Prior experience suggested it was a good guess.

HTML5 Parser?

In all seriousness, that would make a number of things I would like to do with tree-sitter easier. So I'm totally in favor :-D

I continue to maintain that we as an industry heavily discount the importance of modeling the error domains. (Self included). Which is why unchecked exceptions are so popular and why there is such a strong temptation to just `unwrap` or `expect` an error in a language like Rust.

I have embodied the grandma's recipe problem for my children now. I'm having to measure as I cook so I can give them amounts so they can cook. The problem being I almost never measure when I cook. I do the add ingredient, wait, taste loop until it's correct.

#cooking #recipe #children

Yeah, I'm currently getting nerd sniped from the discussion recordings from these.

Technically it only just started but I'm making coffee so it's starting out pretty good.

current status: Listening to Cowboy Bebop (Original Soundtrack) by Seatbelts - 1998 and furiously hitting thumbs up on every track...

I share this story all the time to people. I consider it part of our shared lore in this industry.

This seems counterproductive to me. The tool is here. The tool has real value. The tool will get used. Might as well be part of the conversation that guides it usage productively.

There is a real conversation to be had around responsible and professional use of LLM's in the production of code. It feels like the actual conversation being had is instead an argument over whether it should be used at all.

If what it produced wasn't right then I just rewrite it myself. Tweaking the prompt is the least enjoyable part of the process. I find it boring and entirely un-engaging.

I use neovim's avante plugin and almost exclusively use it's Show a diff in the editor approach. Basically I treat the LLM experience like I'm doing a code review.

That said, I really only use it to get started and then start feverishly editing whatever it did. I *despise* prompt engineering.

Instead, it is an altogether boring, tedious, and frustrating exercise. If this is what the career looked like when I first started out I would probably never have taken the developer path. Probably would be a virus hunter for the CDC or something right now.

I have been using coding assistants for a few weeks at work. They are not bad at reducing the time to first line of code. And I can use them fairly effectively so they can provide value.

However, prompt engineering does not in any way exercise the part of my brain that enjoys writing code.

Felt a very small bit of ground shaking this morning in western north Carolina. #earthquake?

Serious question: How do you tell why a tool isn't getting used by your llm agent graph? These things seem entirely inscrutable and undebuggable to me.

You know how sometimes you are like... this should work why isn't it working. And you just automatically suspect the tool because "It's TLS, there be dragons" and then you realize you were just looking at the wrong callsite the whole time?

Yeah.....

Cloudflare WAF False Positives Cloudflare's Status Page - Cloudflare WAF False Positives.

Some people when they have a security problem think "I know! I'll use a regex!" Now they have "Internet Scale" problems.

www.cloudflarestatus.com/incidents/gs...

Does it do a better job at deep research than GPT? That is basically my biggest use case for LLM's these days.

When your syntax sugar creates ambiguity in the semantics of your language it may not always be pulling it's weight.

I am in this post and I think I like it.

Ohhhh..... Noooo..... I barely know where to begin.

I personally would prefer the rust type system in a compiler but I am *right now* writing a language interpreter in C# and it's fine. A little clunky in places but still reasonably doable.

Lately due to the state of our current politics in the US I've been craving examples of extreme competence. Like, piano virtuosos, incredible builds by craftsmen, and so on. It's nice to be reminded that there are still people out there who have real talent and are capable of good decision making.

Something I like about the Rust `impl TraitName for TypeName` syntax as opposed to the more traditional `class Name: IInterface` syntax is that you know exactly why a given method is there in the `impl` block. It is there to satisfy the trait requirements.

Despite the second point I regularly use Rust for prototyping and hacking on stuff.

Most interesting man meme with the top text saying "I don't always re-use string memory" and the bottom text saying "But when I doMooCowMooCowCow"

I once inspired this meme while using unique_ptr in a C++ codbase. To this day I and several other much better C++ devs than me do not fully understood how I introduced the bug.

This is why people get so excited about the promise of Rust.

Honestly Greg K-H's response in that thread gives me some hope that behind the scenes things are moving in the right direction. I just hope Linux Kernel Leadership makes some public commitments soon.

Like any locally optimized thing moving out of that local optimum is painful. I'm not sure linux can successfully move out of it. But long term if it doesn't it will have difficulty gaining new devs.