Very cool!
07.03.2026 09:45
👍 1
🔁 0
💬 0
📌 0
Very cool!
German Reichstag looks wrapped in white paper
Oh no they wrapped our government building!
That’s actually a great idea 🤔
From the artifacts, it looks like there’s no text box there, just a bitmap image… 🙈
I think/hope there will still be a lot of creativity from the original version recognizable in the remake, and the author of the original version might have copyright on the remake
I’m biased as I get paid to work on source available software
Yep! German used them the other way around. People understand it, but it fell out of fashion here in Germany
Taiwan looks gargeous
This page screams o11y, without mentioning it 😁
I just got --when.environments landed. Planning to use it to conditionally disable the visual editor within my agent, as LLM tends to run `jj commit` without a message, starting a visual editor, locking up my agent
oh no, this goes way too hard
The main reason I didn’t write IdHashMap myself is it’s just soo much typing for very little value. Maybe using IdHashMap saves me one hour in the long run. But I’m not going to make it if making it costs me 2 days
I’ve wanted to have iddqd::IdHashMap before. HashMap is fine too, but often IdHashMap is a more natural fit.
Bi/trijective hashmaps are also pretty common, as a concept. But I would have just sucked it up and used 2-3 HashMaps manually and kept them in sync by hand — hopefully without bugs.
And for LLM to produce code like iddqd, rain also needed to understand what’s possible and describe it in a prompt. I surely hadn’t thought about iddqd but I would definitely use it in the future.
You can understand code by reading and reviewing. Recently someone reviewed my PR for jj and proposed a lot of improvements for me to make. Given the feedback they gave me, they very much understand the code I made
I think there’s no way to make iddqd without really understanding what you’re doing 🤷🏼
And at 12k lines of code, I guess we are in the “slow typing” case here? :)
How many lines of good code can you write per week?
I think it’s a pragmatic way of adding new configuration to your function. Just add new optional Boolean parameters at the end. Existing behavior can be preserved.
This is especially relevant in languages where parameter packs carry a GC cost and are unergonomic in their own way.
Without agentic coding, iddqd wouldn’t exist in its current form.
At Oxide and Friends they had a great example. Crate iddqd contains 4 data structures. Their core was complicated to make, but then they each have a buttload of methods which overlap hugely in implementation and contain no creativity. But rain generics don’t help here.
I find agentic coding helps me a little bit — and only because I manually curate the output and continuously invest in getting more control over the whole thing.
Eight is the new nine
What did they do now 😁
🎧 Nice to hear about culture change at Windows in the ‘10s. And about how it hugely impacted people’s lives
We work remotely, strong comp, meaningful equity. Just ping me or fill out the contact form
I donate to a different news media. Stuff is important.
I thought that meant the LLM itself decided that the message was done. LLM generates end-of-stream.
ok, but Claude sounds wrong about the EOS part 🤔
Consider linking the rendered version of the dada formal model, it looks super useful
This reminds me of lexical effects. Or region type systems. But I’m not aware of anything in the literature that is directly relevant (and I’ve lost touch with academia)
