On Wed, Apr 14, 2010 at 2:49 PM, Uwe Schindler <u...@thetaphi.de> wrote:

> > And 2.9's backwards compatibility layer in
> > TokenStream
> > was significantly slower.
>
> I protest! No, it was not slower, only at the beginning because of missing
> reflection caching! But this also affected the *new* API. With 2.9.x and old
> TokenStreams there is no speed difference, really.
>

but it wasn't like this initially. only after you put even more work into
the backwards compatibility layer, after discovering performance issues with
Solr, all happening in a minor release from major changes.

I guess Marvin is hinting that perhaps major changes could be associated
with major versions. For that example, perhaps more time could have been
instead spent upgrading Solr tokenstreams so it could move to 3.0 (rather
than almost a year later).

And I do think its a good example, you put a ton of work into this, but not
all the backwards compatibility can be done like this, and what if somehow
this one had slipped through without this caching? I think most users would
consider it strange to experience a performance degradation in a minor
release from major changes...

-- 
Robert Muir
rcm...@gmail.com

Reply via email to