Re: [PATCH/RFC] Get rid of deprectated_read_register_gen in i387-tdep.c

[Andrew Cagney <ac131313@redhat.com>]

> I checked in the attached patch.  Unfortunately this introduces a

Thanks for flushing these!

>    FAIL: gdb.base/default.exp: info float
> 
> in the testsuite.  The output now is:
> 
>    info float
>    No registers.
> 
> Which makes sense to me since it's the truth.  However, "info
> registers" and "info vector" print a slightly different message:
> 
>    info registers
>    The program has no registers now.
> 
> The reason is that the bit of code that prints this message:
> 
>       if (!target_has_registers)
> 	error ("The program has no registers now.");
>       if (selected_frame == NULL)
> 	error ("No selected frame.");
> 
> isn't executed if a print_float_info method exists in the target
> vector.  Is there any reason not to move these statements up such that
> they're executed even if a target provides the print_float_info
> method?

I can't think of a reason not to.

--

PS: My things to do real soon includes add a function:

get_selected_frame (thread_info *)
	if (!thread's target_has_registers)
	  error ("...");
	if (thread's selected_frame != NULL)
	  return thread's selected_frame;
	// now for the fun
	// selected_frame = find_frame_by_id (thread's selected frame-id);
	// if (selected_frame != NULL)
	//   return selected_frame;
	error ("No selected frame");

and then use that instead of the selected_frame global.

The bit that is commented out is more interesting.  It would mean 
differentiating between target_changed() and target_run() events.

The former shouldn't change the selected frame but the latter should set 
it back to current frame.  This will cause a behavioral change in GDB 
but I think it will really fix a bug.

Andrew