, since
> previously .debug_info section contained address of some variable/function
> in the current module while with .rel.debug_info it will have to contain
> addend (ie. usually 0).

This will only affect consumers of the debug info, ie. gdb, won't it?

The only reservation I have about rth's argument that emitting the
reloc is correct, is that in practice gdb may not need the relocations
to arrive at a correct result.  For instance, in order to apply
relocations gdb needs to provide a correct symbol table, which
implies proper symbol name resolution.  But the example I gave of
the debug info for a variable location also has the name in
DW_AT_name.  So why should gdb use DW_AT_location?

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre