Python: The Documentary [video]
youtube.com
youtube.com
> I do show up, but not “live”. For various reasons (mostly related to declining health), I didn’t actively participate. I gave the director (Ida Bechtle) a pile of source material at the start, and that was the last of my involvement. She spliced in some recycled video of an interview I did with the PyPy folks at a PyCon some years ago, but I’m there mostly so people could recite my so-called “Zen of Python”, which an actual historian (Joseph Dragovich) assured me is “the most famous values statement for any programming language community”. https://discuss.python.org/t/python-documentary-going-live-t...
This is where the guru whispers (3 times) to you the seed letters which one has to imbibe over decades of sadhana by lacs of repetition
this looks similar
#rememberThis
how did i do on my python exam? 5.6 is also a pass!
With no real recent experience (I fell deep down the hole into C/kernel etc), I wouldn't have any authority to judge how it's adapted to time. But the most common complaint I've observed at companies of all size and sophistication is "the deployment story is a disaster". But it also seems like uv allows people to do all the "I want this specific version of such and such and I don't want $OS to know about any of this" well?
Re math/ai it's an interesting comment because a language is one part the syntax/stuff you receive, and one part the community that tells you to do things a certain way. I'd guess that Python has become such a big tent it is a little hard to enforce norms that other languages seem to have. This somewhat reminds me of Bjarne Stroustrup discussing scaling/evolving a language.
Packaging has been a nightmare. PyPI has had its challenges. Dependency management is vastly improved thanks to uv - recently, and with a graveyard of tools in its wake.
The modern Python script feels more like loosely combining a temperamental set of today’s latest library apis, and then dealing with the fallout. Sometimes parallels the Node experience.
I think an actual Python project - using only something remotely modern like 3.2+ standard library and maybe requests - is probably just as clean, resilient, and reliable as it ever was.
A lot of these things are and/or have been improving tremendously. But think to your point the language (or really the ecosystem) is scaling and evolving a ton and there’s growing pains.
Libraries can overtake aspects of a language for better and worse. Ruby seemed really tied to Rails and that was great for it as an example.
It feels like something broke around 2015-ish. Going back, you could make a whole app and gui with Basic. You could make whole websites simply with HTML+PHP, sometimes using nothing but Notepad. You could make portable apps in Java with no libraries - even Swing or whatever was built in.
Now…? Electron, a few languages, a few frameworks, and a few dozen libraries. Just to start.
Bizzare.
— Ronnie Coleman
But this seems like an apples to oranges comparison. Yes, of course a few scripts are very different than "something largeish" written in Python
I’m happy that python basically took over the role once filled by matlab, and I’m happy that it became the leader in AI dev instead of something worse (probably Java if gpt2 had hit 5-10 years earlier).
But you’re right. It’s not fun anymore. It feels more like a pseudolanguage for expressing tensors now, because of the influx.
I’m exaggerating only in feigned outrage. In my actions, I’ve been coding in rust, go, and zig ever since ChatGPT came out.
I think that moment made me value python less. When I think about why, it’s because python became less challenging, and the problem space shrank.
It’s been fun to go back to low-level and remember how computers actually work.
mypy --strict in CI & don't let dict[str, Any] pass review if the keys are constants, insist on a dataclass or at least a TypedDict.
good luck with justifying that before your manager (reacting to complaints within your team, claiming you being a bottleneck).
This exactly! Worded charitably, but accurately.
Python is a messy and complex language, full of historical baggage and non-essential machinery. I’ve been working in it for many years, so I’m not new or unfamiliar to it… But I know what a clean simple language looks like (eg.: clojure), so I have much less patience for unnecessary complexity now.
It may be arguably-better than some other popular languages (or MATLAB), but it's easy to find an objectively much better high-level language.
On the other hand duck typing is largely a joke. Letting functions take anything as an argument and then just assuming it's a duck with the methods you want is no way to write a robust system. Performance wise you're boxed into a corner. The slow runtime will have you writing native code in another language, complicating your project. The GIL will have you jumping through hoops to work around.
As an offline data exploration and research tool limited to an individual or a small team, or for writing small utilities, I totally get it. For anything mission critical I'd much rather be in something like Java or C# where the typing situation is stronger, the performance is going to be much better, I have better access to threads if I need them, the reasons for dropping into native code are fewer, and the cross platform support works more seamlessly without additional layers like Docker.
Back then the whole OO crowd was with Java though.
Python's Moat was beginner friendliness to write simple scripts, which at the time- Perl was more like a thermonuclear scripting language. Most people who never wanted to graduate to that advanced stage of writing mega million lines of Perl code over a week(which was literally the use of Perl then), realised they needed some thing simpler, easier to learn, and maintain for smaller scripts- Kind of moved to Python.
