Re: Bad patch for gas

[amodra at one dot net dot au]

On Wed, May 23, 2001 at 09:43:47PM -0700, H . J . Lu wrote:
> 
> Under mips-elf, I now got
> 
> # cat foo.s
> 	.set	mips16
> 
> 	.word	0
> 	la	$2,data2
> data2:
> 	.word	0
> # ./as-new -o /tmp/foo.o /tmp/foo.s
> /tmp/foo.s: Assembler messages:
> /tmp/foo.s:4: Error: invalid unextended operand value

Here's the code, from tc-mips.c:md_convert_frag, that is broken by
my change:

      resolve_symbol_value (fragp->fr_symbol);
      val = S_GET_VALUE (fragp->fr_symbol);        (*)

Previously, resolve_symbol_value was actually setting the value of
the symbol to it's final value, ie. including the base address of
the symbol frag as well as the offset.  Now, with md_convert_frag
being called with finalize_syms == 0, resolve_symbol_value doesn't
do much, and S_GET_VALUE just returns the offset of the symbol
within the frag.  Oops.

Fortunately, there is an easy fix.

	* write.c (write_object_file): Set finalize_syms = 1 before
	size_seg is called.


(*) val = resolve_symbol_value (fragp->fr_symbol);
would have done the job here too.

-- 
Alan Modra


Index: gas/write.c
===================================================================
RCS file: /cvs/src/src/gas/write.c,v
retrieving revision 1.33
diff -u -p -r1.33 write.c
--- write.c	2001/05/22 10:23:48	1.33
+++ write.c	2001/05/24 07:51:42
@@ -1574,13 +1574,14 @@ write_object_file ()
       if (!changed)
 	break;
     }
+  /* Relaxation has completed.  Freeze all syms.  */
+  finalize_syms = 1;
+
   bfd_map_over_sections (stdoutput, size_seg, (char *) 0);
 #else
   relax_and_size_all_segments ();
-#endif /* BFD_ASSEMBLER  */
-
-  /* Relaxation has completed.  Freeze all syms.  */
   finalize_syms = 1;
+#endif /* BFD_ASSEMBLER  */
 
 #if defined (BFD_ASSEMBLER) && defined (OBJ_COFF) && defined (TE_GO32)
   /* Now that the segments have their final sizes, run through the