You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, for each TO class we manually implement test_de_json and some classes also implement test_de_json_required and test_de_json_localization. Personally I'm okay with the explicit repitition since automating tests across several classes can be error prone itself (see #4593 and also the convoluted logic in https://github.com/python-telegram-bot/python-telegram-bot/blob/v21.9/tests/auxil/bot_method_checks.py).
However, the findings above show that we should try to ensure that we test all uses cases in all classes, that being
deserialization works if optional arguments are missing
deserialization works if additional arguments are passed (backward compatibility for new arguments)
To be discussed¹: deserialization works if required arguments are missing (forward compatibility for TG removing arguments)
If required for that class: deserialization correctly localizes datetimes
I can see at least three options for ensuring that a test class tests all required use cases
Introducing helper methods that do all that and calling them in the test classes. Adantvage: Less repetition. Disatvantage: Still need to ensure that the helper methods are called and also the downside of increased test complexity.
Introduce abstract base class for TO-Class tests with abstract methods test_de_json, test_de_json_required, test_de_json_locatization(?) and possibly others. This should force subclasses to implement all relevant tests. Advantage: Enforced unified setup without added test complexity. Disadvantage: Still need to check that the ABC is used. Could be done with a meta-test.
Add a meta-test that checks simply that methods with the corresponding names exist. Advantage: Enforced unified setup without much added test complexity. Disadvantage: Less explicit than 2.
Yeah, the abstract base class for tests + meta test sounds pretty robust to me. Could also enforce testing of test_slots, to_dict, equality and friends.
Looking at the discussion in the tdlib issue, I strongly suggest to do 3 as well
by point 3 you mean testing the forward compatibility, right?
This issue is a consequence of #4634 and #4617 (comment).
Currently, for each TO class we manually implement
test_de_json
and some classes also implementtest_de_json_required
andtest_de_json_localization
. Personally I'm okay with the explicit repitition since automating tests across several classes can be error prone itself (see #4593 and also the convoluted logic in https://github.com/python-telegram-bot/python-telegram-bot/blob/v21.9/tests/auxil/bot_method_checks.py).However, the findings above show that we should try to ensure that we test all uses cases in all classes, that being
To be discussed¹:deserialization works if required arguments are missing (forward compatibility for TG removing arguments)I can see at least three options for ensuring that a test class tests all required use cases
test_de_json
,test_de_json_required
,test_de_json_locatization
(?) and possibly others. This should force subclasses to implement all relevant tests. Advantage: Enforced unified setup without added test complexity. Disadvantage: Still need to check that the ABC is used. Could be done with a meta-test.At first glance, I'm in favor of 2.
@harshil21 @Poolitzer do you have any preferences?
¹ So far, Telegram has always made the removal of fields backward compatible (with the exception of
thumb
back in 2015), seeOf course, as per usual, Telegram doesn't document such things.
The text was updated successfully, but these errors were encountered: