Re: [rfc][23/37] Eliminate builtin_type_ macros: Platform-neutral types for internal variables

["Ulrich Weigand" <uweigand at de dot ibm dot com>]

Daniel Jacobowitz wrote:
> On Sun, Aug 31, 2008 at 07:51:08PM +0200, Ulrich Weigand wrote:
> > One problem is the $_ variable, which holds a pointer value.  To
> > allow using a platform-neutral type for this, I'm introducing a
> > new platform-neutral pointer type builtin_type_void_ptr.  The
> > lenght of this type is chosen so that it can hold every CORE_ADDR
> > value.
> 
> I don't think this is a good idea.  Target-specific code should not be
> required to correctly handle jumbo pointers, e.g. in "(gdb) print
> foo ($_)".  In a command function, we must already have some idea of
> what the associated architecture is (e.g. to initialize the
> architecture used by the expression evaluator); why not use that?

I guess that would mean current_gdbarch for now.  However, I had been
wondering what that means for the multi-architecture case -- if you 
set $_ in one frame, and then switch to a frame using a different 
architecture and use $_ as inferior call argument there, we'll still 
get pointers the architecture doesn't understand ...

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com