[ 
https://issues.apache.org/jira/browse/OAK-6378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16058904#comment-16058904
 ] 

Michael Dürig commented on OAK-6378:
------------------------------------

I like the idea of introducing a {{SegmentWriter}} interface! I think this is 
long overdue. 

OTOH I'm not to sure about the changes of the return types of the various write 
methods. I was experimenting with this myself a while ago and came to a similar 
result. What I dislike is the proliferation of various entities (e.g. 
{{BlobStore}}) into the record implementations and related classes (e.g. 
{{SegmentNodeState}}, {{SegmentNodeBuilder}}). In the end this is the 
consequence of those record implementation not being fully stateless. My 
preference back then was to better keep this "mess" inside the 
{{SegmentWriter}} class instead of spreading it around. 
Maybe we could tackle this by going in the opposite direction and have all 
{{SegmentWriter}} methods uniformly return a {{Record}} instance? Alternatively 
we could try introducing a {{RecordFactory}} and use that one to create 
{{Record}} instances from {{RecordId}} s instead of spreading the individual 
dependencies?

> Move the SegmentWriter API to its own interface
> -----------------------------------------------
>
>                 Key: OAK-6378
>                 URL: https://issues.apache.org/jira/browse/OAK-6378
>             Project: Jackrabbit Oak
>          Issue Type: Bug
>          Components: segment-tar
>            Reporter: Francesco Mari
>            Assignee: Francesco Mari
>             Fix For: 1.8, 1.7.3
>
>
> In order to isolate {{SegmentWriter}} from the rest of the implementation, 
> and to match the corresponding {{SegmentReader}}, I extracted the API exposed 
> by {{SegmentWriter}} to its own interface and moved its implementation in a 
> separate class.
> Moreover, I cleaned up the {{SegmentWriter}} a bit by letting. every method 
> of its interface always return a {{RecordId}}. Before the refactoring, some 
> methods returned concrete implementations of record classes. The cleanup 
> improves the uniformity of the {{SegmentWriter}} interface.
> I see potential in this change for the following reasons.
> * The concrete record implementations ({{SegmentNodeState}}, {{Template}}, 
> {{MapRecord}}, etc.) might be implemented directly on top of the 
> {{SegmentReader}} and {{SegmentWriter}} API, moved to a different package and 
> tested separately from the rest of the code.
> * {{SegmentWrier}} and {{SegmentReader}} provide a higher level API that 
> isolates the {{SegmentStore}} and its supporting classes. Code using only 
> {{SegmentWriter}} and {{SegmentReader}} might be able to use {{RecordId}} 
> instances as opaque handles to the underlying records, with beneficial 
> effects on code decoupling.
> I have a working version of the refactoring in [this branch on 
> GitHub|https://github.com/francescomari/jackrabbit-oak/tree/segment-writer].



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to