Kristin

hiii 👋

You're not showing up because you talk about cringe

I just don't post

We are not the same

Thanks for mentioning me. I might just concede at this point that I don't know how to social media lmao

Unfortunately true 😞

Yess!!!

Apparently? I thought I was washed fr getting 33 at first.

Ok, i got a 33 on my first attempt but on my third i got this

How are you doing this? This seems insane to me.

Holy shit, how?

you may have heard of model checking, but what about modal checking? (tool that just wildly clicks buttons that sound vaguely positive)

some of the coolest PRs are those that grow in size to thousands of locs of addition and after some deliberation shrink to like, 200 locs addition and 100 locs subtraction

same problem with transients in clojure. it can be a nightmare. if you find a good workflow, do tell.

I think @astralabyss.m1em.gay has both atproto and clojure experience. I'm gonna be so audacious and ping her here ^^

*cough* c++ devs writing everything themselves in a shitty way instead of using a library unless it's header-only *cough*

when someone tells you you've changed, how do you even react?

I was inclined to just thank them and tell them it took a lot. But that feels inauthentic. I changed because I had to. And in part it's not like I wanted to either.

It was as easy or difficult as staying alive, and about as inevitable.

instant relief surgery free hemorrhoids terminating gel made with authentic CUTTING EDGE GERMAN BIOTECHNOLOGY

lmao

type propagation via go channels. distributed type checking with Kafka. Using Amazon sns / sqs for type checking in the cloud. Type checking as a service. Watch out, I'm going to publish an update for your unification variable. Help my skolem ended up in dead letter queue. I won't ack this type.

Something i noticed in general where I see AI coding assistants struggle a lot: If you start with a certain style, it'll perpetuate it, seldomly going for the more robust, simpler, or plain better approach. Makes me worried for vibe code heavy projects that the quality can at best stay the same.

Solutions built on Communicating Sequential Processes are just so nice. Having to solve problems in this domain without it is such a pain.

have you considered

private <S, T extends S> S generify(T t) {
return t;
}

/j

if you look at List<T> as a type constructor, it should be able to specify that it is considered covariant in T, and forbid write access. If you want write access, you need contravariance, which could be its own type.

idk, invariance with generics is a pain and inheritance was a mistake.

yeah! i'm saying that should work as well in the default case.

i know, covariance and contravariance is specified on the function, not the type, but like, why?

a function that can work with a List<SomeType> should have not problems working with a List<SubType> unless it does mutation.

Java will slowly, but surely, drive me insane one day.

How the hell is a List<? extends SomeType> not compatible with a List<SomeType>? as in, a function taking a List<SomeType> parameter won't accept a List<? extends SomeType> value.

I've started datamining the video game doll's nest. never done this before. It's a fun process!

Becoming increasingly weird on the internet (affectionate)

I've had my paranoia hat on for long enough. Gotta stop doing that.

Some people don't seem to understand what other people value. If you want to make an AI tool for some existing niche, it should *improve* UX.

E.g. why the fuck would i want to write a prompt instead of click a button?

raw sql? I would never. I use Spring Boot, which supports JPA, which nicely abstracts over all databases.

Needless to say, I only write JPQL, unless i need a postgres feature. Then I add "native = true" to my `@Query` annotations.

angy clojure noises

Haskell does a special kind of type checking, which no other language does. But it's not overzealous ivory tower academics — it's elegant and rigorous precision in action ⚡

(/s in case that's needed)

I'm more and more convinced that the main problem types can introduce is that they're not good general purpose specs, but will be used as such.

They're the allowlists approach, but once you want to specify illegalalities, e.g. "only even numbers", they quickly suck at modeling that.