On Tue, Nov 29, 2011 at 05:17:26PM +0200, goran kent wrote:
> On Tue, Nov 29, 2011 at 3:52 PM, Marvin Humphrey <[email protected]> 
> wrote:
> > If the parsed queries differ, that is probably because the Schema being fed 
> > to
> > ClusterSearcher's constructor is incomplete or wrong
> 
> aha!  Thanks for that diagnostic approach - it has revealed that
> indeed I am an ape, gnawing at the monolith.

LOL -- while that's an entertaining image, I have to disagree. :)

This error was an API trap that we laid for you.  Had you avoided it, it would
be to your credit, but our design encouraged silent failure.  I don't like
blaming users for our shortcomings as designers.

I was in fact discussing this very problem with my colleague Dan Markham last
night.  We can solve it by serializing schemas and passing them around during
the handshake phase, silently adding valid field definitions and erroring out
noisily if conflicts are detected.

In fact, Dan's efforts to design a configuration-free search system led us to
the insight that there's no need for ClusterSearcher's constructor to require
a Schema at all.

Marvin Humphrey

Reply via email to