[PATCH] RTEMS: Use strict DWARF-2 on ARM, PowerPC, SPARC

[Sebastian Huber]

On 06/19/2013 02:51 PM, Gedare Bloom wrote:
> On Wed, Jun 19, 2013 at 3:22 AM, Sebastian Huber
> <sebastian.huber at embedded-brains.de>  wrote:
>> >On 06/19/2013 01:54 AM, Chris Johns wrote:
>>> >>
>>> >>Sebastian Huber wrote:
>>>> >>>
>>>> >>>Some debuggers do not cope with the new DWARF3/4 debug format introduced
>>>> >>>with GCC 4.8.  Default to strict DWARF-2 on ARM, PowerPC and SPARC for
>>>> >>>now.
>>>> >>>
>>>> >>>This patch should be committed to GCC 4.8 and 4.9.
>>>> >>>
>>> >>
>>> >>I am not convinced about this change on technical grounds. When I say I am
>>> >>not
>>> >>convinced, I am not sure what we gain and what we give up and I would like
>>> >>to
>>> >>understand that a little better before agreeing to it.
>>> >>
>>> >>I should also point out I am using ARM with gcc-4.8.1 and gdb-7.6 and it
>>> >>is
>>> >>working well (my OpenOCD changes need more work) and any change to DWARF2
>>> >>that
>>> >>alters this would be a regression.
>> >
>> >
>> >The recent ARM GDB seems to have no problem with the new debug format.  I
>> >didn't see a differences in the debug experience between an old (e.g. 4.6.4)
>> >and a newer (e.g. 4.8.1) GCC.  I didn't debug C++.
>> >
>> >
>>> >>I have taken a look at the differences between DWARF2, DWARF3 and DWARF4.
>>> >>There
>>> >>is better language support in the later versions and debug data
>>> >>compression.
>>> >>These improvements are nice. What I am not sure about is the way limiting
>>> >>gcc
>>> >>to DWARF2 effects the debugging experience. If the flag is just a format
>>> >>change
>>> >>and the experience is the same that is ok, if however the C++ or C
>>> >>debugging
>>> >>experience is reduced that would be a regression.
>> >
>> >
>> >Yes, but I didn't see a difference so far.  Debugging optimized code is
>> >still a pain.
>> >
>> >
>>> >>My major concern is locking us into this and it being forgotten and we sit
>>> >>on
>>> >>DWARF2 for ages and we do not see or notice regressions related to
>>> >>DWARF3/4
>>> >>when it breaks on these archs. Can ARM/PowerPC/SPARC tools be built with a
>>> >>target option that limits the target libraries to DWARF2 ?
>> >
>> >
>> >A compromise would be to apply this only to GCC 4.8.  On PowerPC we have the
>> >situation that GCC 4.8 is the first version after 4.3 with all known bugs
>> >(bugs the render GCC useless and have no suitable workaround) fixed and no
>> >new ones (to our knowledge).  It will take some time to upgrade the debug
>> >tools in running projects.
>> >
> I'm fine with this compromise. If we apply it to gcc's development
> head then we need to review the change periodically to determine if
> DWARF 3/4 is ready to adopt for these couple of targets.
>

Chris pointed me to the undocumented CFLAGS_FOR_TARGET feature of GCC which is 
supported by the RSB with the --targetcflags option.  So you can build the 
multilibs with special flags.  I didn't test it yet, but I think this will make 
the patch unnecessary.

-- 
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone   : +49 89 189 47 41-16
Fax     : +49 89 189 47 41-09
E-Mail  : sebastian.huber at embedded-brains.de
PGP     : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.