[OPEN-ILS-DEV] Proposal for search modifications for Evergreen 2.4

[Dan Scott]

On Mon, Oct 15, 2012 at 02:03:09PM -0400, Mike Rylander wrote:
> On Mon, Oct 15, 2012 at 12:52 PM, Jeffrey Bond
> <jbond.evergreen at gmail.com> wrote:
> > Greetings Evergreen Developers,
> >
> > I’d like to propose a few changes to search functionality for the Evergreen
> > application.  Here is some relevant information on the changes. I had read
> > in an e-mail that blueprints may no longer be needed? I added one anyways
> > but I can remove it if needed. Note that all changes proposed here can be
> > turned on or off as defined in the wiki; if and when these changes are added
> > they will not change the functionality of Evergreen unless an organization
> > specific option is selected to do so. Thank you all for considering these
> > changes.
> >
> 
> Without details or some prototype code the community can't really help
> evaluate the design.  That's something that you'll want to happen to
> improve the chances of having your work included in main-line
> Evergreen.
> 
> Also, as Thomas mentioned, there is ongoing search improvement being
> both discussed and developed collaboratively, mostly in the IRC
> channel.  Some of Thomas' might impact yours, and there is higher
> level QueryParser work happening that might also either impact or
> inform your design.
> 
> Jumping in IRC and discussing (and pointing to any code you might be
> experimenting with) might make things easier.

Jeffrey did point to the wiki, where at least pg_trgm was mentioned and
five new tables were mentioned; but beyond that, yeah, more details
would certainly help.

I think it's encouraging that both Thomas and Jeffrey have started by
writing up some of their design thoughts in the wiki; Thomas has done a
good job of also adding in use cases with some specific examples. I
haven't totally bought into the way he wants to try to address those use
cases via the tsvector weights-as-classes, personally, but the use cases
themselves can be kept around for any potential solution. And it's way
easier to get a sense of the overall design & direction of a change of
this size from a document / email conversation than from slowly doling
it out on IRC (and hoping that the pertinent folks are on-channel and
able to pay attention).

While I'm babbling, I would like to take the opportunity to point to
https://bugs.launchpad.net/evergreen/+bug/1054755 primarily for the
"stop adding TSearch2 contrib/extension in new installs".  TSearch2
doesn't offer any functionality that we actually use these days, but it
does break ts_rewrite(), which offers fairly simple on-the-fly thesaurus
functionality that might be useful. Yes, there's still the problem of
getting rid of the TSearch2 extension from existing sites, but let's not
add to that pile of existing sites in the mean time...