RFC: Change of dw2 DWARF_OFFSET_SIZE on 64-bit to POINTER_SIZE

[Joseph S. Myers]

On Tue, 26 Aug 2008, NightStrike wrote:

> On 8/1/08, Kai Tietz <Kai.Tietz@onevision.com> wrote:
> > Hello,
> >
> > The default value of DWARF_OFFSET_SIZE of 4 is for w64 a problem, because
> > on w64 there are only 64-bit imaged based relocations present and the
> > offset is build of this offset is computed by a secrel offset.
> > Additionally there is no way for this target to set VA to zero as done for
> > elf. I prepared a patch for binutils to change this behaviour. As we
> > discussed is the additional memory consume for 64-bit less, so I want to
> > query if there is general agreement in changing this value in general for
> > 64-bit targets?
> > These should affect DW_AT_stmt_list and the arange to use 8 byte offset
> > for 64-bit targets.
> >
> > The change would be in dwarf2out.c and looks like:
> > ....
> >  /* The size in bytes of a DWARF field indicating an offset or length
> >    relative to a debug info section, specified to be 4 bytes in the
> >    DWARF-2 specification.  The SGI/MIPS ABI defines it to be the same
> >    as PTR_SIZE.  */
> >
> >  #ifndef DWARF_OFFSET_SIZE
> > -#define DWARF_OFFSET_SIZE 4
> > +#define DWARF_OFFSET_SIZE (TARGET_64BIT ? 8 : 4)
> >  #endif...
> >
> > As I discussed with binutils, they would be happy to change this, if gcc
> > would change its behaviour, too.
> 
> I haven't seen a response on this yet.  Have I missed it?

Such a patch is obviously wrong; TARGET_64BIT is a target-specific macro 
used in some particular targets and cannot be used in generic code.  In 
general 64-bit offsets are only needed if a single object has more than 
4GB of debug info.  So such a patch is a pessimization for almost all 
normal code.  Please see the text from the DWARF standard I quoted when 
stopping MIPS64 GNU/Linux from using 64-bit offsets 
<http://gcc.gnu.org/ml/gcc-patches/2006-05/msg00142.html>.

-- 
Joseph S. Myers
joseph@codesourcery.com