[
https://issues.apache.org/jira/browse/LUCENE-5608?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14001383#comment-14001383
]
David Smiley commented on LUCENE-5608:
--------------------------------------
One thing about the API I don't love is that to write an efficient Cell impl,
you have to lazy-decode from bytes. This is because the code using the Cell
creates Cells sometimes only to check leaf status and/or the level, or maybe to
check if it's within (underneath, a suffix of) another cell. Alternatively,
there could be some simple methods on SPT like readCellIsBoolean() and
readCellLevel(), and have Cell.isPrefixOf() take a BytesRef instead of another
Cell. Ok, a couple simple new methods, but then there is the conundrum of the
parameters to abstract methods in AbstractVisitingPrefixTreeFilter like
visitLeaf() that take a Cell, which would maybe would be modified to take,
what, a BytesRef; and maybe the length and leaf boolean? That doesn't seem
right. So lazy-decode it is.
> SpatialPrefixTree API refactor
> ------------------------------
>
> Key: LUCENE-5608
> URL: https://issues.apache.org/jira/browse/LUCENE-5608
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/spatial
> Reporter: David Smiley
> Assignee: David Smiley
> Fix For: 5.0
>
> Attachments: LUCENE-5608__SpatialPrefixTree_API_refactor.patch
>
>
> This is a refactor of the SpatialPrefixTree spatial API, in preparation for
> more SPT implementations on the near horizon. These are fairly internal
> APIs; SpatialExample.java didn't have to change, nor the Solr adapters, and I
> doubt ES would have to either.
> API changes:
> * SpatialPrefixTree & Cell had a fairly significant make-over. The existing
> implementations for Geohash & Quad have been made to subclass
> LegacyPrefixTree & LegacyCell shim's, and otherwise had very few changes
> (performance _should_ be the same). Cell is now an interface.
> * New CellIterator which is an Iterator<Cell>. Includes 3 implementations.
> * PrefixTreeStrategy.simplifyIndexedCells was renamed to pruneLeafyBranches
> and moved to RPT and made toggle'able with a setter. It's going to be removed
> in the future but for the time being it remains a useful optimization.
> * RPT's pointsOnly & multiOverlappingIndexedShapes options now have setters.
> Future:
> * The AbstractVisitingPrefixTreeFilter (used by RPT's Intersects, Within,
> Disjoint) really should be refactored to use the new CellIterator API as it
> will reduce the amount of code and should make the code easier to follow
> since it would be based on a well-knon design-pattern (an iterator).
> I wish I had done this as a series of commits on a GitHub branch; ah well.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
