Things that need doing before this can be released
--------------------------------------------------

+ Break requests are pretty temperamental
+ The memory access commands are quite capable of taking out the machine. 
  GDB itself may need teaching about our stack structure to avoid this 
  happening when the user attempts to get a backtrace.
+ Error checking/handling (there's not much at present)

+ The *gdb implementation is using OS_AddCallBack to get called back when
  the RISC OS binary is loaded.  This is very dubious and actually breaks
  for binaries loaded via Sunfish.  It looks it is working but as soon as you
  read the memory word at 0x8000, RISC OS stiffs (while reading word at 0x0
  works).
