Re: Should the section address be ignored when processing DWARF information ?

[Nick Clifton <nickc at redhat dot com>]

Hi Eric, Hi Corinna,

> >   Does Mach-O need to have the section address added in to the offset
> >   calculation when processing DWARF information ?
>
> Right now dwarf information is only used in object files which are a 
> single "section", that then the actual sections are offsets within that 
> so unfortunately, yes, it needs the section address added into the 
> offset calculation when processing dwarf in object files.

Darn.

> Out of 
> curiosity what is the section offset in the coff/pe files? I would think 
> it would be something similar...

Corinna can probably answer this better than I, but I believe that the 
dwarf sections in PE files are given addresses after the end of all the 
other data.  (See ld/scripttempl/pe.sc).  They do not have the Mach-O 
problem however, because the offsets are all section relative to the 
start of the real debug sections, not Mach-O's mega-section.


So - Eric, Corinna - please could you try out the attached, alternative 
patch ?  It changes the value computed for the "address" field of the 
dwarf_section structure to be 0 for all targets, except the Mach-O. 
(Well at least that is its intention...)  So that in theory the current 
code in dwarf.c will continue to add in the section address for Mach-O, 
but no longer use it for Cygwin.

Cheers
  Nick


Index: binutils/objdump.c
===================================================================
RCS file: /cvs/src/src/binutils/objdump.c,v
retrieving revision 1.124
diff -c -3 -p -r1.124 objdump.c
*** binutils/objdump.c	19 Apr 2007 10:43:47 -0000	1.124
--- binutils/objdump.c	20 Apr 2007 12:41:53 -0000
*************** load_debug_section (enum dwarf_section_d
*** 1973,1979 ****
    if (sec == NULL)
      return 0;
  
!   section->address = bfd_get_section_vma (abfd, sec);
    section->size = bfd_get_section_size (sec);
    section->start = xmalloc (section->size);
  
--- 1973,1991 ----
    if (sec == NULL)
      return 0;
  
!   /* Compute a bias to be added to offsets found within the DWARF debug
!      information.  These offsets are meant to be relative to the start of
!      the dwarf section, and hence the bias should be 0.  For MACH-O however
!      a dwarf section is really just a region of a much larger section and so
!      the bias is the address of the start of that area within the larger
!      section.  This test is important for PE and COFF based targets which
!      use DWARF debug information, since unlike ELF, they do not allow the
!      dwarf sections to be placed at address 0.  */
!   if (bfd_get_flavour (abfd) == bfd_target_mach_o_flavour)
!     section->address = bfd_get_section_vma (abfd, sec);
!   else
!     section->address = 0;
!     
    section->size = bfd_get_section_size (sec);
    section->start = xmalloc (section->size);