[RS6000] GNU strip fails to strip "gdb" (Was: Re: [RS6000] GNU strip doesn't remove ".debug" section)

[Christian LESTRADE]

This fix doesn't make problems with page alignment on RS6000 architecture.

I think the comment applies on other architecture.

In fact, page alignment on AIX is somewhat weird (see xcoff*.c and *rs6000*).

There is remaining a bug when GNU stripping GNU gdb 5.2 compiled with GCC 3.1.
The stripped executable won't load (with and without this patch).

The "objcopy" utility, when copying "gdb" makes a new "gdb" double in size 
and with corrupt headers.

I don't know why this problem has only (until now) occurred with "gdb" and 
not other programs.

At 11:16 02/07/02 +0930, you wrote:
> On Fri, Jun 28, 2002 at 10:43:06AM +0200, address@hidden wrote:
> > On IBM AIX 5.1 (RS6000) platform, the GNU "strip" program doesn't 
> remove ".debug" sections (unless when using --remove-section=".debug").
> >
> > I suggest a fix like the following:
> > --- binutils-2.12.1/bfd/coffcode.h~   Thu May  9 16:48:49 2002
> > +++ binutils-2.12.1/bfd/coffcode.h    Fri Jun 28 10:19:12 2002
> > @@ -657,7 +657,7 @@
> >  #endif
> >          || strncmp (name, ".stab", 5) == 0)
> >      {
> > -#ifdef COFF_PAGE_SIZE
> > +#if defined (COFF_PAGE_SIZE) || defined (RS6000COFF_C)
> >        sec_flags |= SEC_DEBUGGING;
> >  #endif
> >      }
>
> I'm no expert with the coff linker code, but did you see the following
> comment in other places where SEC_DEBUGGING is set?
>
>       /* We mark these as SEC_DEBUGGING, but only if COFF_PAGE_SIZE is
>         defined.  coff_compute_section_file_positions uses
>         COFF_PAGE_SIZE to ensure that the low order bits of the
>         section VMA and the file offset match.  If we don't know
>         COFF_PAGE_SIZE, we can't ensure the correct correspondence,
>         and demand page loading of the file will fail.  */
>
> --
> Alan Modra
> IBM OzLabs - Linux Technology Centre