I think the last time that errexit and pipefail were brought up to the fish devs, we were treated instead to an essay about how errexit is a total hack in bash, how it has all kinds of places where it doesn't work as expected, how it's special-cased away for some constructs, etc. None of which is wrong, but it's still a whole lot better than the nothing that we have with fish now.
So we still have this inflection point where scripts eventually have to graduate into a "real" language, and while those languages give us proper data types and structures on one hand, they take away conveniences like pipes with the other. It's 2026, and we managed to crack artificial freakin' intelligence before we got a decently evolved shell into the mainstream. </yells-at-cloud>
The author of CommonMark and Pandoc has a new format called Djot: https://djot.net/ that I've been meaning to check out. Supposedly more sane to parse, and it comes from someone who would definitely know about that sort of thing.
Djot is great. I use it in my project ( client for https://timbran.org/moor.html ). It has all that I needed from markdown without any excess, and it's safe and easy to parse and familiar to people.
> Straight double quotes (") and single quotes (') are parsed as curly quotes
I don't know who actually likes curly quotes, they are clearly excess to me. And as parsing sometimes fails (as the site says it may), you get inconsistent results, and failures stick out like a sore thumb.
Here is another syntax: this is <*bold>. Very unlikely to clash, can be vibe coded in an hour. But it's more of the same.
For reading, I don’t know who prefers straight quotes.
For writing—
There are more than a few people on HN who deliberately type curly quotes and other non-ASCII punctuation, due to a strong preference for them. I’m one of them.
I use Compose sequences: ; ; for left single quote, : : for left double, ' ' for right single, " " for right double.
(Accordingly, I hate being subjected to automatic curlification, partly because it’s not always correct, but more because if I typed ' or " you better believe I meant ' or ".)
How would you possibly enforce this? I can disconnect my laptop from the internet and the local LLM will still autocomplete TW classes. Does JetBrains therefore owe TW every time it does this? What if it was actually completing UnoCSS class names that happen to overlap? How about when it's just simple autocomplete based on what classes are visible and what I've used within the same file?
These might sound like snide rhetorical questions, but when you start demanding payment, they're very real.
If you see a bunch of Tailwind markup on websites without a license key, you can enforce your license. The LLMs can write the code for you, but they either have to negotiate their own license or instruct users to get their own.
The comparable I am familiar with is Font Awesome. Even if you want a free plan, you still have to create an account and get a key.
reply