But then of course simplicity only takes you that far, and its logically impossible to have a simple clean interface to problems that require several variables tweaked. Python had to evolve and is now having the same bloat and complications that plague any other language.
Having said that, I would still use Java if I had to start a backend project.
In fact, many books that the OOP haters attribute to Java, predate its existence.
Everyone keeps harping on type safety, but it just doesn't play out in reality. Linux Kernel is incredibly robust, has never been a broken mess, and has basically no type enforcement, as you can cast pointers into other stuff.
In general, all typing does is move error checking into the compiler/preprocessor instead of testing. And time spent on designing and writing type safe code is almost equivalent to time spent writing tests that serve as an end-to-end contract.
There is a reason why NodeJS was the most used language before all the AI stuff came with Python.
>Performance wise you're boxed into a corner. The slow runtime will have you writing native code in another language, complicating your project.
Most of these performance arguments are similar to arguging that your commuter car needs to be a track spec Ferrari in terms of performance, by people that have very little experience with cars.
Plenty of fast/performant stuff runs on Python. PyPy is a thing also. So is launching small compiled executables like a process, that takes literally one line of code in Python.
>The GIL will have you jumping through hoops to work around.
This is just laughable. Clearly you have extremely little experience with Python.
Yes, as a result of strong development practices, a high skill floor, and where work is done by a small number of people knowledgable in the domain. These things are not mutually exclusive with type checking.
> In general, all typing does is move error checking into the compiler/preprocessor instead of testing
Which is an enormously powerful thing. Having those things be required by tests requires that the test exists for all the cases, which turns your tests into things that not only have to test behaviour but are also simultaneously responsible for checking types.
Compile-time type checking can essentially eliminate all need for type-based tests and is able to do it automatically for any code that exists, whereas this has to be opted in when using tests to check types.
You also don't get the benefit of compile-time tests automatically checking use cases for the types you don't support (which is likely more than the ones you do), whereas leaving this to test-time is practically impossible and you can only test the valid path. I've never seen a codebase that checks the negative paths for the hundreds-upon-thousands of types that aren't supported.
None of this is to say I'm against tests as end-to-end contracts, but moving type checking to compile time gives you a lot of extra kinds of assertions for free that you likely don't get from having tests to check types.
> There is a reason why NodeJS was the most used language
And the reason was? AFAIK Node came along as a runtime option for using a familiar language outside of the browser. Coupled with a single-threaded event-driven concurrency model out of the box it was an enormously practical/easy choice from the perspectives of both language familiarity for developers and fpr the workloads it was given.
I mean if it's a medical device it might not be great?
> If you pass a wrong object type to a function, that mistake is very easy to fix.
And with compile time checks you can avoid ever having to get to the point where you have to fix it.
Do you write tests for every third-party function that interacts with your code, so that it never fails in runtime after a version bump?
How do you guarantee that your own refactoring is exhaustively covered by the prior tests you've written for the old version?
The exact granularity is a debate that has gone on for a long time. Nowadays, people seem to prefer larger tests that cover more code in one go so as to avoid lots of mocking / stubbing. Super granular tests tend to be reserved for libraries with no internal state.
If you ignore the vast number of cases where it does, and cherry pick an extraordinarily non-representative example like the Linux kernel.
> This is just laughable. Clearly you have extremely little experience with Python.
Or you have extremely little experience with the use cases where it applies, extremely little knowledge of the ongoing effort by the Python developers to address it, and think that ignorant mocking is an argument.
But it seemed like a lot of people overlooked why it was loved. Most of us weren't wishing that our 2k+ line projects were all written in it.
Those little 20-500 line scripts were just so much more fun in Python than anything else though, and it showed up in survey after survey.
It's notable that you've written this in the past tense. These kinds of scripts don't seem to be a focus for Python any more.
In fact, I recently heard as justification for a new Python feature: "it helps us with our 100-million-line Python codebase". The language has become extremely complex as it prioritizes the needs of a few teams writing millions of lines over millions of developers each writing relatively few lines.
The reason is that the design team are all working on these large codebases and, understandably, mostly focused on what makes their own lives easier. I was hopeful that in the transition from a BDFL-based governance system to a Steering Council, we would see a larger variety of experience and opinions designing the language. Instead, I don't think there has ever been a single scientist, finance worker etc on the Steering Council - it's always software developers, almost always employees of large software companies.
The other obvious example of complexity is pattern matching. The feature is very complex, especially how it breaks established rules of the language. It also required fundamental changes to the interpreter to accommodate it: a rewrite of the parser from simple LL(1) to PEG.
The PEP 617[1] (the PEG change) doesn't mention pattern matching at all.
