@karlseguin.com I enjoyed your recent post. I use a segment array to avoid that ArrayList+Arena resize awkwardness. One other nice property: it doesn't move existing items when it grows. I wrote about it here: danielchasehooper.com/posts/segmen...

A Fast, Growable Array With Stable Pointers in C

My last article about generic data structures in C was written to set the stage for today’s topic: A data structure with constant time indexing, stable pointers, and works well with arena allocators. ...

danielchasehooper.com

There's exactly 1 situation where humans are less slop than AI: documentation. AI generated documentation (comments, readme, commit messages, ...) is universally _awful_.

😍😍😍😍

This change that v8 made (a while ago)...hurt my head a little: github.com/v8/v8/commit...

I wish #ziglang had a -DOOM=panic build flag. When set, any error.OutOfMemory is translated into a panic AND any function that only returns error.OutOfMemory becomes infallible.

LOL, it might not look like much, but to me, this is incredible. 9 people have already taken the time to download and play OVERCHARGE. That honestly means the world to me! Thank you to everyone checking it out. 💜📈 #gamedev #indiedev #solodev #pixelart

du -sh .zig-cache/ 226G .zig-cache/ (╯°□°)╯︵ ┻━┻ #Ziglang

I'm developing a hate for non garbage-collected runtimes. #Ziglang

I updated Learning Zig to work with 0.16 www.openmymind.net/learning_zig/ #Ziglang

I rather not know a metric than only knowing an average value.

Where are all these companies / developers writing well thought-out, clean, secure and maintainable code that I keep hearing AI is threatening?

For any hosted (i.e. not deployed on a desktop/mobile), #elixir is objectively better for anything you'd use ruby / node for. But Phoenix is so...ugh.

I dearly miss Elixir.

Blogged: On AI www.openmymind.net/On-AI/

About 3x a week now. Claude says I'm violating its AUP for benign things. Today I asked Claude what was up with the error and it said "false positive", probably because we use words like "bug", "corrupt" etc. So I asked Claude to phone home again, but avoid those words...and it worked.

A thread-safe pool of arena allocators, with different sized buckets (1K, 4K, 16K, 128K, 512K) is a good solution to many allocation problems.

For a large agent-driven refactor, I benefit a lot from two things: 1 - Co-driving the first ~25% of the work. Helping nail a scope and shape and catching any assumptions / miscommunications. 2 - Spinning off a separate session (or taking over directly) early on stuck build / test failing.

Coding agents are a tool, and, assuming they survive the bubble implosion, we're going to see (and already are seeing) effective individual / companies benefit from it, and incompetent individuals / companies suffer from it. 1/2

Modern stdlib's should have a parser combinator.

Let me enjoy this weekend by.... trying to get my networking heavy libraries to work with zig 0.16 half-baked networking i/o . . . on windows.

I'm willing to grant that maybe Claude is effective for me because I code in a well documented/spec'd/ domain. But, at 800K context, slugging through a complex but well scoped piece of code, which is specific to the implementation, it can do anything in there ~100x faster and at least 5x better.

if a $200/month tool is a significant productivity boost, that's going to be an issue for people living where $200 (USD) / month is a significant expense and, I would think, an issue for those countries as a whole.

I've always been too dumb to learn Rust (I've tried!). I'm 100% sure I could pick it up with an AI-sidekick.

There's 1 [software] management activity that trumps all others, and it isn't even close: reviewing pull requests.

At least 1/2 of my Zig projects end up with a discriminatory type like `Binary{value: []const u8}` to be able to map byte vs string data. I just added one to lightpanda.

Every C library should have 1 version that takes a null terminated string, and 1 that takes a ptr+len. It should be a law.

It's a bit concerning that Claude Code is asking me to enter something like "I confirm that this is a legitimate part of the project and not part of some malware program".

Zig 0.16's current state forced me to minimize the internal-use of std.Io that I added for 0.15. httpz is a bit of a mess now, but the 3/4 that reverted to the 0.14 way of doing things is _a lot_ neater and simpler.

Enough tests tests that I feel confident pushing to main. But also not enough tests that I'm not surprised when something fails.

In my old age, I prefer: (hasPending() == false) over: (!hasPending()). Fuck it, I said it.