[
https://issues.apache.org/jira/browse/LUCENE-7316?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15316639#comment-15316639
]
Karl Wright commented on LUCENE-7316:
-------------------------------------
Looks like this is a problem because of minimum resolution. Here's some
debugging output:
{code}
[junit4] 1> Looking for coplanar points for: [[lat=0.0425265613312593,
lon=0.0([X=1.0002076326868337, Y=0.0, Z=0.042561051669501374])], [lat=0.0,
lon=-1.7156310907312492E-12([X=1.0011188539924791, Y=-1.7175506314267352E-12,
Z=0.0])], [lat=-0.7766317703682181,
lon=3.141592653589793([X=-0.7128972529667801, Y=8.730473389667082E-17,
Z=-0.7005064828988063])]]
[junit4] 1> planeToFind.evaluate(start) = 0.0
[junit4] 1> planeToFind.evaluate(end) = 0.0
[junit4] 1> failing because planeToFind = [A=1.7156310907312492E-12,
B=1.0; C=-4.031825447240733E-11; D=0.0] doesn't contain
[lat=-0.7766317703682181, lon=3.141592653589793([X=-0.7128972529667801,
Y=8.730473389667082E-17, Z=-0.7005064828988063])]; off by 2.7020217250132315E-11
[junit4] 1> planeToFind.evaluate(start) = 0.0
[junit4] 1> planeToFind.evaluate(end) = -2.0194839173657902E-28
[junit4] 1> failing because planeToFind = [A=1.7156310907312492E-12,
B=1.0; C=-1.7458530603341764E-12; D=0.0] doesn't contain
[lat=0.0425265613312593, lon=0.0([X=1.0002076326868337, Y=0.0,
Z=0.042561051669501374])]; off by 1.6416819695159931E-12
[junit4] 1> planeToFind.evaluate(start) = 0.0
[junit4] 1> planeToFind.evaluate(end) = -7.703719777548943E-34
[junit4] 1> failing because planeToFind = [A=5.543375111216114E-18,
B=-1.0; C=-1.3027230013345064E-16; D=0.0] doesn't contain [lat=0.0,
lon=-1.7156310907312492E-12([X=1.0011188539924791, Y=-1.7175506314267352E-12,
Z=0.0])]; off by 1.7175561810040737E-12
[junit4] 1> ... not coplanar
{code}
Note that the amount it is "off" by is just larger than
Vector.MINIMUM_RESOLUTION (1e-12). The triangle is therefore barely not
degenerate, but it is close enough to degenerate to make plane sidedness not
work well in determining inclusion of a given point.
I solved a similar problem for GeoBBox shapes by computing the bounds of
intersections as well as of edges and points. The only other solution is to
create a pair of cutoff planes for each edge of the polygon. That would be a
more accurate solution that we might want to adopt in the future, since acute
angles between planes otherwise will allow quite a bit of "leakage" of places
that shouldn't legitimately be "in set", but are, due to the MINIMUM_RESOLUTION
value.
> Geo3d test failure
> ------------------
>
> Key: LUCENE-7316
> URL: https://issues.apache.org/jira/browse/LUCENE-7316
> Project: Lucene - Core
> Issue Type: Bug
> Components: modules/spatial3d
> Affects Versions: master (7.0)
> Reporter: Karl Wright
> Assignee: Karl Wright
> Attachments: LUCENE-7316.patch
>
>
> Reproducible on master with:
> {code}
> ant test -Dtestcase=TestGeo3DPoint -Dtests.method=testGeo3DRelations
> -Dtests.seed=EEA08DD7FAE3C688 -Dtests.multiplier=2 -Dtests.slow=true
> -Dtests.directory=MMapDirectory -Dtests.locale=es
> -Dtests.timezone=America/Manaus -Dtests.asserts=true
> -Dtests.file.encoding=UTF-8
> {code}
> Note: I was initially unable to reproduce this, until I pulled up code that
> [~mikemccand] recently committed. It seems possible that encoding/decoding
> changes are triggering it. Of specific concern is the new way
> maximum/minimum decoded values are computed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