It's easy to get started with, but it's certainly not simple. Even small expressions like `a + b` or `a.b` in Python have very complex semantics.
The language is not complex.
(And there is a context here, with claims that the language has become complex ... your examples are obviously not relevant to that context.)
I love reading comments like this because it just makes me feel good about my job security.
You are aware that things like Youtube and Uber back ends all run on Python, right?
You are aware that things like your bank and insurance all run on COBOL, right?
Best of all with this approach, is that what is in the standard library suffices, no need to mess up with venvs, PYTHONPATH, or whatever is the current trend.
Then came Python 3 and it just never felt the same. PyQT projects faltered. Deps split. Strings broke.
However I started using Nim a few years back which has given me that Python 2 feeling – except with static types, better overall language design, meta-programming, and performance!
And except tooling, ecosystem, documentation and practical real-world application.
True the LSP is pretty bad, but Cursor makes up for a lot of that for me. Atlas dependency manager is great. Valgrind and TSan work perfectly. Gdb sorta works now. LLMS solve the docs issue.
The ecosystem is smaller, but I've been able to use it to successfully create tools for an IoT project with Nim running on an embedded esp32, a multi-threaded api server on linux processing images, and a web SPA with Karax in a couple of weeks of programming work. Oh and calling into Python for some datascience.
But then you find out that Nim people still rely on Nimpy, Rust people on PyO3, and even Julia people on PythonCall.
Yeah, no reason to use Python anymore, except by calling Python from your awesome language. The fact that these awesome Python packages are just wrappers over C/C++ doesn't change much - for some reason people using other languages are not willing to write and maintain wrappers of similar quality, including documentation.
> The fact that these awesome Python packages are just wrappers over C/C++ doesn't change much - for some reason people using other languages are not willing to write and maintain wrappers of similar quality, including documentation.
True, the older I get the more I’m amazed at communities of people who just build things and keep them running. It’s impressive. But hey calling Python from other languages is pretty sweet.
Though I recently found the Python libraries for automation stuff to be lacking. So I wrote my own library with Claude. I suspect that the friction of creating good wrappers in smaller languages is lowering. I still used Python for actual clicking with a super small rest api.
Just wanted to say thank you to those people. They cared enough about it to make it what it is today. An incredibly versatile and powerful, yet simple to understand, language.
I remember when an old school C++ guy from a company I worked for showed up at my desk, he had a csv file he output for some complicated case of latencies issues to analyze (that was like ~10 years ago), and he got a glimpse of what a Jupyter notebook / pandas / plotting lib can do. After 15 mins, when we got a clear understanding of what was going on, his face expression was like 'what the hell did just happen ?'. Felt like magic. Still do !
Jupiter Notebooks are a subset of what the experience felt like in those systems.
But in the documentary Guido says he mentored Mariatta?
Reading all the complaints about it recently is a bit surreal. Newer developers can't program without IDEs and "type systems." Packaging was kind of a mess, but feel like uv solved it. Neither issue really affected me, I used setup.py almost exclusively until moving to uv maybe six months ago.
It was always a glue/prototyping language—if you're using it for kernel drivers or multi-threaded Fibonacci servers or worse, that bad decision is yours (or that of your boss). ;-)
Education system in Lithuania had turbo pascal in high-school and mostly java and c/c++ in university and while I really loved pascal in high-school the switch to Java was so jarring - "people really enjoy this? maybe I should do something else" is what I thought during my first year of college. Luckily Python started to become really big online and was such a joy to use and be a part of the community it really cleared up this notion that programming sucks. To this day this experience has stuck to me when approaching any new activity - is there a Python here somewhere that would unsuck this?
I'm quite a polyglot these days and will write Java if needed but Python is still my daily driver and it just feels right. If I'm doing something 10 hours a day, I'd like to feel good while doing it and that's exactly what Python delivers.
He was patient, informative and above all extremely thoughtful in his stackoverflow responses. He was also very generous with his talks, all of which added to the rich ecosystem Python had back in the day.
"Armin Ronacher, Barry Warsaw, Benjamin Peterson, Brett Cannon, Drew Houston, Guido van Rossum, Jessica McKellar, Ken Manheimer, Lambert Meertens, Lisa Guo, Lisa Roach, Mariatta Wijaya, Paul Everitt, Patrice Lyons, Peter Wang, Robin Friedrich, Robert Kahn, Sjoerd Mullender, Steven Pemberton, Tim O'Reilly, Ton Roosendaal, and Travis Oliphant"
I believe he also coined the term duck typing.
e.g.:
Interview: Guido van Rossum Revisits Python’s Life in a New Documentary
https://thenewstack.io/guido-van-rossum-revisits-pythons-lif...
