bug#10963: 24.0.94; `dabbrev-completion' and `completion-cycle-threshold'

[Stefan Monnier <monnier@IRO.UMontreal.CA>]

> It mostly works, but there is one detail that seems to need a fix:
> Consider this scenario:
> 1. M-: (setq completion-cycle-threshold 4) RET
> 2. foo1 RET foo2 RET foo3 RET foo

> Now, repeated `C-M-/' commands cycle like this:
>   foo3/foo1/foo2

> but repeated `M-/' commands cycle like this:
>   foo3/foo2/foo1

> I think that the cycling order should be equal in both cases, and IMO
> the right order is the one followed by `M-/'  (ISTR that I've read
> something about this in the manual, but right now I can't find the
> place).

That will require more changes: the current cycling order of C-M-/ is
completely different from the one of M-/ because the completion table
used is an obarray (a sort of hash-table), so dabbrev's ordering is
completely lost, and after that the completion code uses its default
sorting (which gives preference to shorter completions, which in this
case makes no difference).  So we'll need to stop using an obarray, and
to pass an ad-hoc sorting function to the completion code.


        Stefan