[ 
https://issues.apache.org/jira/browse/JCR-4455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Julian Reschke updated JCR-4455:
--------------------------------
    Fix Version/s: 2.16.5

> condition index-rule handling more broken after JCR-4339
> --------------------------------------------------------
>
>                 Key: JCR-4455
>                 URL: https://issues.apache.org/jira/browse/JCR-4455
>             Project: Jackrabbit Content Repository
>          Issue Type: Bug
>          Components: core
>    Affects Versions: 2.18.2
>            Reporter: Ate Douma
>            Assignee: Claus Köll
>            Priority: Major
>              Labels: candidate_jcr_2_14
>             Fix For: 2.20, 2.16.5, 2.19.4, 2.18.3
>
>         Attachments: JCR-4455.patch, exposing-JCR-4455.patch
>
>
> [~c_koell]
> When reviewing the fix applied for JCR-4339 before going through an upgrade 
> to 2.18.2, I noticed that, while it fixes the reported problem, it does so 
> only in simple ('happy path') scenarios. Now it is broken when trying to use 
> multiple index-rule definitions for one node type.
> The logic for finding the applicable indexing rule for a specific property no 
> longer considers (checks) if the there is a condition for that property. 
> Instead, that check is postponed/moved to the method *calling* 
> #getApplicableIndexingRule.
>  But this is incorrect because now the #getApplicableIndexingRule method 
> returns the first *type* matching rule, regardless of the *property* it 
> should be applicable for.
> For example, it is perfectly feasible, and sometimes even needed, to have 
> multiple index-rules for the same node type, like the following enhanced 
> version of test resource indexing_config6.xml, which can be used to verify 
> the logic now is broken:
> {code:xml}
> <index-rule nodeType="nt:unstructured">
>     <property>other</property>
> </index-rule>
> <index-rule nodeType="nt:unstructured" condition="@foo = 'high'">
>     <property>foo</property>
> </index-rule>{code}
>  The important points to expose the new bug is:
>  * the index-rule for property other is defined *before* the index-rule for 
> property foo
>  * (for this example) the index-rule for property other doesn't have a 
> condition
> With the above, the property foo will *not* be indexed, regardless its value, 
> because the first 'matching' rule returned from #getApplicableIndexingRule 
> for a node of type nt:unstructured will be the rule for property other. But 
> will always return false on the (now postponed/delegated) call to 
> rule.isIndexed(propertyName: foo), because *that* rule doesn't has a 
> propertyConfig for foo (only for other).
> I'll attach a patch (based on trunk) to demonstrate the above failing using 
> the new IndexingConfigurationImplTest#testMatchCondition test.
> Note that the current #testMatchCondition() test itself also is broken: it 
> actually *does not* test the intended condition, but tests for it to *not* 
> match using assertFalse instead of assertTrue.
>  Which indeed is needed to pass the test because the indexing_config6.xml 
> configuration file itself contains an invalid (incomplete) index-rule.
> Instead of the current content:
> {code:xml}
> <index-rule nodeType="nt:unstructured" condition="@foo = 'high'">
> </index-rule>{code}
> it actually should be: 
> {code:xml}
> <index-rule nodeType="nt:unstructured" condition="@foo = 'high'">
>     <property>foo</property>
> </index-rule>{code}
> to pass the test with assertTrue.
> I'll also fix that test method in my patch, which then however will fail, 
> because of the above reported problem.



--
This message was sent by Atlassian Jira
(v8.3.2#803003)

Reply via email to