mdvtools.tests.test_bad_data
============================

.. py:module:: mdvtools.tests.test_bad_data


Functions
---------

.. autoapisummary::

   mdvtools.tests.test_bad_data.test_bad_data


Module Contents
---------------

.. py:function:: test_bad_data()

   Test what happens with NaN, Infinity...
   Previously, although there was code for filtering `na = na[~numpy.isnan(na)]`,
   this didn't help if there also happened to be Infinity.

   The `json.dumps` default behaviour of `allow_nan=True` would then cheerfully output non-compliant JSON,
   which contrary to what the Python documentation states is not, as of this writing (2024-02-05),
   compatible with *any* standard JavaScript based decoders I tested (let alone 'most').

   We now use `allow_nan=False`, so the user would be alerted at project creation time rather than runtime,
   but also correct the error earlier in the process so that the particular case of `isinf` is handled.

   If there are no valid numbers at all in a numeric column, then an exception will be thrown,
   in this example we catch that and as of now, end up with a datasource with 0 columns.