samlaf

🧵 The Unbeknownst Pervasiveness of Immutability This was one of the most fun articles I've written in a while. Didn't realize immutability as a concept was used in such a diverse number of different places.

Why do I feel like the 8 fallacies of distributed computing are much less important in the cloud. Except for 2 and 3, all the rest can be made false in a cloud deployment. Good luck to web3 folks trying to compete with web2.

Go team: "Don't communicate by sharing memory; share memory by communicating" Go runtime: "your channels are implemented using mutexes which share memory" Hardware caches: "your mutexes rely on atomics which only work because of my message-passing MESI cache coherence protocol"

The pure-impure sandwich is great in theory, but in practice I always end up implementing a multi-layered sandwich because the impure data fetching steps are expensive and should be skipped when some previous pure validation step works. Another example of theory != practice.

The composition fallacy The composition of safe components does not automatically yield a safe system Safe components ≠ Safe system x.com/DominikTornow/…

How have I not heard or seen base32 used in the wild until today? Lose 1 bit of efficiency wrt base64, but gain 2 nice properties: 1. Keeps lexicographic encoding of LUIDs 2. Easier to read and Error correcting (O=0, I=1, etc)

If the Accelerate and DORA studies are true, are BFT systems where code is run by external validators at a complete disadvantage here? Higher cost of coordination means less safe system.

Code review: developer tells me what the program is doing. Me: The computer can tell me what the program is doing. You should tell me what your intentions were. How else can I tell whether the whole effort is "correct" or not?

youtu.be/eeOANluSqAE “mcp tests should assert that the llm significantly often calls the right tool/resource given some standard prompts” I feel like REST’s HATEOAS might finally become a thing now that we have AIs to consume them.

The subtle difference between a golang nil interface and interface containing nil is an endless source of nil ptr dereferences... can someone fix this?

github.com/jonrad/lsp-mcp Is anyone using this? Grep (regexp search) Ripgrep (fuzzy search) Astgrep (syntactic search) LLM (pattern matching search) Lsp mcp LLM (semantic search)