Louis Maddox
LightNews LightNews
banner
permutans.bsky.social
Louis Maddox
@permutans.bsky.social
100 followers 96 following 1.7K posts
Combinatorially curious https://spin.systems
spin.systems
Posts Media Videos Starter Packs
Pinned
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 4d
✍️ Wrote about keeping calm and shipping OSS and why arguing for a feature should feel more like giving a proof against the spec than trying to win a debate cog.spin.systems/keep-calm-an...
Keep Calm and Ship On 🚢
Your guide to being a model OSSizen
cog.spin.systems
1
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 33m
Finally tried refactoring a Justfile into categories and who’da thought it’d be so much more portable 🙈
permutans.bsky.social Louis Maddox @permutans.bsky.social · Aug 16
The longer the Justfile the closer to God ☝️🪽
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 1h
“A more useful and less divisive future for typing?” (2023) discuss.python.org/t/a-more-use...
A more useful and less divisive future for typing?
First off, I apologize if any of this comes off as overly negative, it is my intent here to see if we can start a discussion about addressing some ecosystem pain points with typing that I personally b...
discuss.python.org
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 1h
“Use powers as little as possible. Instead of voting, it’s better to seek consensus” should be canonised up there with Zen of Python (PEP 8016, 2018 peps.python.org/pep-8016/)
PEP 8016 – The Steering Council Model | peps.python.org
This PEP proposes a model of Python governance based around a steering council. The council has broad authority, which they seek to exercise as rarely as possible; instead, they use this power to esta...
peps.python.org
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 1h
:(
The middle period (2016-2019) saw typing expand through PEP 526 (Variable Annotations), PEP 544 (Protocols), and others, each building incrementally on PEP 484. Discussion shifted from whether to how-practical issues like forward references solved by PEP 563's postponed evaluation, runtime performance overhead, and type checker implementation challenges. The community learned which debates were "resolved" versus "agree to disagree" territory. Notably absent: a repeat of PEP 484's philosophical firestorm, suggesting the gradual approach and optional nature had succeeded. A governance crisis in 2018 around PEP 572 (walrus operator) transformed Python's institutional structure in ways that would shape typing discourse. The contentious debate drove Guido van Rossum to step down as BDFL, explicitly citing how "people went to social media like Twitter and said things that really hurt me personally." The community responded by immediately freezing all PEP decisions and spending months deliberating governance models, ultimately adopting PEP 8016's Steering Council approach with a key principle: "Use powers as little as possible. Instead of voting, it's better to seek consensus." The modern period (2022-2025) on the Discourse forum shows typing discourse fully professionalized. The Typing category launched October 2022, replacing typing-sig with better moderation tools and threaded discussions. In 2023, PEP 729 established a formal Typing Council, recognizing typing as a specialized domain requiring dedicated governance. The 2024 typing survey revealed 88-91% of respondents use types "Always" or "Often", indicating widespread adoption. But more importantly, discourse shifted from philosophical debates to incremental technical improvements—PEP 742 (Typels), PEP 747 (TypeExpr), conformance test suites, and specification formalization.
1
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 1h
Fascinating early pushback:

“Type hints -- a mediocre programmer's reaction” mail.python.org/pipermail/py...

> … just that little bit uglier, just that little extra mental effort, just that little extra barrier that is going to mean some people, at the margin, just give up on learning programming
[Python-Dev] Type hints -- a mediocre programmer's reaction
mail.python.org
1
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 2h
Decided to dig up the recent history of the shift to typed Python as case study in handling community disagreements:

