New frame migration path ...

New frame migration path ...

[Andrew Cagney]

Below is a sketch for migrating existing ISA's to the new frame unwind code.

(ok, this is a bit light on details :-().

- if you have get_saved_register(this), replace it with 
deprecated_unwind_register(this->next) (don't worry, I'm going to also 
deprecate get_saved_register).

I'm about to send out a patch that adds a new architecture method - 
deprecated_unwind_register() that, while largely equivalent to the old 
get_saved_register(), is paramterized with the next, instead of this, frame.

The problem is that the new code relies on being able to do calls like:
	frame_register_unwind (next);
	... value of register in this, uncreated, frame ...
and that has no equivalent using get_saved_register() vis:
	get_saved_register (next->prev)
	... value of register in this, uncreated frame ...
(er, hold on next->prev == NULL).  MichaelL's fix to x86-64 dummy-id has 
this problem - while the patch looks correct, it can't work without the 
above!

Given that, for the most part, architecture specific 
get_saved_register() functions look like:
	frame = get_next_frame(frame);
	... find value to unwind ....
e.g., sh, and more importantly cfi.  The migration should be `straight 
forward' ...

It's deprecated since it is definitly only an intermediate step.

- implement gdbarch_unwind_pc(), delete frame_saved_pc()

This will likely require further fixes to the register unwinder et.al. 
It needs to correctly unwind the PC.

- delete POP_FRAME.

Again, could involve fixing the register unwind code.

- implement gdbarch_unwind_dummy_id()

- slowly implement custom unwinders

Thoughts?

Anyone interested in trying it?

Andrew