Re: Correct logic for "defined by object"

[Alan Modra <amodra at gmail dot com>]

On Tue, Dec 23, 2014 at 05:25:03AM -0800, H.J. Lu wrote:
> On Tue, Dec 23, 2014 at 5:10 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> > On Tue, Dec 23, 2014 at 5:06 AM, Alan Modra <amodra@gmail.com> wrote:
> >> On Tue, Dec 23, 2014 at 04:46:29AM -0800, H.J. Lu wrote:
> >>> No testcase?
> >>
> >> No.  This patch wasn't in response to a bug report or from observing
> >> wrong linker behaviour myself.  In fact, I don't think you can write a
> >> testcase that fails with the old code and passes with the new, due to
> >> the way def->by_object is used.  The patch was really just a tidy.
> >>
> >
> > So this patch won't change linker behavior in any way whatsoever.
> > Basically the setting of def->by_object is unused.
> >
> > --
> > H.J.
> 
> There are
> 
>          new_number (h != NULL
>                       && (h->type == bfd_link_hash_defined
>                           || h->type == bfd_link_hash_defweak
>                           || h->type == bfd_link_hash_common)
>                       && ((def = symbol_defined (tree->name.name)) == NULL
>                           || def->by_object
>                           || def->iteration == (lang_statement_iteration & 1)));
> 
> and
> 
>          if (!(h != NULL
>                 && (h->type == bfd_link_hash_defined
>                     || h->type == bfd_link_hash_defweak)
>                 && h->u.def.section == bfd_abs_section_ptr
>                 && (def = symbol_defined (tree->name.name)) != NULL
>                 && def->iteration == (lang_statement_iteration & 1)))
> 
> One checks by_object and the other doesn't.   Should they be
> consistent?  Can we find some real usage for by_object?  If not,
> why don't we remove it?

This is a real use of by_object.  The "new_number" expression is
saying the result of DEFINED is true iff the symbol is defined,
defweak or common, and there are no script assignments to the symbol,
or the symbol was actually defined in an object file, or the script
assignment occurred before the DEFINED keyword.

The second hunk of code you show doesn't need to test by_object since
it is dealing with self-assignment of symbols in a linker script.

-- 
Alan Modra
Australia Development Lab, IBM