Skip to content

Extend & Unify Deserialization Testing #4635

Open
1 of 1 issue completed
Open

Description

@Bibo-Joshi

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 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

  1. deserialization works if optional arguments are missing
  2. deserialization works if additional arguments are passed (backward compatibility for new arguments)
  3. To be discussed¹: deserialization works if required arguments are missing (forward compatibility for TG removing arguments)
  4. 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

  1. 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.
  2. 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.
  3. 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.

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), see

Of course, as per usual, Telegram doesn't document such things.

Activity

  1. Bibo-Joshi commented on Jan 2, 2025

    @Bibo-Joshi
    MemberAuthor

    In light of point 3, I've openend tdlib/telegram-bot-api#686

  2. Bibo-Joshi commented on Jan 3, 2025

    @Bibo-Joshi
    MemberAuthor

    Looking at the discussion in the tdlib issue, I strongly suggest to do 3 as well

  3. harshil21 commented on Jan 3, 2025

    @harshil21
    Member

    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?

  4. Bibo-Joshi commented on Jan 3, 2025

    @Bibo-Joshi
    MemberAuthor

    deserialization works if required arguments are missing (forward compatibility for TG removing arguments)

    yes, this :)

  5. Bibo-Joshi commented on Jan 5, 2025

    @Bibo-Joshi
    MemberAuthor

    FYI, I'd like to have #4617 merged before starting on this, as that PR is touching a lot of these tests as well.

  6. github-actions commented on Oct 15, 2025

    @github-actions
  7. updated Priority Medium and Effort Medium on May 20, 2026
  8. added this to the v23 milestone on Jun 16, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    ⚙️ bot-apiaffected functionality: bot-api⚙️ testsaffected functionality: tests

    Fields

    Priority

    Medium

    Effort

    Medium

    Projects

    No projects

      Milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions