best-practices.md

marp
true

Python best practices

Karthikeyan Singaravelan /(Software Engineer at Platform Portal | Python core developer)/

---

Agenda

• Better abstractions
• Developer tooling
• Testing
• Performance

---

Philosopy of Python

• 30 years of active development.
• Zen of Python.
• PEP8 as a standard.

---

Better abstractions

• Learn more about dunder methods and protocols.
• Using properties and caching.
• Abstractions in standard library.

---

Context managers for resource management

• Context manager protocol __enter__ and __exit__ for resource management.
• contextlib module provides handy wrappers.

try:
    db = Database()
    db.query()
finally:
    db.close()

with Database() as db:
    db.query()

---

Properties and caching

• Properties for computational attributes. Cache them if they are immutable.

class Person:
    first_name: str
    last_name: str

    def get_full_name(self):
        return f"{first_name} {last_name}"

    @property
    def full_name(self):
        return f"{first_name} {last_name}"

---

Abstractions in standard library.

• collections - namedtuple, defaultdict, Counter.
• itertools - functional constructs
• command line interface for modules - zipfile, tarfile, json, etc.
• Dataclasses.

---

Collections module example

collection = {}

for student, mark in marks:
    if mark not in collection:
        collection[mark] = []
    else:
        collection[mark].append(student)

from collections import defaultdict

collection = defaultdict(list)

for student, mark in marks:
    collection[mark].append(student)

---

Turn on warnings

• Helps catch potential errors in future.
• Ensures smooth upgrade process.

---

Developer Tooling

• IPython and notebook for interactive sessions.
• timeit for benchmarking. pyperformance from PyPI.
• CProfile for profiling.
• Automatic formatting - Choose one tool - autopep8, yapf, black.
• pdb for debugging. ipdb, pudb from PyPI.
• Optional typing with mypy.

---

Testing

• Use pytest for concise tests.
• Use coverage for code coverage.
• Use unittest.mock for mocking tests.
• Linters : flake8 and pylint.

---

Performance

• pypy is a JIT for computation intensive code.
• Cython for writing C interface code.
• Use native interface libraries. E.g. numpy, pandas, scipy, etc.