Comprehension debt: A ticking time bomb of LLM-generated code
Article URL: https://codemanship.wordpress.com/2025/09/30/comprehension-debt-the-ticking-time-bomb-of-llm-generated-code/

Comments URL: https://news.ycombinator.com/item?id=45423917

Points: 530

# Comments: 334 ⌘ Read more

Can VS Code shortcuts compete with Vim? ⌘ Read more

I finally solved the loading issue in my WIP reader, TwtStrm (and apologies again to anyone that got spammed while I was diagnosing the issue).

After another round of coding this weekend, I’m happy to report that it now renders all the twts (with markdown parsing), complete with localstorage and server-based file caching.

Pretty happy with my zs-blog-template starter kit for creating and maintaining your own blog using zs 👌 Demo of what the starter kit looks like here – Basic features include:

• Clean layout & typography
  
• Chroma code highlighting (aligned to your site palette)
  
• Accessible copy-code button
  
• “On this page” collapsible TOC
  
• RSS, sitemap, robots
  
• Archives, tags, tag cloud
  
• Draft support (hidden from lists/feeds)
  
• Open Graph (OG) & Twitter card meta (default image + per-post overrides)
  
• Ready-to-use 404 page
  

As well as custom routes (redirects, rewrites, etc) to support canonical URLs or redirecting old URLs as well as new zs external command capability itself that now lets you do things like:

$ zs newpost

to help kick-start the creation of a new post with all the right “stuff”™ ready to go and then pop open your $EEDITOR 🤞

#awesome #zs

@prologic@twtxt.net need to work on the CSS. For example, the tags are too big, the code blocks (and the inline ones) are too small, the single posts have no date (intended?), and so on. It’s an alpha start!

Put another way, what you are proposing/pushing for requires hundreds of lines of code to change across a half dozen or so clients and lots of breaking changes, not to mention unknowns.

What I want us to do is make only a few half dozen or so lines of code changes to our clients and minimize the breaking changes and unknowns.

Coding a SHA2 Length Extension Attack - Computerphile ⌘ Read more

GitHub Copilot gets smarter at finding your code: Inside our new embedding model 
Learn about a new Copilot embedding model that makes code search in VS Code faster, lighter on memory, and far more accurate.

The post GitHub Copilot gets smarter at finding your code: Inside our new embedding model appeared first on The GitHub Blog. ⌘ Read more

@bender@twtxt.net Thanks for asking!

So, I’ve been working on 2 main twtxt-related projects.

The first is small Node / express application that serves up a twtxt file while allowing its owner to add twts to it (or edit it outright), and I’ve been testing it on my site since the night I made that post. It’s still very much an MVP, and I’ve been intermittently adding features, improving security, and streamlining the code, with an eye to release it after I get an MVP done of project #2 (the reader).

But that’s where I’ve been struggling. The idea seems simple enough - another Node / express app (this one with a Vite-powered front-end) that reads a public twtxt file, parses the “follow” list, grabs (and parses) those twtxt files, and then creates a river of twts out of the result. The pieces work fine in seclusion (and with dummy data), but I keep running into weird issues when reading real-live twtxt files, so some twts come through, while others get lost in the ether. I’ll figure it out eventually, but for now, I’ve been spending far more time than I anticipated just trying to get it to work end-to-end.

On top of it, the 2 projects wound up turning into 4 (so far), as I’ve been spinning out little libraries to use across both apps (like https://jsr.io/@itsericwoodward/fluent-dom-esm, and a forthcoming twtxt helper library).

In the end, I’m hoping to have project 1 (the editor) into beta by the end of October, and project 2 (the reader) into beta sometime after that, but we’ll see.

I hope this has satisfied your curiosity, but if you’d like to know more, please reach out!

Gartner positions GitHub as a Leader in the 2025 Magic Quadrant for AI Code Assistants for the second year in a row
Our commitment is to empower every developer and stay true to our north star by building an open, secure, and AI-powered platform that defines the future of software development.

The post [Gartner positions GitHub as a Leader in the 2025 Magic Quadrant for AI Code Assistants for the second yea … ⌘ Read more

The big QR code canine, has been one of my favourites - because even after a few months, I still find the pose really cute. Always thought a chibi version is a necessary addition and now I finally drew it.

