[ 
https://issues.apache.org/jira/browse/JCR-2170?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12744549#action_12744549
 ] 

Jukka Zitting commented on JCR-2170:
------------------------------------

+1 to getting rid of the DefIds. IMHO they unnecessarily bind the persisted 
items to specific node type settings. Resolving the item definitions on demand 
makes the typing system more flexible.

Also, I'm somewhat scared about the DefIds being essentially just hash codes of 
the item definitions. What happens when we have a hash collision?

> Remove PropDefId and NodeDefId
> ------------------------------
>
>                 Key: JCR-2170
>                 URL: https://issues.apache.org/jira/browse/JCR-2170
>             Project: Jackrabbit Content Repository
>          Issue Type: Improvement
>          Components: jackrabbit-core
>    Affects Versions: 2.0.0
>            Reporter: Tobias Bocanegra
>             Fix For: 2.0.0
>
>
> the PropDefIds and NodeDefIds are used to quickly lookup a childnode- or 
> property definition in the nodetype registry (or effective nodetype).
> this is heavily used during reading, when calling Property.getDefinition() 
> usually when checking the isMultiple() flag. and of course while writing when 
> getting the definition for the property or childnode. 
> however, this poses problems when a nodetype is changed that is still used in 
> the content. if a property definition is changed due to an altered nodetype, 
> subsequent accesses to that property result in a "invalid propdefid" warning 
> in the log - but the id is recomputed. this is especially a problem when 
> upgrade jackrabbit from 1.x to 2.0, where some of the builtin nodetypes are 
> defined differently.
> i think that it should be feasible to remove the propdefids and nodedefids 
> and compute the definition on demand. i think this can be implemented without 
> performance loss, when some sort of 'signatures' of the items are computed to 
> quickly find the definitions in the effective node type. furthermore, the 
> most common usecase for using the property definition is probably the 
> isMultiple() check - which is now on the Property interface itself - which 
> does not need a definition lookup at all.
> and last but not least, it saves 8 bytes per item in the persistence layer.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to