As someone who enjoys Python, I do wonder where Python will be in 10 years. I felt Guido's abdication came about due to fundamental challenges in the language (i.e., it's un-specifiable), which then were submerged by orthogonal, early 2020's social politics. I think some of the challenges have been addressed piecemeal (performance), but I worry that decisions makers may be losing their holistic understanding of the language.
if ma = re.search(needle, haystack):
print("ok, things are sane")
>>> if (x: str := re.sub("a", "b", "abc")):
File "<python-input-8>", line 1
if (x: str := re.sub("a", "b", "abc")):
^
SyntaxError: invalid syntax
I have always hated that comprehension variables leak for that same reason
1: https://github.com/python/cpython/blob/v3.13.7/Grammar/pytho...
Where it's useful, it tends to just save a line of code or work around other defects (for example, the regex example near the start of the PEP points to either a missing null-coalescing operator — which has been repeatedly proposed and rejected, BTW — or the library's fault in not using the "null object" pattern). It's used twice for example as far as I can tell in all of pip and its vendored dependencies (>100 kloc), and it isn't really doing much in either case (https://github.com/pypa/pip/blob/main/src/pip/_internal/meta... and https://github.com/pypa/pip/blob/main/src/pip/_internal/meta... ; it was also used in `typing_extensions` but that's been dropped as a dependency; there are some false-positives related to code that parses source code and some data formats).
But most importantly to me, it fundamentally violates the strong https://en.wikipedia.org/wiki/Command%E2%80%93query_separati... seen throughout the rest of the language.
Meanwhile, changes that I think might actually be good are nowadays routinely rejected with the same kinds of arguments that PEP 572 faced and survived.
It's unfortunate that Guido felt harassed and ended up leaving the project. It was overall better under his leadership. I wish I'd been paying attention at the time. I suspect that the people who created that feeling must have themselves felt betrayed.
import this
import antigravity >>> from __future__ import braces
File "<python-input-2>", line 1
from __future__ import braces
^^^^^^
SyntaxError: not a chance(https://news.ycombinator.com/item?id=45056377 https://news.ycombinator.com/item?id=45050548)
I was big into Python for over a decade of my career. I spoke at Pycon US and Canada. Wrote a few libraries. It had a great community.
I’ll always be grateful for it.
https://thenewstack.io/guido-van-rossum-revisits-pythons-lif...
And right now, transformers literally can transform any source to any target, so languages are not as valuable as they once used to be and any idioms in one language will eventually become outmoded thanks to the rapidly evolving way we work today.
Python will still be a favorite language for me, just not as much as it used to be.
If someone makes a new modern, tiny, self contained runtime (Bun like) for Python deployments, hit me up. I'll be happy to try it. That's something I've been wanting for a while.
I cannot say for the back and general usage, but at least for AI/ML definitely that moment happened between 2014 and 2016 when Theano, Caffe, and later 2015/16 Tensorflow cemented Python as the default for ML applications.
Processors and memory have gotten cheap enough were development speed matters more than compute cost.
Citation needed
https://survey.stackoverflow.co/2025/technology/#admired-and...
The more I use python, the more I hate it. For the inconsistencies and the short comings and the stuff it glosses over and absurdities like having the default queue be thread safe (sacrificing performance). My personal opinion is it is a garbage language.
Debugger >> language -- next most popular language manifesto slogan (i hope).
Why is this absurd? What are you writing where you need to scrape this sort of performance, but are perfectly fine with the overhead of interpreting bytecode at runtime, heap-allocating objects to represent floating-point numbers, not having homogeneous-type arrays that can store their elements contiguously in memory but which must indirect them, etc. etc. etc.?
Alternately, what world are you living in where everything a program does is CPU-bound and must be done efficiently? In this world, what led you to examine or try Python in the first place?
Java learned this lesson early.
Python has always suffered from the overrepresentation of talkers, and the recent draconian "leadership" that consists of the worst people have driven most developers away apart from some failing corporate projects.
This video tries to revive the spirit that existed in 2000 before many of those developers sold out Python to corporations and turned utterly evil.
This is so further from the truth that it hurts. It sounds like you're fairly new to python, and you probably only saw python in its third or fourth "wave" at which point there were alternatives, and you could ask why was python being promoted.
- The first wave was systems programming (replacing aging perl and bash scripts) - Second wave was web, with django, flask etc. This is also the rise of Ruby on Rails. This cemented python's stature. - Third wave was data analysis, fueled by numpy and pandas - Fourth wave is ML, building on top of third with tensorflow and pytorch.
You can argue we've in the fifth wave with LLMs, but it is mostly a 4.5 in my head.
If you dont remember waves 1 or 2, it is hard to get a perspective. But python was much much much big by wave 2.