Python Type Hints: Take It Easy

The introduction of the type hints supporting Python 2.7 and Python >
3.5 is one of the biggest changes in the history of Python. The best
feature of type hints: they are totally optional. Type hints help
detect bugs and enhance readability—but not always. They also
introduce complexity and make it harder to leverage some of Python's
most powerful dynamic features. The cost-benefit of using type hints
varies from project to project. Sometimes there is value in type
hinting part of a codebase but not all of it. This talk introduces
type hints and discusses the pros and cons of this epic new feature.

Transcript

1. 1 0 0 % o p t i o n a l
   TYPE HINTS: TAKE IT EASY
   Introduction, pros and cons of

   Python's gradual type system
   Luciano Ramalho
   @ramalhoorg
   

   View Slide

2. 2
   https://twitter.com/raymondh/status/1255580152706158592
   

   View Slide

3. TYPE HINTS?
   A brief introduction
   3
   

   View Slide

4. STATIC TYPE DECLARATIONS
   Formalized in PEP 484 — Type Hints (and other PEPs)
   Supported by the typing module in the standard library
   Mostly for the use of external type checkers, not runtime
   4
   def parse_flag(text: str) -> bool:
   return text.upper() in {'Y', 'YES', 'T', 'TRUE'}
   

   View Slide

5. ABOUT GRADUAL TYPING
   Dynamic typing: type checking at runtime only
   Ex. Lisp, Smalltalk, Python, JavaScript, Ruby, Clojure, Elixir, etc.

   Static typing: type declarations in source code, type checked
   before runtime by static analysis of the code
   Ex. C, C++, Pascal, Java, C#, Haskell, Rust, Kotlin, Swift, Elm, etc.

   Gradual typing: hybrid approach—optional static type
   declaration, no effect during runtime
   Ex. ActionScript, Hack, Dart, TypeScript, Python with type hints...
   5
   

   View Slide

6. EXAMPLES
   6
   

   View Slide

7. THE FIRST BUG MYPY FOUND FOR ME
   7
   

   View Slide

8. FIXED!
   random_queens_and_guard.py: https://tgo.li/rnd_queens
   8
   

   View Slide

9. "SUBPASCAL" INTERPRETER
   Example from Programming
   Languages Workshop

   at Garoa Hacker Clube
   https://tgo.li/OfLiPr2018
   9
   

   View Slide

10. SUBPASCAL PARSER
    10
    

    View Slide

11. IMPORTING TYPES FROM TYPING
    11
    

    View Slide

12. DATA CLASSES
    PEP 557, Python 3.7
    12
    

    View Slide

13. NAMED TUPLE (CLASSIC)
    13
    

    View Slide

14. TYPED NAMED TUPLE
    14
    

    View Slide

15. DATA CLASS
    15
    

    View Slide

16. DATA CLASS WITH MORE FIELDS
    16
    

    View Slide

17. OVERLOADS
    17
    

    View Slide

18. OVERLOAD EXAMPLE #1
    18
    

    View Slide

19. MAX OVERLOAD
    19
    

    View Slide

20. MAX: CURRENT TYPE HINTS ON TYPESHED
    20
    

    View Slide

21. RECENT
    DEVELOPMENTS
    21
    

    View Slide

22. USABILITY IMPROVEMENTS
    PEP 585 — Type Hinting Generics In Standard Collections
    Starting with Python 3.9 (backported to 3.7, 3.8)
    

    22
    from __future__ import annotations
    def loads(input: Union[str, bytes], *,
    encoding: Optional[str] = None) -> dict[str, Any]:
    ...
    

