Don't overwrite json package

[indygreg]

While the streaming Jackson API is much faster than the built-in json module, I think it is a bad idea to overwrite the built-in json module with the monkeypatched functions out of principle. There are legitimate cases where someone may want to utilize the additional features of the built-in json module APIs. Using Jackson shouldn't preclude this from occurring.

I think the Jackson JSON API should be exposed under say "import jacksonjson" or similar.

[tarasglek]

No.

jython ships with a lot of python libraries, most are implemented in python. Unfortunately python is damn slow when not using native extensions, jython is even slower. Benjamin wasted a couple of hours trying to figure out why jobs using native jython json would hang, turned out they were so slow the job tracker gave up on them.

[indygreg]

This has nothing to do with jython's json module being slow: it's all about removing a core module from the execution environment. http://docs.python.org/2/library/json.html has a number of handy APIs that jydoop's json module does not. For example, dumps() and the ability to decode classes from JSON.

It's quite conceivable a job may want to employ one of these (albeit slower) APIs. However, because you have overwritten the built-in module, it's impossible.

FWIW, I've already heard at least one other person gripe about this. I think taking away a tool is a bad idea and the decision should be reconsidered.

[tarasglek]

The solution here is to add missing functionality to jydoop.

[indygreg]

Here is the reference implementation of the json package for whoever wants to reinvent the wheel: http://hg.python.org/cpython/file/a26df2d03989/Lib/json