Why You Should Care About Types by Luka Sterbic

By now you have probably all heard about Python static typing. But why should you care? Are types in Python even Pythonic? SPOILER: Guido says so. Is Python turning into Java? In this talk I’ll try to answer these questions and explain how type annotations helped my team scale, made our code more efficient and new hires happier.

We will start with how the typing system in Python came to be, what is the motivation for it and how it looks today. I will demonstrate how to gradually add types to a pre-existing codebase and what benefits will your team get in the long run from a 100% typed project. I will also cover Facebook specific technologies that have now been open sourced like our runtime type collection system, MonkeyType, and the scalable type checker, Pyre!

Transcript

2. • Luka Sterbic • Software Eng @ Facebook London • https://github.com/sterbic • @lsterbic • ✈ #AvGeek #PointsAndMiles whoami
3. 1. Why is typing important 2. Typing 101 3. Types in the real world 4. Lessons learnt Agenda
4. Why should I even care about typing?
5. What jobs?
6. What jobs?
7. What jobs?
8. What jobs? • Grep for it?
9. What jobs?
10. What jobs?
11. What jobs?
12. What jobs?
13. • Bootcamp • New hires do tasks for multiple teams • Team decision after ~6 weeks • Engineers can pick any team with headcount • eng productive –-≥ likely to join • Applies to Open Source too! Scale my team @FB
14. Are Python types Pythonic?
15. • If nothing else, because Guido said so • Are Python types Pythonic? Yes! > import this ...
 Explicit is better than implicit. ... Readability counts. ... In the face of ambiguity, refuse the temptation to guess.
16. • Inspired by types in other languages • Typing is still optional • Has no impact at runtime (*) • Typing errors can be ignored Is Python turning into Java? No
17. Typing 101
18. • PEP 3107 - Function Annotations (2006) • By Collin Winter and Tony Lownds • PEP 484 - Type Hints (2014) • By Guido van Rossum, Jukka Lehtosalo, Łukasz Langa How did it start?
19. What needs to be annotated?
20. What needs to be annotated?
21. • In some cases, variables • What needs to be annotated?
22. • In some cases, variables • What needs to be annotated? • For Python ≤ 3.6
23. Collections
24. Collections
25. Collections
26. Collections
27. Unions
28. Unions
29. Unions
31. TypeVars
32. The Type Checker $ mypy test.py test.py:4: error: Argument 1 to "sqrt" has incompatible type "str"; expected "float" • Pyre --≥ https://github.com/facebook/pyre-check
33. Typing in the real world
34. 1. 100% typed coverage from Day 1 2. Type checker in CI 3. Profit! The Easy Bit New projects
35. 1. ... first understand gradual typing The Hard Part Existing (Large) Projects
36. utils.py job.py
37. utils.py job.py
38. error: Argument 1 to "pretty_print" has incompatible type "str"; expected "Job" utils.py job.py
39. 1. Type common libraries --≥ utils.py 2. Try running the type checker 3. Type checker in CI 4. Ask people to add types to new code 5. Use MonkeyType The Hard Part Bit Existing (Large) Projects
40. 1. Collects types at runtime and logs them 2. Generates type stubs 3. Can even apply back type information to your code MonkeyType https://github.com/Instagram/MonkeyType
41. Stubs
42. $ pip install MonkeyType $ monkeytype run main.py $ monkeytype stub utils def pretty_print(job: str) -> None: ... $ monkeytype apply utils MonkeyType https://github.com/Instagram/MonkeyType
43. Lessons learnt
44. Where did the ducks end up? Duck typing in the typed world
45. Protocols error: Argument 1 to "duck_things" has incompatible type "PharaohDuck"; expected "Duck"
46. The Typeshed • Typing for the standard library is not part of it ?! • Separate repo: https://github.com/python/typeshed • Bundled with your type checker • Still far from perfect --≥ send a PR!
47. # type: ignore • Silences typing error on one statement • Not necessarily evil • https://github.com/python/typeshed/pull/1885
50. JSON
51. NamedTuple
52. Summary
53. • Type annotations are Pythonic • Start using them for the readability win • Get the extra safety of the type checker for free • The easy bit, new projects should be 100% typed • The hard part, existing large codebases • Start small with gradual typing • Use MonkeyType • Avoid JSON blobs and use NamedTuple instead Summary
54. Luka Šterbić Software Engineer @ Facebook London luka@fb.com
 github.com/sterbic https://github.com/Sterbic/PyConZA-2019