Re: Dwarf 1 or 2?

Re: Dwarf 1 or 2?

[Jason Merrill]

>>>>> Mike Stump <mrs@wrs.com> writes:

> If one was going to switch to dwarf, what would be the better dwarf to
> pick?  Dwarf 1, or 2?  I ask because quite a few different groups
> (processor abi groups) seem to picking up dwarf 1, and no one seems to
> be behind 2 (except Linux and sgi).

Dwarf 2 is clearly technically superior.  Dwarf 1 has more existing
implementations, and therefore more political clout.

> I'm interested in user experiences with dwarf 2 systems, (sparc v9
> linux, iris6) with gdb and how well it holds up in relationship to
> stabs.

One important win of dwarf2 over stabs on x86 linux is that it allows for
debugging of frame-pointerless code.  In theory, that is; gdb doesn't
grok the frame unwind info yet, but all the necessary info is there.

Jason

Re: Dwarf 1 or 2?

[meissner]

| > The dwarf 2 spec can be found in
| 
| >   ftp://sgigate.sgi.com/pub/dwarf
| 
| > if anyone's interested in doing that work.
| 
| If one was going to switch to dwarf, what would be the better dwarf to
| pick?  Dwarf 1, or 2?  I ask because quite a few different groups
| (processor abi groups) seem to picking up dwarf 1, and no one seems to
| be behind 2 (except Linux and sgi).

Dwarf 2 is a better debugging language.  The ABI groups are going with Dwarf 1
because it is a standard, and the Dwarf 2 group got caught in the changes in
the System V world and went under before having it become a standard.  The
PowerPC ABI group for instance has a bunch of changes to dwarf 1 to support C++
better.  From a practical point of view, the dwarf-2 code in gcc (and possibly
gdb) is still new enough that it may still have bugs.

| I'm interested in user experiences with dwarf 2 systems, (sparc v9
| linux, iris6) with gdb and how well it holds up in relationship to
| stabs.
| 

Re: Dwarf 1 or 2?

[Robert Lipe]

Hi, Mike.

> If one was going to switch to dwarf, what would be the better dwarf to
> pick?  Dwarf 1, or 2?  I ask because quite a few different groups

I certainly have no preferences or really even any useful information,
but in case you're compiling a list, the currently shipping SCO native
x86 stuff uses Dwarf 1 for ELF.   The new tools due to ship Real Soon 
Now (this is public knowledge, not an NDA violation) will generate 
Dwarf 2.   This would be their "Universal Development Kit" which hosts
on OpenServer, Unixware 2, or Unixware 7 (a.k.a. SVR5) and generates
binaries that will run on any of the three, obvious exceptions noted.

> I'm interested in user experiences with dwarf 2 systems, (sparc v9
> linux, iris6) with gdb and how well it holds up in relationship to
> stabs.

I'd like to hear the results of your survey.   I've been building
our embedded MIPS --wiht-stabs becuase GDB wouldn't play nice with
G++ any other way.   I have no compelling reason to change, but I 
always like having more options! :-)

Later,
RJL

Dwarf 1 or 2?

[Mike Stump]

> To: egcs@cygnus.com
> From: Jason Merrill <jason@cygnus.com>
> Date: 12 Dec 1997 15:28:13 -0800

> The dwarf 2 spec can be found in

>   ftp://sgigate.sgi.com/pub/dwarf

> if anyone's interested in doing that work.

If one was going to switch to dwarf, what would be the better dwarf to
pick?  Dwarf 1, or 2?  I ask because quite a few different groups
(processor abi groups) seem to picking up dwarf 1, and no one seems to
be behind 2 (except Linux and sgi).

I'm interested in user experiences with dwarf 2 systems, (sparc v9
linux, iris6) with gdb and how well it holds up in relationship to
stabs.