Re: 3 dw2gencfi tweeks

[Andreas Jaeger <aj at suse dot de>]

Richard Henderson <rth@redhat.com> writes:

> (1) .cfi_remember_state/.cfi_restore_state didn't work as I'd
>     have expected.  Namely, we didn't remember the cfa offset
>     across the push/pop, which make it sort of confusing to
>     use .cfi_adjust_cfa_offset.
>
> (2) Data offsets not a multiple of the divisor were not detected, 
>     This leads to descriptions that are not what the user intended.
>
>     I suppose a more user-friendly course of action would be to
>     choose a gcd from the set of all offsets actually used.  This
>     might lead to more compact codes based on being able to use
>     DW_CFA_offset more often than DW_CFA_offset_extended.  However,
>     that seems like a lot of work, and I seriously doubt that
>     someone is *actually* going to be storing a register at an odd
>     offset.  That just seems like a typo to me.
>
> (3) Uli pointed out that .cfi_offset isn't particularly easy to use.
>     One has to remember the current CFA offset.  Which is silly
>     because we've got this handy .cfi_adjust_cfa_offset directive.
>     So this adds a .cfi_rel_offset directive that's relative to the
>     current CFA offset.

Please update the manual for this change,

Thanks,
Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj