Re: write_register_bytes() confusion

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

> I'm in the middle of bringing BDM for m68k support into current gdb, and
> while I have fixed the most blatant calling convention changes (pid
> changing to pid_t, etc) write_register_bytes() in regcache has got me
> horribly confused.

(BDM being a target (or backend) to GDB.)

> not to mention all this m68k craziness happening elsewhere.  ;)
> 
> in v1.61 of valops.c (which appears to be current) there is the snippet:
> 
> write_register_bytes (VALUE_ADDRESS (toval) + VALUE_OFFSET (toval),
>                       VALUE_CONTENTS (fromval), TYPE_LENGTH (type));
> 
> I'm writing a register -- it seems in my case this should be:
> 
> write_register_bytes (VALUE_REGNO (toval), VALUE_CONTENTS (fromval), 1);
> 
> the 1 of course is machine dependent

The above code behaves as per the designers intent.  The ADDRESS is an 
offset into registers[] array.

However, GDB is trying to move away from that mechanism and more towards 
something like what you describe.

> what I'm seeing is the register number getting completely thrashed on
> the way through.
> 
> yet all calls to write_register_bytes are using VALUE_ADDRESS?
> 
> insane.  I'm starting to think I need to back off -current and just
> stick with gdb-5.0...

An up-to-date target backend should only be using:

supply_register()
regcache_collect()

it should not contain code that refers to registers[].

enjoy,
Andrew