[OAK-3991] Incorrect resultset from XPATH, multiple ORs and Lucene full-text

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 1.2.10, 1.3.15
• Fix Version/s: 1.0.37, 1.2.15, 1.3.16, 1.4
• Component/s: lucene, query
• Assignee: Thomas Mueller
• Reporter: Davide Giannella
• Created: 05/Feb/16 15:32
• Updated: 08/Oct/19 15:21
• Resolved: 14/Feb/16 07:14

Description

In case of more complex xpath queries with a mix of ORs, ANDs and
full-text the returned resultset is wrong. Some of the conditions are
lost during the query execution. Most probably during the Filter
creation.

The query is:

/jcr:root/test//element(*, nt:unstructured)[ 
  (
    jcr:contains(., 'cinema') 
    or @tags = 'architecture-keywords:building/cinema' 
    or @tags = '/tags/architecture-keywords/building/cinema' 
  ) 
  and @status = 'amber' 
  and ( 
    not(@id) 
    and ( not(@types) 
    or (
      not(@types = 'published') 
      and not(@types = 'page') 
      and not(@types = 'asset') 
    ) 
  )
)]

In this case if you replace the @tags conditions with a
jcr:contains rather than = it will make the query works. Which
suggests the same root cause of OAK-2660.

OAK-2660 was indirectly addressed by OAK-1617 but the OR->UNION
conversion is not applied to subqueries of a UNION set.

See attached patch for a reproducing test case.

Issue Links

• is related to: OAK-4324 Backport OAK-3991 (Incorrect resultset from XPATH, multiple ORs and Lucene full-text) to 1.2
• relates to: OAK-5507 Backport OAK-3991 (Incorrect resultset from XPATH, multiple ORs and Lucene full-text) to 1.0