Re: RFA: modernization of ia64-tdep.c with new frame model for gdb-6.0 branch

[Kevin Buettner <kevinb at redhat dot com>]

On Aug 25,  2:12pm, J. Johnston wrote:

> Kevin Buettner wrote:

> > Okay, I see that you're turning r32-r127 and (not shown) p0-p64
> > into pseudo registers.  Is there any reason to leave big "holes"
> > in the register number space?  I.e, why not just get rid of all
> > of the empty strings above?
> > 
> > (Most of the time, the reason NOT to do this is because remote
> > targets depend on the order.  The only remote target that I'm
> > aware of is gdbserver, and I'm not particularly worried about
> > breaking compatibility.)
> > 
> > If I'm not mistaken, removing these holes will somewhat decrease
> > the size of struct ia64_frame_cache:
> > 
> >     +struct ia64_frame_cache
> >     +{
> >     ...
> >     +  /* Saved registers.  */
> >     +  CORE_ADDR saved_regs[NUM_IA64_RAW_REGS];
> >     +
> >     +};
> > 
> 
> Actually, number of real raw registers went down to the last non-pseudo
> register anyway.  My preference regarding renumbering registers would be
> to sync this up with gdbserver later.

Okay, so long as "later" isn't too much later.  It'd be a shame if someone
suddenly wrote a stub which depended on the holes being there...

> > Have you tested the nat bit related code in ia64_pseudo_register_read()
> > and ia64_pseudo_register_write() ?  My recollection is that my original
> > code didn't handle the unat bits correctly.  I was wondering if you
> > had fixed this problem.  (I'm curious about the other NaT bits too.)
> >
> 
> Could you elaborate about what problems you think existed in the previous
> code?

I'll send you a thread which describes the problems.  After rereading
that thread, it looks like the fix isn't as difficult as I had
remembered.  In fact, it should be pretty easy.  (But it still needs
to be tested.)

With regard to your current patch, I'm okay with you checking it in
after adding the ia64_frame_cache comments.

Thanks for doing these cleanups.

Kevin