Anybody want to talk about DWARF?

rfg@ncd.com (Ron Guilmette - C++ Entomologist)
Wed, 22 Aug 90 05:24:40 GMT

          From comp.compilers

Related articles
Anybody want to talk about DWARF? rfg@ncd.com (1990-08-22)
| List of all articles for this month |

Newsgroups: comp.compilers
From: rfg@ncd.com (Ron Guilmette - C++ Entomologist)
Keywords: debug, design
Organization: Compilers Central
Date: Wed, 22 Aug 90 05:24:40 GMT

I've noticed over time that the messages in this newsgroup tend to focus
mostly on the more intellectually interesting aspects of compiler
construction (e.g. the recent discussion of intermediate representations and
their relationship to optimization techniques).


If I may, I'd like to see if any of the readership (other than me) would
perhaps have an interest in discussing (what may be viewed as) one of the
more mundane practical considerations of compiler construction in this day
and age.


In particular, I'd like to solicit people's views regarding the new format
for symbolic debugging information which is now being promulgated by AT&T as
a part of System V Release 4.


For those who don't already know, back in the the good ol' days, the COFF
`standard' (and I use the term loosely) specified the general format for
object files (i.e. headers, relacation information, etc.) and also specified
the format for symbolic debugging information.


As most COFF users already know, COFF left a bit to be desired in terms of
the symbolic debugging information. Therefore, I (and others) actually
welcomed the introduct (by AT&T) of yet another object file format (i.e.
ELF). I (and others) had high hopes that ELF would be dramatically better
than COFF, especially with regard to the provisions for symbolic debugging
information within the new format.


Well, unfortunately, all is not well here in the new age.


First, for reasons which are not entirely clear (but which may relate to the
internal opinions and external clout of one particular hardware
manufacturer) the ELF object file format does *not* include any
specification of symbolic debugging information. Rather, this information
has been segregated out and is covered (separately) under the name of DWARF.
The AT&T folks are not really going out of their way to promote DWARF.
Rather, their attitude seems to be more like "Here it is. Use it if you
want to. If you don't like it, that's OK too. We don't demand that anybody
use it."


Now this seems to be very broad-minded attitude on the part of AT&T (and
perhaps it is), but it seems that most folks who are implementing V.4 or
stuff related to V.4, or who are otherwise interested in ELF are too small
to try to swim against the tide and implement something other than the de
facto standard (i.e. ELF+DWARF). So the net result seems to be that most
companies who are doing ELF are also doing DWARF.


Now ELF+DWARF, while being an improvement over COFF, still leave some things
to be desired. I'd like to know if other people, companies, or organizations
have noticed any of the problems in ELF+DWARF that I have seen, and if so
what (if anything) they plan to do about them.


I'm posting this message in comp.compilers because I believe that this is
probably the best place to make contact with the kind of folks that would
have an interest in this subject (i.e. compiler & debugger people).


Please E-mail responses to me unless you think that you have something of
general interest to say on this subject.


// Ron Guilmette - C++ Entomologist
// Internet: rfg@ncd.com uucp: ...uunet!lupine!rfg
[I'd be interested to hear what DWARF's problems are, other than its name.
-John]




--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.