Doing a bit of 2018 Advent of Code now to relax. 🎄

Working on a project that does Augmented Reality and computer vision object detection and QR code and image recognition inside a Web application. Pretty neat what can be done today with a few thousand lines of JavaScript.

@arne@uplegger.eu Hm, noch nie gemacht. 🤔 Machst du das von Hand oder mit Code?

@kat@yarn.girlonthemoon.xyz @kat@yarn.girlonthemoon.xyz Pretty sure I have many more mentions in the database than the one and only one I see hmmm 🤔 – I’ll have a look at the code when I can and the SQL query it’s using

Erlang Solutions: ElixirConf US 2025: Highlights from My First ElixirConf
Joining conferences is one of the best perks of working as a Developer at Erlang Solutions. Despite having attended multiple Code BEAM conferences in Europe, ElixirConf US 2025 was my first. The conference had 3 tracks, filled with talks from 45+ speakers and 400+ attendees, both in-person and virtual.

ElixirConf is one of the great occasions to connect with other Elixir ent … ⌘ Read more

@movq@www.uninformativ.de this seems like a bit of an overkill, that would also harm modding and power users - who often need to see the exact implementation of new features and benefit from the ability to pull up the history of code changes, in their browser. Sure they could clone the repo and do that locally, but if it has dependencies, they’d also have to clone those, to see how those get updated and it’d soon be a mess.

@movq@www.uninformativ.de Holy shit, that’s insane! :-D I tried it, but i’m absolutely terrible at these type of games. I’m having trouble with the keys to move around. Maybe after ages I would pick it up and it becomes natural. I just was never a real gamer.

I will definitely try to read through the code, though! This looks sick. 8-)

@eric@itsericwoodward.com Hmm, the images are all 404ing. Also, I reckon that lots of code blocks are broken, too.

Haha, fun! I browsed your gopher hole a little bit. I noticed some entries are fully justified (formatting), while others are not. I didn’t notice a pattern, though it makes sense not to use justification on entries with code. Yet, some prose entries are, and some are not. A mystery. :-)

@bender@twtxt.net The address is/was correct but probably got mangled by the Markdown renderer. Let’s try again in a code block:

gopher://uninformativ.de/0/phlog/2025/2025-09/2025-09-03--roophloch.txt

Mathieu Pasquet: slixmpp v1.11
This new version includes a few new XEP plugins as well as fixes, notably
for some leftover issues in our rust JID code, as well as one for a bug that
caused issues in Home Assistant.

Thanks to everyone who contributed with code, issues, suggestions, and reviews!

Nicoco put in a lot of work in order to get all possible wheels built in CI. We now have manylinux and musl builds of everything doable within codeberg,
published to the codeberg pypi repo, and published on pypi. … ⌘ Read more

“But all your stuff is MIT licensed! They are allowed to do that!”

Haha. As if they would care. They crawl everything they get their hands on.

Besides, that’s not true, the license states that the copyright notice must be retained. “AI” breaks that. They incorporate my code and my articles in their product and make it appear as if it was their work.

** To the surprise of literally no one, I’m working on implementing a programming language all my own **
Inspired by conversation at a recent Future of Coding event, I decided I’d write up a little something about the programming language I’ve been working on (for what feels like forever) before I’ve gotten it to a totally shareable state. I have a working interpreter that I’m pretty pleased with, but I don’t yet have an interact … ⌘ Read more

@kat@yarn.girlonthemoon.xyz On the one hand, all these programs have a very long history and the technology behind manpages is actually very powerful – you can use it to write books:

https://www.troff.org/pubs.html

I have two books from that list, for example “The UNIX programming environment”:

It’s a bit older, of course, but it looks and feels like a normal book, and it uses the same tech as manpages – which I think is really cool. 😎

It’s comparable to LaTeX (just harder/different to use) but much faster than LaTeX. You can also do stuff like render manpages as a PDF (man -Tpdf cp >cp.pdf) or as an HTML file (man -Thtml cp >cp.html). I think I once made slides for a talk this way.

On the other hand, traditional manpages (i.e., ones that are not written in mandoc) do not use semantic markup. They literally say, “this text is bold, that text over here is italics”, and so on.

