Re: New ARI warning Sat Mar 19 01:54:11 UTC 2011

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

Pedro Alves wrote:
> On Wednesday 23 March 2011 17:16:01, Joel Brobecker wrote:
> > > This was a large diff, but in fact, there is only one new warning:
> > > 
> > > > gdb/i386-tdep.c:1693: obsolete: frame_register_read: Replace frame_register_read() with get_frame_register(), or possibly introduce a new method safe_get_frame_register()
> > > gdb/i386-tdep.c:1693:     && frame_register_read (this_frame, cache->saved_sp_reg, buf))
> > 
> > I just had a look at this ARI warning.  The comment on
> > frame_register_read says:
> > 
> > /* FIXME: cagney/2003-02-02: Should be deprecated or replaced with a
> >    function called get_frame_register_p().  This slightly weird (and
> >    older) variant of get_frame_register() returns zero (indicating the
> >    register value is unavailable/invalid) if either: the register
> >    isn't cached; or the register has been optimized out; or the
> >    register contents are unavailable (because they haven't been
> >    collected in a traceframe).  Problem is, neither check is exactly
> >    correct.  A register can't be optimized out (it may not have been
> >    saved as part of a function call); The fact that a register isn't
> >    in the register cache doesn't mean that the register isn't
> >    available (it could have been fetched from memory).  */
> > 
> > I have had this feeling that we have way too many ways to read/write
> > frame registers, but I'm wondering if this comment might not be
> > too zealous in this case. This function seems useful, because it
> > returns a status as opposed to get_frame_register, which has the exact
> > same profile except that it throws instead of returning. So I'm thinking
> > we should remove the "deprecation" fixme, and just keep the FIXME for
> > fixing whatever incorrectness might be left, and then remove this from
> > the ARI.
> > 
> > Thoughts?
> 
> Agreed.

I think that all users that require this additional status information
should just use the (new) get_frame_register_value, and look at that
value's properties.

So I do think that frame_register_read ought to stay deprecated;
we need to remove those extraneous frame register routines ...

Bye,
Ulrich

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