document the changes

Author: mattip

doc/source/whatsnew/v0.21.0.txt

@@ -332,13 +332,11 @@ Performance Improvements
 Bug Fixes
 ~~~~~~~~~
 
-
 Conversion
 ^^^^^^^^^^
 
 - Bug in assignment against datetime-like data with ``int`` may incorrectly convert to datetime-like (:issue:`14145`)
 - Bug in assignment against ``int64`` data with ``np.ndarray`` with ``float64`` dtype may keep ``int64`` dtype (:issue:`14001`)
-- Fix :func:`DataFrame.memory_usage` to support PyPy. Objects on PyPy do not have a fixed size, so an approximation is used instead (:issue:`17228`)
 - Fixed the return type of ``IntervalIndex.is_non_overlapping_monotonic`` to be a Python ``bool`` for consistency with similar attributes/methods.  Previously returned a ``numpy.bool_``. (:issue:`17237`)
 - Bug in ``IntervalIndex.is_non_overlapping_monotonic`` when intervals are closed on both sides and overlap at a point (:issue:`16560`)
 - Bug in :func:`Series.fillna` returns frame when ``inplace=True`` and ``value`` is dict (:issue:`16156`)
@@ -420,6 +418,15 @@ Categorical
   the ``.categories`` to be an empty ``Float64Index`` rather than an empty
   ``Index`` with object dtype (:issue:`17248`)
 
+PyPy
+^^^^
+
+- Compatibility with PyPy in :func:`read_csv` with ``usecols=[<unsorted ints>]`` and
+  :func:`read_json` (:issue:`17351`)
+- Split tests into cases for CPython and PyPy where needed, which highlights the fragility
+  of index matching with ``float('nan')``, ``np.nan`` and ``NAT`` (:issue:`17351`)
+- Fix :func:`DataFrame.memory_usage` to support PyPy. Objects on PyPy do not have a fixed size,
+  so an approximation is used instead (:issue:`17228`)
 
 Other
 ^^^^^