Skip to main content.
home | support | download

Back to List Archive

Re: Future: Collation of Meta Tags?

From: Bill Moseley <moseley(at)>
Date: Sun Nov 06 2005 - 17:36:31 GMT
On Sun, Nov 06, 2005 at 10:25:55AM -0600, Peter Karman wrote:
> Bill Moseley scribbled on 11/6/05 12:48 AM:
> >man SWISH-RUN
> >
> >    swish-e -w not dkdkd -L date <= 17760705
> >
> ah. The -L option is not in the -h (help) output, which is where I looked 
> for it. I'll add that.
> Is it ready to be declared 'un'-experimental?

Here's the deal.  Swish has pre-sorted arrays that represent the sort
order of a given property.  Back when days had more hours in them, I
said, "well, if we have an ordered list of each property, then it
would be easy enough to use those same pre-sorted lists for limiting
results as well as sorting results."

So I implemented the feature.  But, that implementation is a by-product
of the pre-sorted arrays and it's not very scalable how it works
because it's has to create a duplicate array, sort it, do two binary
searches on it (which also means fetching actual property values from
the property file while doing the binary search), flagging items in
the array in the range, and then doing a second sort of the array.
All of that is very dependent on the size of the index.

It works, but probably not how one would design that feature, if that
feature was the goal and not a by-product.  Plus, the pre-sorted
arrays really get in the way of incremental indexing, so they may also
go away.  So if it's not "experimental" then it would need to be
reimplemented if the pre-sorted arrays went away, and that requirement
might get in the way of making incremental work better. ;)

That said, I depend on the -L feature.

Bill Moseley

Unsubscribe from or help with the swish-e list:

Help with Swish-e:
Received on Sun Nov 6 09:36:34 2005