This method allows bots to validate their config info
in a standardized way. Adding the method to a bot is
optional, but recommended for bots with config options
that can be invalid, like api keys. The giphy bot serves
as an example.
The primary reason behind this is to allow the zulip
backend to validate config data for embedded bots. The
backend does not have a permanent bot class instance, so
validate_config() must be a static method.
bot_handler.quit() should be used whenever a bot
wishes to terminate. This allows a flexible reaction
suited to the bot's environment: For external bots,
sys.exit() will be called, whereas for embedded bots,
different code can be executed.
The new module mocks out calls to the requests library, which
is used by many of our bots that use third party services.
Mocking of requests.py is mostly orthogonal to our other
testing concerns.
All of the functionality of BotTestCase was either directly
moved to StubBotTestCase or replaced by similar functions,
and all bots have been ported to StubBotTestCase.
This is mostly a pure refactoring, but it also ensures
that `initialize` is called in a consistent way by most
of our test helpers. (This didn't cause problems before,
since some bots don't require initialization.)
From @showell:
We had a PR here with lots going on, and the commits weren't
very well organized, and then there were some tricky merge
conflicts from another PR. So I just squashed them all into
one commit.
What this does:
* allow you to configure your followup stream
* provide help in followup stream
* add more testing to followup stream
* add get_response() helper for tests
Fixes#173Fixes#174
Move `get_response` inside of `mock_http_conversation`, as it is not
used anywhere else. Also create `assert_called_with_fields`.
`assert_called_with_fields` calls the `assert_called_with` method of a
mock object by using an HTTP request and a list of fields to look for.
This sets us up to validate more aspects of the conversation,
and it also introduces the more rigorously checked
`unique_response` helper.
(This also fixes a minor copy/paste error from a prior commit
that was harmless.)
This method had two pretty easy-to-separate concerns:
* find the fixture data using our directory conventions
* use the fixture data to simulate a real HTTP request
Part of the goal here is to make the extracted functions a
bit easier to use in other TestCase-based classes without
needing to subclass from BotTestCaseBase, which is kind of
complex with its setUp/tearDown.
Before this change, we were looking for config files in
default locations in source control, which is not a good
place to look for them. Now `run.py` and friends have a
command line argument where users can specify the config
files.
Note that the change to server.py is only a partial fix
to make it so that bots that don't use third party config
files won't crash. That program needs an overhaul, anyway.
This makes the StateHandler functional. To reduce the
number of server roundtrips when fetching/updating the
state, the entire state is fetched ocne at bot
initialization and cached. All changes are stored in the
cache and only saved externally after handle_message()
has been executed.
Fixes#141.
With this change, StateHandler.put() does only accept JSON-able
objects by default. The incrementor test tried to store the return
value of send_reply(), a non-JSON-able MockObject, in the state.
Therefore, this commits also sets functional default test return
values for send_message() and send_reply().
Finally, it fixes the tictactoe bot which relied on directly
modifying the state_ attribute.
unittest includes by default all module-level classes that inherit
from TestCase and implement at least one method starting with 'test'.
Since it doesn't provide a convenient way for excluding TestSuites,
we need to manually filter out the unwanted testing of our test base
class itself.
This simplifies testing stateful bots by integrating the StateHandler
into the test library. As a side-effect, the mock bot handler gets
reused during a test, making the tests more realistic. The
StateHandler now keeps its state during a call to check_expected_responses,
forcing some stateful tests to be more verbose and explicit.
