Have you tried github.com/ish-app/ish? Uses normal Alpine, but probably isn’t as well integrated into iOS.

Same! My read on the error handling feedback is pretty much that people want to “throw” errors and not have to care. Error handling code should take up space – because it’s just as important as the happy path. All the whining about having to do error handling makes zero sense to me.

Better to discuss that with friends and colleagues, where it actually makes a difference. For the wider internet: live and let live. The alternative is just going to annoy annoying people who then will annoy me.

Yeah, how do you feel about that trade-off? 😂

Ah, the good old days when iPhones were wonderful instead of addictive timesucks, social media was upcoming and exciting, the dream of ”global village” wasn’t dead, and the tech bro fascists still were in their closets?

Doesn’t look like that kind of colloquialism. Maybe something uncached in the action on first run?

Nice! You noted in a comment that the Tailscale setup took 2 minutes. Is that normal @apenwarr.ca ? In Lambda-style setups that’s amortised across invokes, but it would be kind of expensive (dev waiting time) for GH builds that may take about 2 min themselves.

Yeah, but it doesn’t *really* help because the values aren’t actually limited to the defined constants. You still need a default failure case. But it requires a fair bit of API misuse on behalf of the caller, so 🤷‍♂️

I miss being able to define ”must be one of x, y, z”. Type+const+iota looks like enums, but anyone can use a literal or cast a value of the underlying type to introduce random values. We have a clear idiomatic pattern of pseudo enums, just a bit annoying that they aren’t the genuine article.

I wonder if it’s actually per *account*, I hope that the randomness is at least coordinated at the organisation level so that sub-accounts get the same physical AZs.

Yeah, I guess it’s something that will have to be investigated empirically. The current plan is to evict old documents to S3, so if that is something that could be abstracted away with Postgres backed by S3 and a SSD cache that would be an application simplicity-win.

Are there any scenarios where you would recommend using an analytics cluster as the primary application database, and what would the trade-offs be? The use-case I'm thinking of is f.ex. a document database that's pretty much append only / new document version results in new row + book-keeping.