On 01/08/12 17:15, Stephen Allen wrote:
+1 on release.


On Wed, Aug 1, 2012 at 5:46 AM, Andy Seaborne <[email protected]> wrote:
Do we want to release after removing (or merely switching off) the
NodeCache?  TripleCache?

As the system seems slightly faster without, I'm minded to switch them off.

The Node cache might give a space advantage.  But there is another cache
(which does make a difference) in RIOT for IRI resolution, and so the string
for the IRI is shared for the same IRI input when using RIOT.

Integrating RIOT properly into jena-core is ready-to-go but I'm not
comfortable doing that juts before a release.

Without presuming the decision on the Node cache for this release, I've made
minimal changes (Node, Triple) to run without caches.  My expectation is
that no-one will, or even can, notice the difference from outside.  The Node
cache can be turned on explicitly.


I'd say totally remove both the Node and Triple caches.  Just
switching off the cache in Node still leaves the create() method
synchronized, and unnecessary checks into the empty cache on node
creation.

Done.

The word 'synchronized' does not appear in Node.java anymore.

A quick-and-dirty test: parsing BSBM 25M from compressed :: 132K TPS (2.7.2 jars) -> 136K (with changes, and some classes not jars)

        Andy


-Stephen


Reply via email to