https://t.co/1m1vOuuv91 Early days yet, but I finally solved the issue of how to represent the userspace memory buffer that the kernel also rights into, which confused the hell out of me last year.

On the off chance the X algorith annoints this post, my answers to "Y NOT USE ANOTHER LIB LOL": - no libc - no std - no cutesy java builder pattern bullshit

It's year later and I'm working on a rust io_uring library again - and wow everything is so much clearer to me now. I understand the Zig code better, I understand Rust better, and I understand Linux better. (writing this before I implement the tests and they crash my machine)

Rust really is fantastic if you ignore about 30% of the language and 90% of the popular crates.

It would be NZ software engineering coded but they don't make enough money usually, and have to be content with copying what google does + 100s of micro services for their startup with 2 whale customers.

Australian software engineering coded

There is basically no digital note taking system in existence that is as ergonomic as just writing things with different coloured pens. No way I could recreate this in html/css, or typst, or anything else.

Why does Zig use vtables everywhere instead of generic (comptime) parameters? Ie, std.mem.Allocator. My suspicion is that they are Go and C people, and find the concept of constraining types at compile time like in Ocaml, Rust, Haskell, C# etc etc alien and bewildering.

I asked a paid LLM model: "can you think of a better name than add_buff in src/linux.zig? relevant man page is io_uring_buf_ring_add, to save you trawling through zig docs/source" After 45 seconds, and paragraphs of thoughts, it recommended "addBuffer"

The more I reflect on my recent agentic coding adventures, the more I realise I would have been better off printing a hard copy of the relevant man pages and reading them for an hour rather than consulting the stochastic oracle. Still see a lot of value in AI code review though.

Actually, maybe I am an artisanal craftsman. I feel completely alienated from the code when a machine writes it. Like I'm doing legacy software work.

I don't like AI writing code for me. It's not just that I'm an ArTiSaNaL sOfTwArE cRaFtSmAn, it's that the act of actually editing it builds a mental model. I lose my connection to it. I very much like it for making suggestions and reading it though. A great rubber duck.

@grok is this true???

Once people thought there was a bubble, they were wrong. This means that no bubble will ever pop ever again.

Imagine if Trump went on TV and bragged that he was shaping US foreign policy to suit Russian purposes in order to earn the $250 million given to him by a prominent Russian-American dual citizen.

Hmmm I notice it's enjoyable when I know what change I wanna make roughly but I am feeling very lazy or inbetween coffees.

I am definitely faster at coding at the agent. But it does let me do other things. This is also a small code base. i wonder if it becomes more time efficient over a larger one or not.

I'm enjoying agentic coding. I wonder how much of it is a novelty, like when I first started copy pasting code into chatbot interface LLMs too much. regardless, an agent thingee that runs in the terminal and reads your code for you is a better way to interact with LLM