So when you run man foo, it has no other choice but to show it in black, white, bold, underline – showing it in color would be wrong, because that’s not what the source code of that manpage says.

Colorizing them is a hack, to be honest. You’re not meant to do this. (The devs actually broke this by accident recently. They themselves aren’t really aware that people use colors.)

If mandoc and semantic markup was more commonly used, I think it would be easier to convince the devs to add proper customizable colors.

@kat@yarn.girlonthemoon.xyz @movq@www.uninformativ.de Sorry, I neither finished it nor in time. :-( That’s as good as it’s gonna get for the moment: https://git.isobeef.org/lyse/gelbariab/-/tree/master/rss-proxys?ref_type=heads

The README should hopefully provide a crude introduction. The example configuration file is documented fairly well, I believe (but maybe not). You probably still have to consult and maybe also modify the source code to fit your needs.

Let me know if you run into issues, have questions, wishes etc.

In 1996, they came up with the X11 “SECURITY” extension:

https://www.reddit.com/r/linux/comments/4w548u/what_is_up_with_the_x11_security_extension/

This is what could have (eventually) solved the security issues that we’re currently seeing with X11. Those issues are cited as one of the reasons for switching to Wayland.

That extension never took off. The person on reddit wonders why – I think it’s simple: Containers and sandboxes weren’t a thing in 1996. It hardly mattered if X11 was “insecure”. If you could run an X11 client, you probably already had access to the machine and could just do all kinds of other nasty things.

Today, sandboxing is a thing. Today, this matters.

I’ve heard so many times that “X11 is beyond fixable, it’s hopeless.” I don’t believe that. I believe that these problems are solveable with X11 and some devs have said “yeah, we could have kept working on it”. It’s that people don’t want to do it:

Why not extend the X server?

Because for the first time we have a realistic chance of not having to do that.

https://wayland.freedesktop.org/faq.html

I’m not in a position to judge the devs. Maybe the X.Org code really is so bad that you want to run away, screaming in horror. I don’t know.

But all this was a choice. I don’t buy the argument that we never would have gotten rid of things like core fonts.

All the toolkits and programs had to be ported to Wayland. A huge, still unfinished effort. If that was an acceptable thing to do, then it would have been acceptable to make an “X12” that keeps all the good things about X11, remains compatible where feasible, eliminates the problems, and requires some clients to be adjusted. (You could have still made “X11X12” like “XWayland” for actual legacy programs.)

Since Fastly acquired and recently shut down glitch.com, some of my ancient webapps are no longer available, nor do I have any plans to make them available again - all had either zero, or very few monthly visits, used outdated libraries and would be a waste of money, to continue hosting and updating elsewhere.

All art archives remain unaffected and all projects shut down before 2025, were already permanently deleted, but if there’s someone out there, still relying on the recently discontinued projects, somehow - you can reach out and request their source code.

These requests will only be honoured, until the end of this year, when we plan to permanently delete, all of this data (both webapps and files only hosted on Amazons CDN).

Canine out °_°

Heck yeah, that’s damn cool: Reading QR codes without a computer! https://qr.blinry.org/

@lyse@lyse.isobeef.org “Advanced”, well, probably more “mature”. There aren’t a ton of crazy features and that icon thing is the largest code addition in the last 10 years. %)

Speaking of OS/2 … I just realized that Windows 3.x didn’t have icons, either. If I’m not mistaken, this only got added in Windows 95. In other words, OS/2 had this feature before Windows did, because at least OS/2 2.1 from 1993 had icons. Who would have thunk.

(Now I kind of want to know which system really introduced this feature.)

@kat@yarn.girlonthemoon.xyz NVM i stole other peoples code to make a dictionary lookup script https://bytes.4-walls.net/kat/dotfiles/src/branch/main/config/.local/bin/dict

@lyse@lyse.isobeef.org @kat@yarn.girlonthemoon.xyz I spent so much time in the past figuring out if something is a dict or a list in YAML, for example.

What are the types in this example?

items:
- part_no:   A4786
  descrip:   Water Bucket (Filled)
  price:     1.47
  quantity:  4
- part_no:   E1628
  descrip:   High Heeled "Ruby" Slippers
  size:      8
  price:     133.7
  quantity:  1

items is a dict containing … a list of two other dicts? Right?

