Hello,

The attached patch introduces a ``struct frame_id'' for recording 
frames.  Previously (as JimI discovered) GDB was using a confused 
combination of a frame's relative level and its base/fp address.

A ``frame_id'' is a frame-{addr,base,fp} + frame-{pc,contaddr} pair. 
Two methods are provided:

	get_frame_id()
and 
frame_find_by_id()

I've also made this new function more robust then the current 
find_frame_addr...() code - it stops searching (I hope :-) as soon as it 
has gone past the relevant frame.

At present the frame_find_by_id() doesn't do a PC based test that is 
needed to tie-break frameless (?) frames (er, frames belonging to 
frameless function instances).  Can anyone come up with a testcase that 
demonstrates the need for the PC comparison test?

varobj, MI and insight should all be updated to use this more reliable 
frame handle.

I'm going to leave this on the table for a week or so.

Thoughts?  JimI, does it fix your bug?

enjoy,
Andrew