[RFA/PATCH] auto-import errors & DWARF 2 (was Re: _cooked_size usually set when/where?)

[Brian Ford]

Ok, I give up.  I can't find the "proper" fix for this.

2004-04-13  Brian Ford  <ford@vss.fsi.com>

        * simple.c (bfd_simple_get_relocated_section_content): Handle a
	_cooked_size of zero when auto-import errors look up DWARF 2 data.

Comments, alternatives, suggestions *please*?  Thanks.

On Wed, 31 Mar 2004, Brian Ford wrote:

> On Wed, 31 Mar 2004, Alan Modra wrote:
>
> > On Tue, Mar 30, 2004 at 04:36:42PM -0600, Brian Ford wrote:
> > > On Tue, 30 Mar 2004, Alan Modra wrote:
> > > > On Mon, Mar 29, 2004 at 04:33:37PM -0600, Brian Ford wrote:
> > > > > So, my question is, when is _cooked_size usually set up and by whom?
> > > >
> > > > ldlang.c:lang_size_sections_1, and perhaps earlier, eg.
> > > > ldemul_before_allocation.
> > >
> > > Thanks for the reply.  Those set the output section's _cooked_size right?
> >
> > No, they set input section _cooked_size too.
> >
> Ok, I was going by the comment above lang_size_sections_1:
>
> /* Set the sizes for all the output sections.  */
>
> and by the observation that in my previously reported problem, neither
> have been called before the abort.
>
> > As an aside, I think the current use of _raw_size and _cooked_size in
> > binutils is a bit of a mess.  My main dislike is numerous places that
> > use _cooked_size if non-zero and use _raw_size otherwise.  In the past
> > I've advocated using _raw_size and an adjustment, initially zero, in
> > all places that use _cooked_size.  You could do the same thing by
> > writing/reading _cooked_size in all places *except* for places that
> > really want the original on-disk size of input sections.  This is a
> > simple conceptual change, but changing bfd to use this scheme is tedious
> > and likely to introduce errors as there are many places that
> > (incorrectly) change _raw_size to something other than the original
> > section size.
>
> I whole-heartedly agree.  And, I was excited by Hans-Peter Nilsson's
> proposal to revamp this back in October.  I do understand the fragile
> nature of this sort of change, though.
>
> As an "outsider", I can tell you that this causes me no end of confusion
> when I try to properly fix bugs and add features.  In some ways, I guess
> this is how companies like Red Hat get business (ha, ha).  Well, that
> kind of confusion caused us to require their assistance recently at
> least ;-).
>
> However, given my current problem and lack of "big picture" understanding,
> the simplest way for me to fix it is to add just such a hack.  ie:
>
> reloc.c (bfd_generic_get_relocated_section_contents) line 4302:
>
> if (input_section->_cooked_size == 0)
>   input_section->_cooked_size = input_section->_raw_size;
>
> Please suggest an alternative.  I'm all ears.  But, so far, I've been
> unsuccessful in finding one, at least not by comparing PE/COFF to Linux
> ELF and trying to find differences :-).
>
> Thanks again.  I've been looking hard to find the "right" fix, but it is
> becoming prohibitively expensive for me to do so.

-- 
Brian Ford
Senior Realtime Software Engineer
VITAL - Visual Simulation Systems
FlightSafety International
Phone: 314-551-8460
Fax:   314-551-8444
-------------- next part --------------
Index: simple.c
===================================================================
RCS file: /cvs/src/src/bfd/simple.c,v
retrieving revision 1.14
diff -u -p -r1.14 simple.c
--- simple.c	30 Nov 2003 18:40:41 -0000	1.14
+++ simple.c	13 Apr 2004 22:19:36 -0000
@@ -142,6 +142,12 @@ bfd_simple_get_relocated_section_content
   void *saved_offsets;
   bfd_boolean saved_reloc_done = sec->reloc_done;
 
+  /* Foul hack to prevent bfd_section_size aborts.  The reloc_done flag
+     only controls that macro (and the related size macros), selecting
+     between _raw_size and _cooked_size.  We may be called with relocation
+     done or not, so we need to save the done-flag and mark the section as
+     not relocated.  */
+
 #undef RETURN
 #define RETURN(x)				\
   do						\
@@ -151,17 +157,20 @@ bfd_simple_get_relocated_section_content
     }						\
   while (0)
 
-  /* Foul hack to prevent bfd_section_size aborts.  The reloc_done flag
-     only controls that macro (and the related size macros), selecting
-     between _raw_size and _cooked_size.  We may be called with relocation
-     done or not, so we need to save the done-flag and mark the section as
-     not relocated.
-
-     Debug sections won't change size while we're only relocating.  There
-     may be trouble here someday if it tries to run relaxation
-     unexpectedly, so make sure.  */
-  BFD_ASSERT (sec->_raw_size == sec->_cooked_size);
   sec->reloc_done = 0;
+
+  /* Auto-import error messages trying to look up DWARF 2 data
+     call us before _cooked_size has been set. */
+  if (sec->_cooked_size == 0)
+    sec->_cooked_size = sec->_raw_size;
+  else
+    {
+      /* Debug sections won't change size while we're only relocating.  There
+         may be trouble here someday if it tries to run relaxation
+         unexpectedly, so make sure.  */
+
+      BFD_ASSERT (sec->_raw_size == sec->_cooked_size);
+    }
 
   if (! (sec->flags & SEC_RELOC))
     {