It is quite hard for me to grasp the structure of YAML docs. 😢

The big advantage of YAML (and JSON and TOML) is that it’s much easier to write code for those formats, than it is with XML. json.loads() and you’re done.

@movq@www.uninformativ.de Yeah, it’s a shitshow. MS overconfirms all my prejudices constantly.

Ignoring e-mail after lunch works great, though. :-)

Our timetracking is offline for over a week because of reasons. The responsible bunglers are falling by the skin of their teeth:

1. The error message neither includes the timeframe nor a link to an announcement article.
   
2. The HTML page needs to download JS in order to display the fucking error message.
   
3. Proper HTTP status codes are clearly only for big losers.
   
4. Despite being down, heaps of resources are still fetched.
   

I find it really fascinating how one can screw up on so many levels. This is developed inhouse, I’m just so glad that we’re not a software engineering company. Oh wait. How embarrassing.

@prologic@twtxt.net Yeah, this really could use a proper definition or a “manifest”. 😅 Many of these ideas are not very wide spread. And I haven’t come across similar projects in all these years.

Let’s take the farbfeld image format as an example again. I think this captures the “spirit” quite well, because this isn’t even about code.

This is the entire farbfeld spec:

farbfeld is a lossless image format which is easy to parse, pipe and compress. It has the following format:

╔════════╤═════════════════════════════════════════════════════════╗
║ Bytes  │ Description                                             ║
╠════════╪═════════════════════════════════════════════════════════╣
║ 8      │ "farbfeld" magic value                                  ║
╟────────┼─────────────────────────────────────────────────────────╢
║ 4      │ 32-Bit BE unsigned integer (width)                      ║
╟────────┼─────────────────────────────────────────────────────────╢
║ 4      │ 32-Bit BE unsigned integer (height)                     ║
╟────────┼─────────────────────────────────────────────────────────╢
║ [2222] │ 4x16-Bit BE unsigned integers [RGBA] / pixel, row-major ║
╚════════╧═════════════════════════════════════════════════════════╝

The RGB-data should be sRGB for best interoperability and not alpha-premultiplied.

(Now, I don’t know if your screen reader can work with this. Let me know if it doesn’t.)

I think these are some of the properties worth mentioning:

• The spec is extremely short. You can read this in under a minute and fully understand it. That alone is gold.
  
• There are no “knobs”: It’s just a single version, it’s not like there’s also an 8-bit color depth version and one for 16-bit and one for extra large images and one that supports layers and so on. This makes it much easier to implement a fully compliant program.
  
• Despite being so simple, it’s useful. I’ve used it in various programs, like my window manager, my status bars, some toy programs like “tuxeyes” (an Xeyes variant), or Advent of Code.
  
• The format does not include compression because it doesn’t need to. Just use something like bzip2 to get file sizes similar to PNG.
  
• It doesn’t cover every use case under the sun, but it does cover the most important ones (imho). They have discussed using something other than RGBA and decided it’s not worth the trouble.
  
• They refrained from adding extra baggage like metadata. It would have needlessly complicated things.

@prologic@twtxt.net Hm, I wouldn’t say that. Go code could fall into that category as well.

Maybe this topic could use a blog post / article, that explains what it’s about. I’m finding it hard to really define what “suckless-like software” is. 🤔 (Their own philosophy focuses too much on elitism, if you ask me.)

In all fairness, GOG says that Forsaken is only supported on Ubuntu 16.04 – not current Arch Linux. If you ask me, this just goes to show that Linux is not a good platform for proprietary binary software.

Is it free software, do you have the source code? Then you’re good to go, things can be patched/updated (that can still be a lot of work). But proprietary binary blobs? Very bad idea.

It annoys me when I clone a git repository A in order to build and self-host some software, only to realize later that I also needed to clone repos B, C and D. I’m not saying that’s a bad thing–logical separation of code between, say, a client and a server is very handy–but some projects do not communicate very well when you need multiple tools to get it running independently.

