[
https://issues.apache.org/jira/browse/TINKERPOP-1722?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16084631#comment-16084631
]
Marko A. Rodriguez commented on TINKERPOP-1722:
-----------------------------------------------
This is complicated given how integrated {{StarGraph}} is with
{{VertexWritable}} and Hadoop (InputFormats) and Spark (RDD). There are many
methods beyond standard {{Vertex}} (for {{StarVertex}}) as well as some
assumption about what {{StarGraph}} (as {{Graph}}) can and can't do. In
principle we could make {{StarGraph}} an interface and allow people to
implement that interface, but that would definitely require a {{TestSuite}} to
verify valid semantics for all such implementations and this may then get us
back to where we are where consistent behaviors are strongly required. Perhaps
the biggest gain would be a different serializers would be allowed.
> Allow graph implementations to supply their own implementation of StarGraph
> ---------------------------------------------------------------------------
>
> Key: TINKERPOP-1722
> URL: https://issues.apache.org/jira/browse/TINKERPOP-1722
> Project: TinkerPop
> Issue Type: Improvement
> Components: hadoop
> Reporter: Bryn Cooke
>
> Currently there is only one version of StarGraph and it is impossible to
> supply a different implementation.
> Allowing an implementation to supply their own would allow the following:
> * More consistent behaviour between OLTP and OLAP behaviour where behaviour
> is not well defined in TP.
> * Better optimisation such as allowing schema information to be shared across
> many star graphs.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