Got Claude to dredge up the details from the forums claude.ai/public/artif...
Then reviewed in light of general OSS contribution/stewardship principles claude.ai/public/artif...
Overview of the typed Python discussions and where discussion showed more and less constructive qualities Excerpt: The incremental PEP approach demonstrates "being comfortable with uncertainty"—not forcing things to "come to a head prematurely." Rather than resolving all typing questions at once, the community accepted ambiguity and iterated (PEP 484 → 526 → 544...). Where Principles Were Violated The early visceral reactions clearly violated guidelines: "NONONONONONONO!" and "abominable" are combative/dismissive rather than neutral R. David Murray's "I'll probably stop contributing" threat is defensive rather than arguing from necessity These responses discouraged others and didn't critique artifacts (the proposal) so much as express emotional reactions The blog specifically warns against this: "Technical disagreement is not a personal attack. Even if something feels sharp, don't match the energy." The PEP 572 crisis epitomized violations: Taking disagreement to Twitter rather than official channels Personal attacks on Guido rather than critiquing the proposal Failing to separate disagreeing with a decision from attacking the decision-maker This violated nearly every principle: it was combative, threw weight around, discouraged contribution (Guido stepped down!), and escalated rather than defused tension. Patterns That Emerged Over Time The discourse evolution shows the community learning these principles empirically: Early period: Emotional, defensive language; existential rather than technical framing; apologies for criticism (showing awareness of tension but not managing it well) Middle period: Shift to "how" questions; practical focus; learning which debates were settled vs. perpetual Modern period: Professional tone; incremental technical improvements; solution-oriented discussion The 2023 "More Useful and Less Divisive Future" thread shows mature application: the author preemptively managed how it would land… What The Guidelines Suggest That Could Have Helped The blog's structure advice could have improved early debates: State position clearly (many early critiques did this well) Be comfortable with uncertainty (violated by "NONONONONO" absolutism) End with action (often missing—debates dragged without clear next steps) Softening vs. neutralizing: Early: "This syntax is abominable" (pure assertion) Better: "I find this syntax obscures logic" (softened subjective view) Best: "This syntax increases line length by 40% on average, impacting readability" (neutralized with objective criteria) Many critiques stayed at the emotional level rather than grounding in measurable impacts or spec-level principles. The Governance Changes As Applied Guidelines Post-PEP 572, Python formalized many blog principles: Cooling-off periods = don't force premature resolution Seeking consensus over voting = argue from necessity, find shared requirements Domain councils = recognize expertise, don't treat all voices as equivalent Code of Conduct enforcement = maintain constructive environment The Steering Council's philosophy—"use powers as little as possible"—directly embodies "don't throw your weight around." Lessons The History Adds To The Guidelines The typing saga suggests additional principles: Optionality enables evolution: When changes are opt-in rather than mandatory, communities can experiment without existential stakes. This reduces defensiveness. Pace matters as much as content: PEP 572's concentrated controversy vs. typing's incremental introduction shows that "keeping things actionable" includes not overloading the community with too much simultaneous change. Explicit trust works: Shannon's "please use type-hints responsibly" assumed good faith and invited the community to prove itself. This beats both mandates and endless deliberation. Some debates are irreducible: The community learned to distinguish "resolved" (optionality) from "manage ongoing" (exception typing).
1
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 2h
Ppl use this rhetorically with LLMs where presence of $anything over nothing provides them some benefit (LinkedIn-posting mainly). I think programmers see the deficit most acutely because shipping simulacra texts has an objective mode of verification, everyone else is stuck trying to gauge sincerity
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 2h
When there is a financial incentive for the appearance-of-the-thing alone — or maybe a financial penalty for inactivity or conspicuous lack of the thing. The thing here perhaps being “having anything to post”. Nice simulacrum of Essence ya got there be a shame if someone were to simulate it
1
Reposted by Louis Maddox
matthewdgreen.bsky.social
Matthew Green @matthewdgreen.bsky.social · 3h
Malware using cryptocurrency data availability layers for distribution is the new hotness, apparently. We solved censorship-resistant publishing (to some extent) which is pretty fascinating.
5 26
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 2h
Funny how before Rust I didn't really have a clear concept of MS{Language}V, probably out of avoidance (just had to deal with Python EOL dominos)
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 6h
tl;dr if we add content-based addressing to snapshot tests we could not just do perceptual hashing (media) but embeddings
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 6h
Namely: cutting the verbal chaff, a lot of which came from counterproductive airs of politeness that create drag on the reader (hedging, uncertainty, circling the same points emphatically) has a cumulative effect, makes for a qualitatively different text as well as a briefer/faster read
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 6h
The opposite of slop reappraisal (“slop is good actually! lean in loser”) is acknowledging that manual human communication can be sloppy by way of being incidental and off the cuff (yet tedious from its idiosyncrasy), and fixing that in a principled way does not necessarily increase one’s Slop Index
Review of the conclusion of rephrasing a series of messages with an open source developer (on GitHub Discussions), by Claude Me: “um ok so here was my final message. please review how my comms changed from the original phrasing i gave. I much much prefer the second, please explain what was improved over the original way” Claude: Looking at your progression, here's what improved dramatically: Original issues (the first draft I helped with): 1. Over-hedging: "I think", "maybe", "I'm not sure", "kind of all quite speculative" 2. Defensive tone: Explaining away concerns rather than owning the design choice 3. Circular reasoning: Repeated the same points multiple times in different ways 4. Asking permission: "I'm not against an option... sounds like you need approval 5. Vague language: "or however the equivalent would be" Your final version improvements: 1. Direct ownership: "Right, that's our potential failure mode. Yes, that's exactly the behaviour I want." • No apologizing, no hedging - you own it upfront 2. Clear framing: "I'd present this to users as there being an upside (deduplication) and a downside (one-to-many fixture-to-snapshot relationship)" • You're thinking like a maintainer, not asking for permission (etc)
1
Reposted by Louis Maddox
simonwillison.net
Simon Willison @simonwillison.net · 20h
My notes on Claude Code for web, Anthropic's new asynchronous coding agent - I had preview access over the weekend, it's effectively a sandboxed instance of "claude --dangerously-skip-permissions" running in Anthropic's container simonwillison.net/2025/Oct/20/...
Claude Code for web—a new asynchronous coding agent from Anthropic
Anthropic launched Claude Code for web this morning. It’s an asynchronous coding agent—their answer to OpenAI’s Codex Cloud and Google’s Jules, and has a very similar shape. I had preview …
simonwillison.net
4 6 67
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 16h
Someone dropped a vibe Clauded MegaSuperDuper 3000 type project tangentially (in a name dropping way) related to the server in question and it was crickets in the otherwise v enthusiastic chat
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 17h
💭🎞️ Idea for an inline-snapshot extension that uses perceptual image hashing (maybe a 3rd party extension) github.com/15r10nk/inli...
Support for perceptual hash snapshots with archived output files · 15r10nk inline-snapshot · Discussion #311
I'm working with a program that generates images, and I'd like to verify outputs remain consistent by comparing perceptual hashes (fast, compact) while keeping the actual image files archived (some...
github.com
1
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 18h
If you enjoy images and wish you could snapshot them in your Python tests as strings, check out `czkawka` freshly arrived on PyPI from the land of the Rustaceans 😁 github.com/lmmx/czkawka