I did a “lecture”/“workshop” about this at work today. 16-bit DOS, real mode. 💾 Pretty cool and the audience (devs and sysadmins) seemed quite interested. 🥳

• People used the Intel docs to figure out the instruction encodings.
  
• Then they wrote a little DOS program that exits with a return code and they used uhex in DOSBox to do that. Yes, we wrote a COM file manually, no Assembler involved. (Many of them had never used DOS before.)
  
• DEBUG from FreeDOS was used to single-step through the program, showing what it does.
  
• This gets tedious rather quickly, so we switched to SVED from SvarDOS for writing the rest of the program in Assembly language. nasm worked great for us.
  
• At the end, we switched to BIOS calls instead of DOS syscalls to demonstrate that the same binary COM file works on another OS. Also a good opportunity to talk about bootloaders a little bit.
  
• (I think they even understood the basics of segmentation in the end.)
  

The 8086 / 16-bit real-mode DOS is a great platform to explain a lot of the fundamentals without having to deal with OS semantics or executable file formats.

Now that was a lot of fun. 🥳 It’s very rare that we do something like this, sadly. I love doing this kind of low-level stuff.

@lyse@lyse.isobeef.org lol – I explicitly kept them in there so that the code is easier to understand for non-Rust people 🤪😂

Saw this on Mastodon:

https://racingbunny.com/@mookie/114718466149264471

18 rules of Software Engineering

1. You will regret complexity when on-call
   
2. Stop falling in love with your own code
   
3. Everything is a trade-off. There’s no “best” 3. Every line of code you write is a liability 4. Document your decisions and designs
   
4. Everyone hates code they didn’t write
   
5. Don’t use unnecessary dependencies
   
6. Coding standards prevent arguments
   
7. Write meaningful commit messages
   
8. Don’t ever stop learning new things
   
9. Code reviews spread knowledge
   
10. Always build for maintainability
    
11. Ask for help when you’re stuck
    
12. Fix root causes, not symptoms
    
13. Software is never completed
    
14. Estimates are not promises
    
15. Ship early, iterate often
    
16. Keep. It. Simple.
    

Solid list, even though 14 is up for debate in my opinion: Software can be completed. You have a use case / problem, you solve that problem, done. Your software is completed now. There might still be bugs and they should be fixed – but this doesn’t “add” to the program. Don’t use “software is never done” as an excuse to keep adding and adding stuff to your code.

AI-assisted coding for teams that can’t get away with vibes
Comments ⌘ Read more

Claude Code: Game Changer or Just Hype? ⌘ Read more

Tell HN: Help restore the tax deduction for software dev in the US (Section 174)
Companies building software in the US were hit hard a few years ago when the tax code stopped allowing deduction of software dev expenses. Now they have to be amortized over several years.

HN has had many discussions about this, including The time bomb in the tax code that’s fueling mass tech layoffs - https://news.ycombinator.com/item?id=44180533 - (927 comments) a few days ago. Other thr … ⌘ Read more

GitHub for Beginners: Code review and refactoring with GitHub Copilot
Learn how to use GitHub Copilot to help review and polish your code.

The post GitHub for Beginners: Code review and refactoring with GitHub Copilot appeared first on The GitHub Blog. ⌘ Read more

Field Notes From Shipping Real Code With Claude
Comments ⌘ Read more

What next after vibe coding
One interesting possible future of the emergence of “vibe coding” as common terminology is the possibility to position an alternative.

“Real coding”?

Future think pieces:

• ‘You can get yourself up and running quickly with “vibe coding” but when you get traction you’re going to want have people doing “real coding”’

• ‘switching from vibe coding to real coding will typically cost you X% of you initial development, so don’t put off switching too late’

[Comments](https://lobste.rs/s/m … ⌘ Read more

Context7: Up-to-date documentation for LLMs and AI code editors
Comments ⌘ Read more

105 vibe-coded tools
Comments ⌘ Read more

That’s the code, it’s surprisingly simple: https://movq.de/v/81dd5649be/

Assigning and completing issues with coding agent in GitHub Copilot
Have you tried the new coding agent in GitHub Copilot? Here’s how developers are using it to work more efficiently.

The post Assigning and completing issues with coding agent in GitHub Copilot appeared first on The GitHub Blog. ⌘ Read more