Re: [i386newframe] New branch

[Andrew Cagney <ac131313 at redhat dot com>]

> Per Andrews suggestion, I just created a new branch, and checked in
> the attached on it.  This stuff works for me on
> i386-unknown-freebsd4.7.  Interix and x86-64 are almost certainly
> broken on the branch.  Let's fix those targets and merge this into
> mainline when Andrew is finished diddling with the interfaces :-).

s/diddling/deleting/

> +{
> +  /* FIXME: kettenis/20030302: I don't understand why the cache isn't
> +     already initialized.  */
> +  struct i386_frame_cache *cache = i386_frame_cache (frame, cachep);
> +

On the branch, while the frame code notionally uses the sequence:

	pc = frame_pc_unwind(next);
	id = frame_id_unwind(next);

it expands to:

prev->pc = next->unwind->prev_register (next->next, 
&next->prologue_cache, PC_REGNUM);
prev->id = prev->unwind->this_id (next, &prev->prologue_cache);

So while the PC is unwound first, it is the ID unwind call that is first 
to see prev's uninitialized prologue cache.

Thinking about it, the new frame's PC can be determined solely from 
unwound register values.  For the ID, however, the prev's frame's 
prologue first needs to be examined to determine, for instance, which 
register from the next frame needs to be unwound.

(the mainline doesn't do this).

Andrew