Re: [PATCH] mips gdbarch-ify REGISTER_NAMES, SKIP_TRAMPOLINE_CODE...

[Andrew Cagney <ac131313@ges.redhat.com>]

> Andrew Cagney wrote:
> 
>> 
> 
>> > Only tm-mips.h seems to define REGISTER_NAME, so this should be safe.
>> >
>> > 2002-08-20  Michael Snyder  <msnyder@redhat.com>
>> >
>> >       * mips-tdep.c (mips_skip_stub, mips_in_call_stub): Make static.
>> >       (mips_gdbarch_init): Set skip_trampoline_code,
>> >       in_solib_call_trampoline.
>> >       * config/mips/tm-mips.h (REGISTER_NAME): Delete.
>> >       (IN_SOLIB_CALL_TRAMPOLINE, SKIP_TRAMPOLINE_CODE): Delete.
> 
>> 
>> Just add a comment pointing out that most embedded mips targets don't
>> use this, instead they override it using the REGISTER_NAMES macro.
> 
> 
> OK.
> 
> 
>> I would also steer well clear of REGISTER_VIRTUAL_TYPE,
>> REGISTER_RAW_SIZE and REGISTER_VIRTUAL_SIZE and anything else related to
>> register numbers.
> 
> 
> When you say "steer well clear of", do you mean "don't touch them"?
> All but register_virtual_type have already been gdbarch-ified, and
> I was building up my courage to do that one (see separate thread).

Then please don't change that!   Mult-arching the mips properly depends 
on changes to remote.c and on figuring out exactly what all the include 
header spagetti in config/mips actually does.

It is simply too easy to introduce a change that modifies the layout of 
the register cache and, as a consequence, break GDB's ability to talk to 
specific remote targets.

Andrew