Subject: Re: CVS commit: src/gnu/dist/gcc/gcc/config/i386
To: Matthias Drochner <>
From: Krister Walfridsson <>
List: tech-toolchain
Date: 07/16/2004 23:02:17
[Sorry for digging up this old issue.  I was traveling at the time, and
did not notice this until now...]

On Tue, 1 Jun 2004, Matthias Drochner wrote:

> The "double" should be 4-byte aligned. (It is in the code, but
> the dwarf2 info is wrong.)
> I've checked a gcc-3.3.3 on Linux, and it fails in a similar way.
> There are some related PRs in gcc's bugzilla, eg ##6522, 10360,
> 14142, 14190, 14191.
> They are all more or less put down by the gcc people, there is
> not much hope that the problem will be fixed in gcc-3.3.x.
> (gcc-3.4 works, but there are serious structural changes which
> let a backport appear infaesible to me.)
> So it seems we have to live with either buggy dwarf debugging
> or incomplete stabs debugging for now...

GDB have problems with stabs; I don't think I have managed to
debug any non-trivial program on NetBSD 2.0 without GDB crashing
or misbehaving in other ways.  Switching to dwarf has solved these

And you really want dwarf when debugging C++ too.

I see this as choosing between a working debugger (that of course
have bugs, but it is the same bugs as on e.g. Linux), or a limited
debugger with lots of problems that handles one corner case that
the working debugger fails on...

So I think the quote from PR 25094:

  [It is] somewhat embarrasing if the system ships with a compiler that
  produces output that the systems debugger can't use.  I'd say that
  shipping 2.0 with this would be a Bad Thing.

is even more true for the current situation...

I think that gcc must be changed back to use dwarf.