Thomas Mueller created OAK-1689:
-----------------------------------
Summary: XPath and union queries with "or" can return the same
node multiple times
Key: OAK-1689
URL: https://issues.apache.org/jira/browse/OAK-1689
Project: Jackrabbit Oak
Issue Type: Bug
Components: query
Reporter: Thomas Mueller
Assignee: Thomas Mueller
XPath queries that contain "or" with can in some cases return the same node
twice. Those queries are converted to "union", as follows, where the same
problem exists:
{noformat}
select [jcr:path], [jcr:score], * from ...
where contains(*, 'square')
union
select [jcr:path], [jcr:score], * from ...
where [x] = '1'
{noformat}
Currently, the results of both subqueries are combined with
"Iterators.concat(left.getRows(), right.getRows())" (Iterators is from Google
Guava). That means results of the first part of the union is executed, then the
result of the second query. This result is wrapped / filtered using
"FilterIterators.newCombinedFilter(it, distinct, limit, offset, orderBy,
settings)", which is Oak code. "distinct" is true for "union" and false for
"union all".
Now, "distinct" operates on the result row object, which contains the path, the
score, and all property values. In previous versions, the "jcr:score" was
always a fixed value, but in the latest Oak version "jcr:score" is the value
returned from the index, which means the two rows are not equal:
{noformat}
path: /test/a, score: 0.5 (from the first subquery; Lucene index)
path: /test/a, score: 0.0 (from the second subquery; property index)
{noformat}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
