Re: [PATCH] Don't trim addresses in Dwarf debug info

[Jim Blandy <jimb at zwingli dot cygnus dot com>]

> > I've committed this:
> > 
> > 2000-03-27  Jim Blandy  <jimb@redhat.com>
> > 
> >         * dwarf2read.c: Revert Andrew Cagney's change of Dec 15, 1997.
> >         Don't include "bfd-elf.h".
> >         (address_significant_size): Delete variable.
> >         (dwarf2_build_psymtabs_hard): Don't set it, or check for
> >         consistency between it and the Dwarf 2 compilation unit header
> >         address size.
> >         (read_address): Don't mask off bits above
> >         address_significant_size.
> >         * Makefile.in (dwarf2read.o): Don't depend on $(elf_bfd_h).
> >         (elf_bfd_h): Remove variable; it's no longer used.
> 
> Jim, have you confirmed this doesn't cause any regressions?

It fixes problems for me, and it fixes the problems Stephane Carrez
reported.  It might possibly cause regressions on 64-bit hosts, but
here's why I committed it anyway:

The purpose of the original change is to ignore information from Dwarf
2 data.  It changes the meaning of the debugging data in a way which
is incompatible with the Dwarf specification.  This can't possibly be
correct.

I asked for an explanation of why it was the correct change, but your
explanation doesn't at all establish that it's GDB's problem.

So, since we no longer know exactly what the erroneous behavior was,
the best way to resolve this, I think, is to revert the change, and
then fix the problem properly when we discover what it was.  Since
reverting this change fixes bugs experienced by Stephane and myself,
I'd say we've reduced the number of bugs overall, and made the code
cleaner.