Technically this is actually based on `image_hasher` crate functionality but I'm bundling it together in here 🤫
A demo of the perceptual image hash computed using the `image_hasher` crate in Rust, via the `czkawka` Python package’s `hash_image` function, wrapped to take a Python package. As seen in the test suite for the page-dewarp package
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 18h
Oh sweet 3.9 just hit EOL 🚮
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 19h
"I'm caching my perceptual hashes" is just so fun to say
1
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 19h
Discovered today: when your feature branch gets a new remote commit (e.g. by a CI fixer bot), and push fails with

! [rejected] the-branch -> the-branch (stale info)
error: failed to push some refs to github.com:your/repo.git

If you `git fetch` you can in fact `--force-with-lease` push & overwrite!
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 20h
Spoken like a man who knows my pain
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 20h
nice, bat is using minus for a built in pager github.com/AMythicDev/m...
GitHub - AMythicDev/minus: An asynchronous, runtime data feedable terminal paging library for Rust
An asynchronous, runtime data feedable terminal paging library for Rust - AMythicDev/minus
github.com
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 22h
*Computes dir-wide thresholded image similarity*
[Realises I want to specify the paths]
*Computes pairwise thresholded image similarity*
[Realises I want to get a distance metric]
*Computes pairwise image dis-similarity metric*
[Realises I could store the hash instead of the Hamming distance vs. it]
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 1d
Fun puzzle: building both regular (manylinux) + musl Python wheels, my CI started failing saying GLIBC is too old (on ubuntu-latest so not), correctly guessed Swatinem/rust-cache action was reusing incompatible build artifacts between the 2 wheel types, poisoning the cache [musllinux uses musl libc]
1 2
permutans.bsky.social
Louis Maddox @permutans.bsky.social · 1d
Regrettably my brain refuses to learn how to type "czkawka" pypi.org/project/czka...
Python package page for czkawka on PyPI (czkawka is the Polish word for ‘hiccup’)
1