[Dwarf-Discuss] Possible ambiguity with DW_CFA_remember_state/store_state

John DelSignore jdelsign at totalviewtech.com
Sat Jun 13 10:19:08 PDT 2009


Paul Pluzhnikov wrote:
>> On Fri, Jun 12, 2009 at 07:46:34PM -0400, John DelSignore wrote: FWIW,
>>> TotalView saves the CFA along with the explicitly-numbered registers. So,
>>> it seems that TotalView and GDB agree, and that GCC is the outlier. IMHO,
>>> it's two against one, so GCC should be fixed.
> 
>> If TotalView saves CFA along with explicitly-numbered registers on
>> DW_CFA_remember_state and restores CFA state in DW_CFA_restore_state, then
>> it matches what GCC expects and what GCC unwinder does, it is GDB that
>> wasn't saving/restoring CFA state (and GCC unwinder before 2006-10-26,
>> but GCC really started emitting these two unwind ops last month).
> 
> That's my understanding of John's message as well.

To set the records straight, I think Cary got it right, and I got it backwards. Anyway...

> John, could you run the test case from
>  http://sourceware.org/bugzilla/show_bug.cgi?id=10275
> under TotalView, and confirm that TV produces correct stack trace for it?

I'll be happy to run the test case, but I don't have a GCC 4.5 available. If you can email me the source and executable files, I'll give it a try.

Cheers, John D.




More information about the Dwarf-Discuss mailing list