We need to do this because when I tride to go to the JSON serializer a
bunch of tests started failing because various parts of our code are
putting things into the session that are not JSON serializable.
We can't keep using the default pickle serializer because it defaluts to
using the highest available protocol and that will cause issues with the
python 2 to 3 upgrade since both will be running in production at the
same time. We need to use a version of the pickle protocol that both
can use interchangably.
We also need to make sure we read with latin1 encoding to make datetimes
work correctly between the two versions of python.
This will force a logout as sessions fail to load but this should be a
more performant and secure serializer moving forward. The reason we
overwrote it in our config is that it previously used to be the default
and we didn't want things to breake and force logouts when we changed
it. We're no more okay with people getting logged out.
The API documentation decorators do not have to leak which solution we
use to generate the docs. Here, and as discussed in PR #21820, we rename
the `openapi` module to `apidocs`, and we make sure that this module
includes all the right functions to document API Views without referring
to Open API.
The _id comes back as a dict and it's in the wrong order compared to the
dict stored in mongo. This results in lookups for deletion failing when
we use the '_id' as the lookup key.
Luckily the delete function also takes the asset_key as a valid key to
delete by so we just use that instead.
There is a bug in the underlying management/base code that tries to make
all manageent command output be unicode. This management command
outputs the binary tar file data and so breaks in python3. In python2
the code is happy to pass bytes back and forth and in later versions of
django this is fixed. Howevere it's not possible to get this test to
pass in Python3 and django 1.11
Treat transcript content as unicode strings and convert them at any edge
where we encounter them. One decision made here was to not update
edx-val it treats transcripts as byte but is much closer to the actual
files so it makes more sense over there. But within the platform they
are generally passed around as serialized json and so it's much better
for them to be unicode.
