Hi,
We've done some additional testing of the patch I sent earlier this week that handles the removal of the "dotted" symbols from elf64.  We tested opannotate on PPC64/RHEL 4, both with and without the patch.  Without the patch, opannotate was not producing annotated source for 64-bit programs.  With the patch applied, we were seeing a segmentation fault in BFD code, elf_find_function, called by _bfd_elf_find_nearest_line, called by op_bfd::get_linenr.  Investigation of this problem is indicating a probable bug in BFD, so our in-house BFD person is working on that. 
 
A workaround is to check the symbol prior to calling _bfd_elf_find_nearest_line;  if it's a "no-name" section symbol, then we skip the call to _bfd_elf_find_nearest_line.  This should be a valid workaround since there's no possibility of finding a line number in the source for a section symbol anyway. This fix is working fine in our testing on RHEL 4, so I've opened a bug to get the fix pushed into RHEL 4. 
 
However, as before, I'm sure the Red Hat folks will want the patch accepted upstream before they accept it.  The problem is that opannotate is broken in CVS right now.  I get a segmentation fault trying to use it as is on either Intel or PPC64 platforms.  The seg fault is coming from line 682 in op_bfd::has_debug_info().  In the initialization of the for-loop, 'ibfd', may be null.  If I put a check for null prior to that and return debug_info.reset(false), I get all the way through the program without apparent errors, but no annotated source is created.
 
I see that the changes involving debug info were just committed into CVS this week, so maybe I'm grabbing something out of CVS that's not quite completely baked yet.  Is there a snapshot or some CVS branch I can use to test my patch so I can submit it to the mailing list?
 
Thanks, and sorry for the long-winded note.
 
P.S.  By the way, I also tested opstack and opgprof on RHEL 4.  opstack appears to work fine, with or without the "dot" symbol patch applied.  opgrof is not functioning properly, as it turns out, because gprof itself is broken.  I opened a bug for that problem to have our in-house binutils person fix that.  This is a problem only for 64-bit programs on PPC64 (as far as I am aware) -- no doubt, another casualty of the removal of the "dot" symbols. 
 
Regards,
Maynard Johnson
IBM Linux Technology Center