[
https://issues.apache.org/jira/browse/AVRO-637?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12903752#action_12903752
]
Scott Carey commented on AVRO-637:
----------------------------------
bq. I don't follow this logic. If they pass a List, then its order will be
preserved. But if someone passes in an unordered data structure, why should
they expect order to be preserved?
Producers and Consumers are sometimes decoupled and really different users.
User A wants to create a list from a Set and persist that later, user B
intercepts the serialization process and tries to append something to the list.
The Avro contract for an array type preserves order. So, for example one
helper method might alter a type and assume that the data is ordered and
appendable (it was in 1.3, and comes out of deserialization that way), then a
different user applies this helper method to an object constructed some other
way, and the result is a surprise.
The raw type is Collection, which does not guarantee this so such a user would
be mistaken, but the semantics are more complicated and error prone.
If we are going as far as Collection and abandoning the idea that the data in
memory is consistently ordered, sortable, and appendable we should probably
consider going one step further up the interface inheritance tree, past
Collection to the top: Iterable. Then you could serialize arrays that don't
fit in memory too.
bq. 1.4 already makes changes to generated APIs (Utf8 -> CharSequence)
Good point, another change in 1.4 along these lines doesn't add much
incremental work to upgrade.
> GenericArray should implement Collection
> ----------------------------------------
>
> Key: AVRO-637
> URL: https://issues.apache.org/jira/browse/AVRO-637
> Project: Avro
> Issue Type: New Feature
> Components: java
> Reporter: Doug Cutting
> Assignee: Doug Cutting
> Fix For: 1.5.0
>
> Attachments: AVRO-637.patch, AVRO-637.patch
>
>
> It would be nice if Avro arrays were better integrated with Java collections.
> The GenericArray interface permits array element reuse, which is awkward
> with java.util.Collection. But if GenericArray implemented Collection and
> the Avro runtime permitted arbitrary Collection implementations to be passed
> for Arrays then it would simplify many applications. The runtime could still
> reuse elements if an array implemented GenericArray, so performance would not
> suffer for applications that, e.g., loop over a data file, reusing instances.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
