RFC: stack trace generation

Tue Jul 20 15:36:00 GMT 2004

Andrew Haley wrote:
> David Daney writes:
> > Arn't there two issues?
> > 
> > 1) Generate the raw stack trace.
> > 
> > 2) Make a pretty printout.
> > 
> > These patches deal with #2.
> > 
> > #1 is easily done where ever DWARF 2 unwinding is available (shameless
> > plug: libjava/sysdep/dwarf2-backtrace.cc).
>> #1 is no longer a problem because we have functions in libgcc to do
> it, and these functions work with any unwinder format, not just DWARF
> 2.
>
Yes. The dwarf2-backtrace.cc is probably mis-named as it uses the
functions that I believe you are referring to.
Being a very lazy person, I am only expending GCJ effort on the
platforms that I am currently paid to support (mipsel-linux*). But I
have to wonder if it would not be a win to use dwarf2-backtrace.cc with
the i386-linux-gnu target and enable -fomit-frame-pointer. From reading
some of the posts to this list, it seems that having the extra register
would be a good thing. It is a small amount of configure hackery to see
if it would work, but something I don't have on my to-do list.
David Daney