    View Slide

23. STATIC DUCK TYPING
    PEP 544 — Protocols: Structural subtyping (static duck typing)
    23
    

    View Slide

24. STATIC DUCK TYPING
    PEP 544 — Protocols: Structural subtyping (static duck typing)
    24
    

    View Slide

25. MORE REALISTIC EXAMPLE
    25
    

    View Slide

26. MAX: PROPOSED TYPE HINTS FOR TYPESHED
    26
    

    View Slide

27. MAX: PROPOSED TYPE HINTS FOR TYPESHED
    27
    

    View Slide

28. BENEFITS
    28
    

    View Slide

29. BENEFITS OF TYPE HINTS
    Document function signatures
    Support the IDE for better code completion and refactorings
    Finding bugs before running the program
    Automating object serialization/desseralization
    They are optional, avoiding the problem of diminishing returns.
    29
    

    View Slide

30. COST
    There is no free lunch
    30
    

    View Slide

31. DAVID BEAZLEY IS NOT ON BOARD...
    https://twitter.com/dabeaz/status/981699763996450817 31
    

    View Slide

32. A TSUNAMI OF PEPS (OUTATED! 17 PEPS AS OF JUN/2020)
    32
    Ano PEP Título Python
    2006 3107 Function Annotations 3.0
    2014 484 Type Hints 3.5P
    483 The Theory of Type Hints N/A
    2015 482 Literature Overview for Type Hints N/A
    2016 526 Syntax for Variable Annotations 3.6
    2017 544 Protocols: Structural subtyping (static duck typing) 3.8
    557 Data Classes 3.7
    560 Core support for typing module and generic types 3.7
    561 Distributing and Packaging Type Information 3.7
    563 Postponed Evaluation of Annotations 3.7
    2018 586 Literal Types 3.8
    2019 585 Type Hinting Usability Conventions 3.8D
    589 TypedDict: Type Hints for Dicts with a Fixed Set of Keys 3.8
    591 Adding a final qualifier to typing 3.8 3.8
    

    View Slide

33. DOWNSIDES
    Effort to learn a whole new set of abstractions
    Generics, variance, structural typing, type variables etc.
    Effort to write type hints
    Most of the time is not hard, but it may be hard or even impossible
    Better support in recent versions of Python
    But also: possible to use in Python 2.7, easing transition
    Slight delay in program startup
    Not important for long running processes
    33
    

    View Slide

34. COST X BENEFIT
    Benefits grow when...
    The codebase is large
    The team is made of professional software developers
    Bernat Gabor* gives a good rule:

    if you invest in automated tests,

    you should invest in type hints
    

    * link at the end
    34
    

    View Slide

35. FAQ
    Some answers
    35
    

    View Slide

36. WILL PYTHON BECOME A STATICALLY TYPED LANGUAGE?
    No.
    Section Non-goals of PEP 484:
    36
    

    View Slide

37. WHAT IS THE RUNTIME EFFECT OF TYPE HINTS?
    None.
    The Python interpreter and standard library ignore type hints*
    Also from PEP 484:
    37
    * a few exceptions, ex.: dataclasses and functools.singledistpatch at import time
    

    View Slide

38. SHOULD EVERYONE USE TYPE HINTS?
    No.
    Python users are very diverse: engineers, financial analysts,
    students, scientists and researches from almost every field
    (physics, biology, sociology, artificial intelligence).
    The cost-benefit analysis is not the same for all of these groups.
    To learn how to program or do exploratory programming, type
    annotations introduce complexity, reduce the flexibility of the
    code, and do not bring much value.
    Type hints is most useful in teams of professional developers,
    with large code bases, used to a more complex tooling (IDEs,
    linters, automated tests, continuous integration etc.)
    38
    

    View Slide

39. REFERENCES
    Before reading all those PEPs
    39
    

    View Slide

40. BERNAT GABOR: THE STATE OF TYPE HINTS IN PYTHON
    https://www.bernat.tech/the-state-of-type-hints-in-python/
    40
    

    View Slide

41. REAL PYTHON: PYTHON TYPE CHECKING GUIDE
    https://realpython.com/python-type-checking/
    41
    

    View Slide

42. Let's keep this conversation going!
    Twitter: @ramalhoorg
    E-mail: lramalho@thoughtworks.com
    DANKE SCHÖN
    

    View Slide