Black is very opinionated, but sometimes those opinions change. Run the
new black version across the codebase, and lock down the version used by
pre-commit so we don't get surprised like this again.
When calculating totalCount for PaginatedList, if the links data does
not contain a last page, we can't know how many elements are contained
in the list. Set it to 0 in that case.
Fixes#1614
It turns out if an argument is optional for an overloaded method, the
type hints require the default argument specified to successfully match
against the correct signature. Furthermore, the second kind of
create_pull() does not support the draft argument at all, so remove it.
Fixes#1634
There is a bunch of repeated code checking reactions which has the issue
that it can easily get out of date with what GitHub accepts. Drop the
check, and rely on GitHub returning an error.
Fixes#1554
To build on the previous work supporting querying Workflows for GitHub
Actions, add a class and relevant methods to support querying runs of a
workflow.
The typing information for Repository.create_pull did no type checking
of the arguments at all. Using the overload functionality, specify both
styles of create_pull(). Drive-by a nearby function from using a generic
dict to Dict[str, str].
Use typing.overload for MainClass.get_user() since it can return two
disparate types. Correct the docstring for it, as well as locking down
Sphinx to <3 due to built docs issue and drive-by link correction.
Fixes#1550
The typing information for the update_file() method's content parameter
states it will only accept str, whereas it will accept both str and
bytes.
Fixes#1542
When Workflow support was added, the method calls were not added to
the typing file for Repository, and somehow the class name for the
Workflow typing file was incorrect.
While chasing coverage failures, I discovered PullRequestReview's do not
send back a URL at all, which means firstly, the url property is not
required because it will always be None, and secondly, the object can
never be completed. I'm not certain why this change broke the test, but
it looked brittle, refactor it to be clearer.
To increase coverage, sprinkle in some asserts for attributes that were
not checked -- this uncovered a bug in Issue that two attributes were
not properly initialized to NotSet.
This is the only class that includes property setters for two of the
attributes, which are untested, and even if they were used, would have
no impact since there is no method to submit the changes back to GitHub.
Drop them.
There were multiple calls checking if an element of a list was a string
or a string -- I suspect this pre-dates use of six in the code base, but
let's clean it up.
A lot of our missing coverage is caused by not checking repr(), which is
trivial enough to test against -- do so, and also clean up existing repr
tests to look the same by calling repr(). Where it was trivial, add a
few assertions about missing attributes.
The API endpoint Repository.get_workflows() calls returns data in a list
item, not directly, leading to some head scratching when calling it
against GitHub. Re-record the replay data.
client_id and client_secret are sent to GitHub encoded on the query
string, which GitHub has firmly deprecated, so much so as to email
people who are doing so. To discourage its use, raise a FutureWarning if
they are specified.
Fixes#1500
Previously you could call it only on Repository or NamedUser. Now you can call it with any GithubObject subclass -- and the type is carried through to the return value.
Andy Grunwald