Indexing proposal

[Brad Crittenden]

On Sep 26, 2013, at 15:44 , Benji York <benji at benjiyork.com> wrote:

> On Thu, Sep 26, 2013 at 2:37 PM, Gary Poster <gary.poster at canonical.com> wrote:
>> 
>> That's an interesting question.  Autocomplete behavior is different than
>> search behavior: If I type in "ra", autocomplete should give me
>> "rabbitmq" but not a hypothetical "corral" charm, right?  Search should
>> give me both.  Do you, and others, agree?
> 
> That would be fine.  At a minimum the autocomplete results should be a
> subset of the "real" search results.  My intuition is that they should
> be exactly the same.
> 
> Why a subset?  Because if a person starts typing something into the
> search box and sees the result they are interested in, they should be
> able to hit enter and that thing be one of the search results.  That is
> not the case now and I believe this behavior is the root of one of the
> bugs.  I see the autocomplete functionality more as a quick search
> result preview.
> 
> I slightly prefer non-prefix matching for autocomplete so so that if
> someone types "apache" they get both "apache2" and "my-custom-apache" in
> the autocomplete.

I was skeptical but I am with Benji here.  I see that Google and Amazon work basically as he described.  At Amazon "flash" yields "flashlight" and "usb flash drive".  Google seems more heavily weighted towards prefixes but does return non-prefix results occasionally.

--bac