This is the mail archive of the gdb@sourceware.cygnus.com mailing list for the GDB project. See the GDB home page for more information.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Forwarded: Dwarf changes


FYI, for those of you not also on PLSIG (because we are the only
embedded toolchain that implements Dwarf-2, I will be the secondary
contact for it in the PPC-EABI):



I attended the PowerPC Embedded ABI meeting last week.
The PPC EABI committee, if you are not aware, is an ad hoc
committee of companies supporting IBM/Motorola PowerPC.
The PPC EABI endorses Dwarf 1.1 as its debugging format.
Dwarf 1.1 is Dwarf 1 with some minor changes in the processor
dependent areas. 

There have been other changes to Dwarf 1 proposed and apparently
adopted by the PPC EABI committee, I believe to improve support
for C++, but these have not been well publicized and I don't know
the details.

There was a long discussion about whether to extend Dwarf 1
or to endorse Dwarf 2.  Some comments were raised that Dwarf 2
needs modifications, although with only a brief verbal presentation
of the proposed changes I was not able to see a strong rationale 
for adopting them.  I volunteered to "champion" changes to Dwarf 2
(with the assistance of Mike Meissner of Cygnus).  

Part of this discussion also revolved around changes to Dwarf 1.
Some of the suggested changes were to pick out pieces of the
Dwarf 2 specification and add them to Dwarf 1, and to add various
other functionality to Dwarf 1 which is duplicated in Dwarf 2, but
which would be implemented in a significantly different fashion.
These extensions sounded like they would be incompatible with 
the current Dwarf 1 standard.

As undercurrent for this discussion was the expressed opinion 
that since UI no longer exists and the PLSIG no longer has a sponsor, 
that the Dwarf spec has been abandoned and anyone can claim ownership.
This is not the first time I have heard this opinion -- the last
time was with the Tools Interface Standards committee, a Intel
x86-oriented ad hoc industry group, which has since folded.

I am concerned that a proliferation of different versions of Dwarf 1
or Dwarf 2 will make producing tools which use these formats more 
difficult, and that there will be no readily available and 
authoritative source of documentation which describes the format.
We have to look no further than COFF (Common Object File Format) to
find a format which is "common" but has several incompatible variations.
The same holds true for IEEE 695, where the debug information has
been changed significantly over the years, and Stabs, which have a
number of variations, none of which seem well documented.

I do believe that industry groups which are focused on a specific 
processor, whether it be PowerPC, x86, or other, should determine
the processor dependent portions of the Dwarf specifications.  I am
far less sanguine about the appropriateness of such a narrow, 
processor oriented group deciding on structural changes in the
specification.

Is there interest in reconvening the PLSIG, looking for a sponsor, 
and asserting ownership/control over the Dwarf specification?

-- 
Michael Eager	 eager@eagercon.com
1960 Park Blvd., Palo Alto, CA 94306  650-325-8077