Re: [GOLD] assertion on more than one PT_NOTE

[Alan Modra <amodra at gmail dot com>]

On Mon, Apr 15, 2013 at 11:20:56PM -0700, Ian Lance Taylor wrote:
> On Sun, Apr 14, 2013 at 11:30 PM, Alan Modra <amodra@gmail.com> wrote:
> > I see no reason why an executable should be limited to one PT_NOTE
> > header.  This patch removes that limitation in gold.
> >
> > As far as I can see there is no code in gold that needs segments to be
> > unique like this.  OK to apply?
> >
> >         PR gold/15355
> >         * layout.cc (Layout::segment_precedes): Allow more than one
> >         segment with the same type and flags.
> 
> This should only happen when using a linker script.  Otherwise
> something is wrong.  So I would rather see something like this:
>       gold_assert(flags1 != flags2 ||
> this->script_options_->saw_phdrs_clause());
> 
> That patch is OK if it works.

Committed.

Index: gold/layout.cc
===================================================================
RCS file: /cvs/src/src/gold/layout.cc,v
retrieving revision 1.249
diff -u -p -r1.249 layout.cc
--- gold/layout.cc	15 Apr 2013 16:40:59 -0000	1.249
+++ gold/layout.cc	21 Apr 2013 23:17:17 -0000
@@ -3175,12 +3175,14 @@ Layout::segment_precedes(const Output_se
 
   // The order of non-PT_LOAD segments is unimportant.  We simply sort
   // by the numeric segment type and flags values.  There should not
-  // be more than one segment with the same type and flags.
+  // be more than one segment with the same type and flags, except
+  // when a linker script specifies such.
   if (type1 != elfcpp::PT_LOAD)
     {
       if (type1 != type2)
 	return type1 < type2;
-      gold_assert(flags1 != flags2);
+      gold_assert(flags1 != flags2
+		  || this->script_options_->saw_phdrs_clause());
       return flags1 < flags2;
     }
 

-- 
Alan Modra
Australia Development Lab, IBM