Incorporate json config file to make mango-types pluggable. about mango

[cjnolet]

I like this. I'd also like to make a ticket to have the mango types perform (at least in function) more similar to Keith Turner's lexicoders which were added to the Accumulo API recently. He's got some neat optimizations for different types (like doubles especially)

[eawagner]

-1
I am fine with this being pluggable, but we need to get away from using static mutable contexts. They severly limit the reuseability of the code. So my gripe is more with the json config file. It is fine if an application defines their configuration using a json config file themselves, but this should not be built into the types API.

[cjnolet]

Yes. I like being able to plug in xml/json configs to represent the types. And I agree, the contexts are bad.

[eawagner]

With the contexts being non static coming up with a serialization strategy is actually key for accumulo-recipes. The reason for this is that this information must be passed to the Accumulo iterators for proper iterator operation for anything but the default set of types. Therefore the iterators are not able to handle custom types (which was one of the points to this api).

I am must more inclined to go for a json representation to xml. Mostly for space and readability reasons. However, since the TypeNormalizer/TypeEncoder interface are fully self describing, we really only need a list of class names. This means we can use something as simple as csv as the default config representation.

[cjnolet]

I'm not sure if we need this, do we?

[eawagner]

No I don't think we need this. Even if we did try this it would be difficult to create a generic configuration that could account for encoders with a configurable state. For example a BigInteger encoder with a maxSize parameter. We should still make the encoders serializable to allow them to be shared between distributed systems, but no config is needed.