Re: PATCH: undefined symbols in shared libraries

["H. J. Lu" <hjl at lucon dot org>]

On Fri, Mar 12, 2004 at 05:12:41PM +1030, Alan Modra wrote:
> On Thu, Mar 11, 2004 at 09:34:19PM -0800, H. J. Lu wrote:
> > unresolved_syms_in_shared_libs is for unresolved references in input
> > shared libraries. unresolved_syms_in_objects is for unresolved
> > references in input relocatable objects. When building shared libraries,
> > we allow unresolved references in input relocatable objects by default.
> > They are disallowed for executables. With "-z defs", we also disallow
> > unresolved references in input relocatable objects when building
> > shared library.
> 
> I agree with all of this.  It's exactly what I think ld should be doing
> instead of the current code.
> 
> What we currently have is messy code in lexsup.c that sets _both_
> unresolved_syms_in_shared_libs and unresolved_syms_in_objects, and then
> code in elf-bfd.h and elsewhere that tests one or other of the flags
> depending on info->shared or info->executable.  That's what I think is
> confusing.  The unresolved_syms_in_* are all about input files, not
> the particular output ld is producing, so there should be no need to
> test output related flags like info->shared.  Instead, use
> unresolved_syms_in_* along with information about what sort of input
> was unresolved.  That's where I was heading with the untested patch
> I wrote.
> 

This is the patch to clean up this mess. unresolved_syms_in_objects is
used on input relocatable files. unresolved_syms_in_shared_libs is only
used when there is an undefined reference from a shared library.


H.J.

Attachment: binutils-elf-defs-2.patch
Description: Text document