Moving to GDB_MULTI_ARCH_PARTIAL

[Andrew Cagney]

> this suggests REGISTER_CONVERTIBLE doesn't need to be immediatly
>> converted.
> 
> 
>> However, it also explains why, at level one, it isn't strict about
>> overriding macros.  Given a case like the i386, the base cpu may be
>> multi-arched but the variants (tm-ptx.h and tm-symmetry.h) may not.
>> This again lets them continue to build and work. 
> 
> 
> 
> So if they can continue to work with the existing definitions, why aren't 
> REGISTER_CONVERTIBLE and PUSH_ARGUMENTS level 2?


(I'm currently going through and reviewing these - I've found two other 
candidates: PUSH_RETURN_ADDRESS (bug) and CALL_DUMMY_LOCATION (not a bug))

It is historic.  The very original gdbarch.sh was an all or nothing 
afair - the developer had to convert every single macro before they 
could turn on multi-arch.  Strangly (:-^) this got serious push back. 
The process was changed to be tolerant to partially multi-arch targets. 
  A few artifacts from the old way still linger.  When someone notices 
they get tweeked.

thanks for the pointers,
Andrew