RE: `completion-in-region'

["Drew Adams" <drew.adams@oracle.com>]

> > The question is whether matching like this should be the 
> > default. IMO, no.
> 
> It's not even a question.

Good. It seemed to be an implied question from Leo:

Leo> BTW, some modern text editors (such as textmate) have
Leo> started offering fuzzy completion as default.

> I could imagine using such a matching scheme by default at some
> point in the future, but it would need to be significantly refined.
> The main problem is that by turning "abc" into a global like 
> "*a*b*c" we tend to get many matches of variable quality.  So we'd
> need some way to compare the quality and only keep the ones that
> have the best quality.
> 
> At least that's what I try to do in my "forgive" completion-style
> (don't look for it, I don't think I've sent it out anywhere yet)
> which allows "abc" to match not only "albacore" but also "yacc"
> (i.e. it can insert&delete chars): the ability to delete chars
> means that fundamentally any input string can match any
> completion candidate, so there it's pretty obvious that you need
> to sort the candidates by "how close" they are to the input
> string.

Take a look at the URL I posted before:
http://www.emacswiki.org/emacs/Icicles_-_Fuzzy_Completion 

You'll see several fuzzy-matching techniques described there. All except the one
we're talking about here (scatter aka TextMate matching) have some _measure of
fit_ such as you describe.

And all have their uses and their drawbacks. I would not recommend using any of
them as the default matching method.