The bare return statement returns None by default. If nothing is to be upgraded, the original notebook should be returned.

It is not actually required to run the tests.

It is only used when explicitly enabled via the --cov option,
e.g. on GitHub actions, where we install the package manually now.

In Fedora, we don't run the tests with --cov, so we don't need the dependency.

…ors on read (#241)

* Catch AttributeErrors thrown by ipython_genutils during reading and conversion as ValidationErrors

* fix docstrings

* catch attribute errors as validation errors while reading json

* revert changes to converter (in another PR)

… ValidationErrors (#240)

* Catch AttributeErrors thrown by ipython_genutils during reading and conversion as ValidationErrors

* fix docstrings

* revert changes to reader and extra test files, update error message

* fix test_nbformat

* Ensure nbformat minor version is present when upgrading
Motivation:
  - We have seen a number of notebooks which do not contain the nbformat
    minor version, this throws a key error when trying to update. This
    pr adds a check to ensure it is present and throws a validation error
    if not.

* fix test

* Only fix cell ID validation issues if asked

Adds a new kwarg to validate called `repair` which will fix well-known validation
errors during the validation process, but only when set to `True`.

Updates existing functionality to not mutate the notebook contents, but instead
fail validation.

* Remove python 3.5 from ci test matrix

* explicitly exclude jupyter-trust script from manifest

* Ensure that validate default mutates notebook content to repair cell ids

* Correct kwarg name for clarity

Co-authored-by: Paul Ivanov <pi@berkeley.edu>

Strip metadata that is causing notebook validation to fail

We need write permissions on the contents to create a GitHub release per their [example](https://github.com/ncipollo/release-action#example)