Establish a PandasBaseObject for all pandas objects

[ghost]

generic.PandasObject is not it since it applies to data containers, not Period or TimeStamp for example.

Should leverage subclasse's __unicode__ to do the right thing
on py2/py3 when it comes to repr,str,bytes #3183

I suggest putting in a mandatory self.random_example() (name?) method
that generates a random instance of the object. Convenient for testing, and for code snippets
in test cases/issues. Complements the more general makeX generator functions in util.testing.

Side benefit, can more easily add one-size-fits-all sanity check for a pickle roundrip,
which would have caught #2891.

More ideas?

[jreback]

with #2915 (and others), will be able to unify all primary objects with PandasObject (mainly putting Series under this umbrella), @wesm already seems to be remove numpyness from Index

I would just add needed methods to PandasObject and maybe make a PandasScalar base for things like Period/Timestamp (as a mixin)?

[ghost]

PandasObject references AXIS*, which doesn't apply to things like Period.
Could rename things and take over PandasObject, but right now there's no
single superclass that all relavent pandas classes derive from, and that entails
some code dupe and inconsistent behaviour.

[wesm]

Sure, I'm +1 on having a base object to be able to say concretely "here be pandas". Have to be careful about how that class makes its way into, say, tslib.pyx if you want it to be a superclass of Timestamp

[jreback]

no I agree.....

I think may be able to entirely drop PandasObject in favor of NDFrame after Series unification
so can reuse the name (or maybe PandasScalar?) I actually like PandasObject as the base of everything

[ghost]

Great, that's quite the concensus. names... whatever. we'll draw lots.

[jreback]

closed via #4092