RFC: Improve error message for unresolved symbols because of --no-add-needed

[Nick Clifton]

Hi Guys,

  As a follow up to my previous post about changing --add-needed to
  -copy-dt-needed-entries, here is a second patch that addresses
  another problem with the --no-add-needed option.

  The problem in this case is that if an unresolved symbol is defined
  in a dynamic library which would have been included via following a
  DT_NEEDED link, but which is denied because the user has specified
  --no-add-needed on the command line, then a rather unhelpful error
  message is generated:

    % gcc -Wl,--no-add-needed -o foo1 foo1.o foo2.so -Wl,--rpath-link=.      
    ld: ./foo3.so: invalid DSO for symbol `foo' definition

  Note in this case foo3.so is the dynamic library defining 'foo' and
  that it would have been included via a DT_NEEDED entry in foo2.so.

  The attached patch changes the error message to this:
 
    ld: foo1.o: undefined reference to symbol 'foo'
    ld: note: 'foo' is defined in DSO ./foo3.so so try adding it to the linker command line

  Which identifies the object containing the unresolved reference as
  well as providing the user with a hint as to how they can resolve
  the problem.

  What do you think ?  Any objections to this change ?

Cheers
  Nick

bfd/ChangeLog
2009-10-29  Nick Clifton  <nickc@...>

        * elflink.c (elf_link_add_object_symbols): Improve error
        message generated when a symbol is left unresolved because a
        --no-add-needed command line option has prevented the
        inclusion of the DSO defining it.

Index: bfd/elflink.c
===================================================================
RCS file: /cvs/src/src/bfd/elflink.c,v
retrieving revision 1.357
diff -c -3 -p -r1.357 elflink.c
*** bfd/elflink.c 13 Oct 2009 04:06:20 -0000 1.357
--- bfd/elflink.c 29 Oct 2009 12:17:48 -0000
*************** error_free_dyn:
*** 3877,3882 ****
--- 3877,3883 ----
        bfd_boolean common;
        unsigned int old_alignment;
        bfd *old_bfd;
+       bfd * undef_bfd = NULL;
 
        override = FALSE;
 
*************** error_free_dyn:
*** 4108,4113 ****
--- 4109,4128 ----
       name = newname;
     }
 
+  /* If this is a definition of a previously undefined symbol
+     make a note of the bfd that contained the reference in
+     case we need to refer to it later on in error messages.  */
+  if (! bfd_is_und_section (sec))
+    {
+      h = elf_link_hash_lookup (elf_hash_table (info), name, FALSE, FALSE, FALSE);
+
+      if (h != NULL
+  && (h->root.type == bfd_link_hash_undefined
+      || h->root.type == bfd_link_hash_undefweak)
+  && h->root.u.undef.abfd)
+ undef_bfd = h->root.u.undef.abfd;
+    }
+
   if (!_bfd_elf_merge_symbol (abfd, info, name, isym, &sec,
       &value, &old_alignment,
       sym_hash, &skip, &override,
*************** error_free_dyn:
*** 4448,4456 ****
       if ((elf_dyn_lib_class (abfd) & DYN_NO_NEEDED) != 0)
  {
   (*_bfd_error_handler)
!    (_("%B: invalid DSO for symbol `%s' definition"),
      abfd, name);
!  bfd_set_error (bfd_error_bad_value);
   goto error_free_vers;
  }
 
--- 4463,4474 ----
       if ((elf_dyn_lib_class (abfd) & DYN_NO_NEEDED) != 0)
  {
   (*_bfd_error_handler)
!    (_("%B: undefined reference to symbol '%s'"),
!     undef_bfd == NULL ? info->output_bfd : undef_bfd, name);
!  (*_bfd_error_handler)
!    (_("note: '%s' is defined in DSO %B so try adding it to the linker command line"),
      abfd, name);
!  bfd_set_error (bfd_error_invalid_operation);
   goto error_free_vers;
  }
 

[Nick Clifton]

Hi Guys,

> bfd/ChangeLog
> 2009-10-29  Nick Clifton  <nickc@...>
>
> * elflink.c (elf_link_add_object_symbols): Improve error
> message generated when a symbol is left unresolved because a
>         --no-add-needed command line option has prevented the
> inclusion of the DSO defining it.

Well I had no objections to this, so I have gone ahead and checked it in.

Cheers
   Nick