I'm at the PyConUS sprints, would appreciate a high-level review / if the approach I took here looks like what you expect. In particular I tried to keep behavior around all the unexpected potential return values from buffer.write but also add DeprecationWarning/RuntimeWarning for many. I'm planning to work on expanding tests for this tomorrow (clone of BufferedIO blocking test, tests for underlying writers that return all the bad values, partial write tests, etc).

cc: @vstinner, @serhiy-storchaka (commented on the underlying issue, but also was quite a while ago. If there's better experts or someone more appropriate let me know)

As I'm working on BufferedIO seeing a lot of cases where the buffering logic I'm testing moving towards there is very similar to what TextIOWrapper is doing internally, and hope I may be able to reduce a bit between the two (and allow utf-8 containing/encoded unicode objects to be buffered without a memcpy to make a temporary bytes object). I think that'll be a lot simpler than this, although may revisit depending on timelines.