woah gnarly

this may be naive, but can't we just extend our current encoder with a check for is NaN, Inf? like how we do with datetimes https://github.com/plotly/python-api/blob/master/plotly/utils.py#L224

situation kinda stinks, I actually documented it because this same thing was confusing to me as well:

https://github.com/plotly/python-api/blob/master/plotly/utils.py#L135-150

We only get a crack at stuff if the JSONEncoder can't encode it. What's more, as is mentioned in the SO post, when allow_nan=False, we also don't get a crack at it in our encoder. bummer!

Here's an example:

import json
import numpy

class Foo(object):
    pass

class NullEncoder(json.JSONEncoder):
    def encode_as_true(self, obj):
        return None

    def default(self, obj):
        return self.encode_as_true(obj)

json.dumps(Foo(), cls=NullEncoder)  # here, it works
# 'null'
json.dumps(numpy.ma.core.masked, cls=NullEncoder)  # and again here
# 'null'
json.dumps(float('NaN'), cls=NullEncoder)  # no shot! this doesn't work as expected!
# 'NaN'

wow, that's pretty wild. well, that looks good then, nice diggin!