[
https://issues.apache.org/jira/browse/LUCENE-7168?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15229927#comment-15229927
]
Karl Wright commented on LUCENE-7168:
-------------------------------------
[~mikemccand]: I realized suddenly how to do this properly.
The trick is to line up the BKD tree cells that it descends through with the
"fuzzy" cells from the point resolution. So you never would generate an
XYZSolid that did not correspond exactly to an integral number of "fuzzy" point
cells. This implies that you also could not have a BKD cell that was smaller
than a fuzzy point cell, and thus the situation we're trying to address could
never arise.
To clarify, imagine the space that you are searching as consisting of a large
number of small cubes. Each cube has an integer address, e.g. (xi, yi, zi).
You can put an actual GeoPoint on the world surface into an individual cube via
quantization, so you then get to describe it by (xi, yi, zi). When we do the
BKD descent, the way we must describe the solid areas we are searching is not
by ranges of (x,y,z), but by ranges of (xi,yi,zi). Then you can convert those
ranges trivially into (x,y,z) ranges and create the XYZSolid objects you use to
relate with the shape. This guarantees you will not ever create an XYZSolid
that contains a point which does not intersect with the world, since the
original point *had* to have been on the world in the first place.
Sound like it might work?
> Remove geo3d test leniency
> --------------------------
>
> Key: LUCENE-7168
> URL: https://issues.apache.org/jira/browse/LUCENE-7168
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Michael McCandless
> Fix For: master, 6.1
>
> Attachments: LUCENE-7168.patch, LUCENE-7168.patch, LUCENE-7168.patch,
> LUCENE-7168.patch, LUCENE-7168.patch
>
>
> Today the test hides possible failures by leniently handling quantization
> issues.
> We should fix it to do what geo2d tests now do: pre-quantized indexed points,
> but don't quantize query shapes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
