incubator-graffito-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Christophe Lombart (JIRA)" <j...@apache.org>
Subject [jira] Resolved: (GRFT-106) Avoid INFINITE RECURSION when Object Model has cycles.
Date Wed, 18 Apr 2007 19:35:15 GMT

     [ https://issues.apache.org/jira/browse/GRFT-106?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Christophe Lombart resolved GRFT-106.
-------------------------------------

       Resolution: Fixed
    Fix Version/s: ocm-1.0

Solve by using an object cache. This is not yet a complete cache system but this small cache
is used per persistenceManager and only for retrieve requests. It allows to perform circular
lookups (as by crossreferenced objects) that would result in non-terminating loops without
such a cache.

Later I would like to implement a complete object cache support. Something like in Hibernate
or OJB. There is a good description here : http://db.apache.org/ojb/docu/guides/objectcache.html.
I think we have similar needs but we have to define a proposal for this kind of feature with
the most useful use cases.

> Avoid INFINITE RECURSION when Object Model has cycles.
> ------------------------------------------------------
>
>                 Key: GRFT-106
>                 URL: https://issues.apache.org/jira/browse/GRFT-106
>             Project: Graffito
>          Issue Type: Improvement
>          Components: JCR-Mapping
>            Reporter: Dan Connelly
>         Assigned To: Christophe Lombart
>            Priority: Critical
>             Fix For: ocm-1.0
>
>
> The default ObjectConverterImpl is restricted to acyclic graphs in the object model.
> Many Java object models are NOT acyclic.   For instance, I am on your Friends list. 
 Yoar are on my Friends list.     Java encourages such structures.   Almost any large object
model in Java will have hidden cycles.
> Saving an Object Model that contains cycles using Graffito causes an infinite recursion.
> Clearly, it is important to maintain a 1-to-1 correspondence between Nodes and Objects
to prevent this.   In the absence of Multiple Parent Nodes, it will be necessary to use REFERENCE
or UNDEFINED Items in place of the 2nd (or greater) Node representing a given Object.   My
preference si that the default ObjectConverterImpl should support REFERENCE.,    Failing this,
use of UNDEFINED also solves this problem and would  acceptable (as default).  Whether or
not REFERENCE is used, both insertion and retrieval must provide a reasonable result.   A
custom ojbect converter should be available to switch UNDEFINED to REFERENCE, or vice versa.
> Also, it is probably best to keep the targeted, well-defined Nodes close to the Root
Node.    This implies that the default ObjectConverterImpl should implement a Breadth-First,
rather than a Depth-First, traversal of the Object Model on both insertion and retrieval.
  Again, if the default is Depth-First, a custom object converter should be available that
implements Breadth-First.
> Admittedly, support for (2 representations) X (2 traversals) implies a drastic refactoring
and/or rewriting of the ObjectConverterImpl class.

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


Mime
View raw message