[
https://issues.apache.org/jira/browse/LUCENE-8860?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16911753#comment-16911753
]
Adrien Grand commented on LUCENE-8860:
--------------------------------------
I suspect that the confusion is that we don't use minPackedValue and
maxPackedValue directly but a combination of them. For instance the current
logic short circuits the query on Inner nodes when the query shape fully
contains all indexed triangles, using the minimum (x,y) coordinates from
minPackedValue and the maximum (x,y) coordinates from maxPackedValue to compute
a rectangle that contains all MBRs. In your example this rectangle is [(0,0),
(30,20)].
The idea of this issue is that there are weaker conditions we can check that
allow us to make the same conclusion. For instance if we take the minimum (x,y)
coordinates from minPackedValue, the maximum x from minPackedValue and the
maximum y from maxPackedValue we get a smaller rectangle that contains the left
edge of all MBRs on the leaf block. If the query fully contains this rectangle
then it is guaranteed to intersect all triangles of the leaf given that
triangles have at least one point on the left edge of their MBR. In your
example this rectangle is [(0,0), (20,20)]. We can do the same test for the
right, top and bottom edges.
With your red box query, we wouldn't be able to make a decision on the inner
node since it doesn't fully contain any of the 4 rectangles that contains all
left, right, top and bottom edges.
> LatLonShapeBoundingBoxQuery could make more decisions on inner nodes
> --------------------------------------------------------------------
>
> Key: LUCENE-8860
> URL: https://issues.apache.org/jira/browse/LUCENE-8860
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Priority: Minor
> Attachments: fig1.png, fig2.png
>
>
> Currently LatLonShapeBoundingBoxQuery with the INTERSECTS relation only
> returns CELL_INSIDE_QUERY if the query contains ALL minimum bounding
> rectangles of the indexed triangles.
> I think we could return CELL_INSIDE_QUERY if the box contains either of the
> edges of all MBRs of indexed triangles since triangles are guaranteed to
> touch all edges of their MBR by definition. In some cases this would help
> save decoding triangles and running costly point-in-triangle computations.
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
