Re: [Libunwind-devel] local vars

[Lassi Tuura]

Hi,

> I can't think of anything obvious that'd break if we returned a
> different answer for UNW_X86_64_CFA. Let me think about this a bit
> more before attempting to implement something.

Hm, are you certain about this?

The unw_step() output you quoted is on entry to unw_step(). As I understand
it, c->dwarf.ip points to the *next* frame's IP address, but c->dwarf.cfa to
CFA of the *previous* frame, and then the unw_step()/dwarf_step() performs a
sort of an edge transition which leaves c->dwarf.ip again pointing at the
next frame or 0 to mark the end. I think c->dwarf.cfa does point to the CFA
of the current frame (cf. apply_reg_state()) at the end of the step, but
'current' is a bit strange concept as the state straddles two frames.

I think the CFA is correct in that all the dwarf expressions evaluated
relative to CFA for that frame are correct. If you ask for CFA after
unw_step() it should be returning the right value; I am not sure it
returns the right value after unw_getcontext() before first unw_step().

Obviously libunwind callers shouldn't see the state straddling.

I'm definitely interested in any thoughts on this, as it will have some
impact on the fast trace implementation.

Regards,
Lassi