allow raw json to be used without parsing

[costin]

There are cases where the raw json might be used as input or sent as output (without transforming it into an object).
This should be supported either by trying to automatically identify the time or by looking at a flag.

[ash211]

I'd appreciate this for my use cases -- deserializing JSON in Scala is much easier than the nested MapWritable objects, especially with the JVM's type erasure.

[costin]

Note: The Writable class will still be used whether you see them or not because the data needs to be serialized/deserialized by the Hadoop infrastructure

[tzolov]

I have an use case that needs such feature as well.
Can't we use org.apache.hadoop.io.Text to wrap the raw json?

[Downchuck]

How about using ByteWritable (seems like this proposal stalled on Text) and then in the addToIndex use scratchPad.setBytes? I've noticed on some pull requests in other bug reports that there have been design considerations stopping acceptance. Is there a particular design consideration that would get in the way of a patch here?

[Downchuck]

Any updates on this bug -- is this something that's just waiting on a pull request?

[Downchuck]

This bug seems to be stagnant -- issue #75 refers to the Pig interface, this issue is only about the MapReduce interface.

[costin]

Supported in master and 1.3 M2