65 Python Developer Jokes Every Pythonista Knows

I spent an hour debugging an IndentationError. It was one tab where there should have been four spaces. The file looked identical.

Python is the only language where pressing the spacebar wrong can take down production.

My code ran in Python 2. My code ran in Python 3. My code did not run in both at the same time. That took six years to admit.

The Python 2 to 3 migration started in 2008. I finished mine in 2023. I was not the last.

I typed pip install. Half the internet downloaded. The package I actually wanted was three dependencies deep.

requirements.txt said it worked on her machine. Pipfile said it worked on his machine. poetry said it worked yesterday. uv finally said it worked, and now nobody trusts it because it was too fast.

I created a virtualenv. Then I created a venv inside the virtualenv. Then conda activated something. I have not used my real Python in two years.

import this is a poem about clarity. The codebase I inherited has not read it.

Beautiful is better than ugly. Explicit is better than implicit. My coworker just imported everything with a star.

I asked about the GIL in an interview. The room went quiet. The interviewer poured himself a drink.

Threading in Python is a polite suggestion. The GIL is the parent in the room.

I gave the data scientist a CSV. She made it a DataFrame. I gave her a single integer. She made it a DataFrame. I gave her a coffee. I am now in row 0.

NumPy is not a library. It is a worldview.

I tried to write a for loop in front of a NumPy user. She closed her laptop and walked out.

I picked Django because it had everything. I picked Flask because it had nothing. I picked FastAPI because it had types. I am now maintaining all three.

Django: here is your admin panel, your ORM, your migrations, and your opinions. Flask: here is a function. Good luck. FastAPI: here is a type hint. Worship it.

The data scientist said tests slow her down. The model she shipped predicts that every house costs negative four hundred dollars.

I added type hints to the codebase. Everyone said thank you. Nobody ran mypy.

Type hints in Python are like seat belts in a parked car. Technically present.

The notebook has two hundred cells. None of them run in order. The kernel has been alive for nine days. We do not restart it. We have built a civilization on top of it.

I opened her Jupyter notebook. Cell 47 imports pandas. Cell 12 uses it.

I asked why a variable existed. She said it was defined in a cell she deleted in March. The variable is still in memory. It runs the company.

This used to work, he said, pointing at code that has never worked in any Python version that has ever existed.

I rewrote the loop as a list comprehension. It was three times slower. Nobody warned me. The book just said comprehensions are faster.

Nested list comprehension is read-only code. You write it once. You never read it again. The code reads itself, weeping.

I forgot to put __init__.py in the folder. Python refused to acknowledge the folder existed. The folder filed for divorce.

Then I added __init__.py and it still did not work because of the sys.path. I cried in a virtualenv.

What is the difference between __str__ and __repr__? One is for humans. One is for developers. I never remember which.

Duck typing means if it walks like a duck and quacks like a duck, it is a duck. In production, it is a goose with a duck costume and a NoneType in its pocket.

I read the asyncio docs. I read them again. I read them a third time. I wrote a synchronous version and shipped it.

async def, await, asyncio.run, asyncio.gather, asyncio.create_task, asyncio.ensure_future. I picked one at random. It deadlocked.

Decorators take three reads to understand. The first read, you panic. The second read, you almost get it. The third read, you write your own and inflict it on the team.

A decorator that takes arguments is a function that returns a function that returns a function. Nobody on the team knows which return is which.

Mutable default arguments. The footgun that ships with every Python install.

def add(item, items=[]). I will let you discover the rest yourself.

I imported pandas at the top of a script that does not use pandas. The cold start went from 0.2 seconds to 1.4 seconds. The script runs in a cron every minute. I am the reason for the cloud bill.

Someone in the standup said dunder. Half the room nodded knowingly. The other half pretended to nod knowingly.

I caught an exception with bare except. Six hours later I learned the keyboard interrupt was being swallowed.

ImportError: attempted relative import with no known parent package. This is the error that ends careers.

f-strings made Python beautiful. Then someone discovered f-strings can run arbitrary expressions, and now my logs contain function calls that have side effects.

Pythonic means whatever the loudest person on the team says it means today.

I shipped a one-line lambda. The code review comment said this is not Pythonic. I rewrote it as a four-line function. The code review comment said this is too verbose. I quit and learned Go.

The walrus operator landed in 3.8. I hated it. I refused to use it. Two years later I wrote (n := len(items)) without thinking. The walrus had won.

I added a match statement to a codebase. The PR sat for nine days. The reviewer said he would learn it next sprint. He has not learned it next sprint for three years.

Structural pattern matching is elegant. It is also the only Python feature I have to look up every single time. The elegance has a cost and the cost is my memory.

List, tuple, set, frozenset. I picked tuple because I read somewhere it was faster. The function mutates the collection in three places. The tests do not run.

a is b returned False. a == b returned True. The junior on the team submitted a five-page postmortem. The senior said welcome to Python.

if x is None. Not if x == None. The linter will tell you. The reviewer will tell you. Your dreams will tell you.

Someone wrote True = False at the top of a Python 2 file. The bug took a week to find. Python 3 made True a keyword. I owe the core developers a beer.

A new hire wrote print without parentheses and looked surprised when nothing happened. He had not touched Python since 2011. I welcomed him to the future.

10 / 3 returned 3.333. 10 // 3 returned 3. The data pipeline rounded a million rows the wrong way. The auditor was not amused.

range(10) does not return a list anymore. I learned this by trying to index into it and getting a TypeError. I have been writing Python since 2009.

Dicts preserve insertion order since 3.7. I built a feature around this. It worked. The senior reviewer said do not rely on it. I sent him the PEP. He stopped responding.

I replaced a list comprehension with a generator expression to save memory. The function downstream called len on it. The function crashed. The memory was fine. The career was not.

yield from is one of those features that looks magical until you need to debug it. Then it is just from, with extra steps.

I wrote a context manager with __enter__ and __exit__. It worked. I rewrote it with @contextmanager and a yield. It also worked. I do not know which one is correct anymore.

@dataclass replaced 40 lines of boilerplate with one decorator. Then someone wanted a custom __eq__. We added 38 lines back.

NamedTuple, dataclass, TypedDict, pydantic, attrs. Five ways to do the same thing. The team uses all five. In the same module.

I opened the typing module to look up one thing. Two hours later I was reading about ParamSpec and TypeVarTuple. I never found the thing I was looking for.

mypy said the type was wrong. The runtime did not care. The runtime never cares. mypy is a yelling shadow.

A junior committed a .pyc file. Then a __pycache__ folder. Then the entire venv. The PR was 12,000 files. The CI ran for an hour.

I added sys.path.insert(0, '..') to make an import work. Three years later the entire app loads in the wrong order and nobody knows why. It was me. It was always me.

The conftest.py exists. It just exists nowhere near where you expect. The fixture you need is in a conftest two directories up that the docs never mentioned.

I wrote a pytest fixture that depended on a fixture that depended on a fixture that depended on the first one. Pytest told me about it. Politely.

The data team shipped a Pandas UDF that runs for four hours on 200 rows. I rewrote it with iloc instead of loc. It runs in 11 seconds. Nobody asked me. Nobody thanked me.