*ptr

Altho config langs like Cue, Jsonnet, Nix, Nickel seem like they’re perfect for this sort of thing

Sick, thanks! So that’s what the sync.* queries were for

Wait what can you explain this, I've been puzzled by how to easily backfill data from the firehose for relevant schemas if my DB/backend ever has down time (it will 💀)

I wish there was an #atproto action-to-api lookup for popular libararies like the @/atproto and @/atcute npm packages.

This lookup of knowing what you want to do → the right library code to call feels like it should be a solved problem, but library docs are almost always in the reverse direction 😩

Holy real, why is atproto so hard to hack on 😭

Real until you want to do anything more interesting than manipulating simple structures, at which point I really really miss nice abstractions and stronger type safety

Jsonnet2lexicon

Oh interesting, I wonder how edit history’ll work with directly editing the post record (no history in user repo)

Oh bummer, yeah i only use nixos on x86_64-linux (and nix on aarch64-darwin), i imagine a lot less ppl use other platforms

(Not super familiar with Google cloud monitoring so this might be apples to oranges but) Grafana supports public view-only dashboards iirc

Pdsld is awesome, I like how seeing the frontend ui gave me a really solid mental model of the “shape” of a user repo

Is it possible to edit a post on bsky by editing its record?
- If you see this, it doesn't seem possible.

You love to see it🤩

Wait this is a good point. What’s stopping new appviews from just not writing to the PDS and using their own storage? This is already happening with some (very popular) apps (tbf, for decent reasons)

So many decisions come with pros and cons. I want to use the NixOS service modules bc they’re so easy to configure, but afaik you can’t run systemd in containers on k8s properly💔

Yeah I use k3s on NixOS using their experimental tailscale integration for networking

I think it depends on your setup; I’ve had a lot of trouble getting flannel to NOT suddenly take over as the main ip route in the host OS😭

Oh wow I didn’t realize this wasn’t a super stable thing yet

Do you deploy containers that somehow use images that use NixOS service modes like services.<service>.<options>? I really want to easily use the NixOS modules system, but they depend on systemd and it seems like running images that use systemd in k8s is not really something people do

Old post lol but what’s the use case? I have k3s on nixos but I don’t provision machines cuz it’s a homelab

Interesting combo, I run k3s on nixos machines

I imagine this has got to be really niche lol

I just played this galaxy last week and it’s one of the most notable

Yup exactly. I'm mostly interested in implications of this, since it seems like adversarial PDSes in the atproto network can impersonate any handle from the pov of appviews, eg. poison Bsky's DNS resolver ➡️ a post can show that the handle is @someone when in reality the backing PDS is not-someone

Something like this: bsky.app/profile/star...

Atproto handles might be susceptible to impersonation Proving you are who you say you are isn't enough

I'm pretty sure you can impersonate a handle in atproto, as long as you can spoof DNS records

I had the same question as that comment, and I think you can actually use arbitrary handles with DNS poisoning.
It’s easier to think about from an adversarial POV. They could create a DNS record for a handle they don’t own, point it at their PDS, and have their PDS use that handle.

Tangled is cool because its user-to-user ("social") aspects (including issues, PRs) are stored in Atproto, but Radicle is cool because its user-to-user ("social") aspects are automatically replicated. Maybe Atproto should have some way to hold data outside of a PDS like secure scuttlebutt..

Atproto lexicons remind me of Radicle's COB types. If you squint a bit, they're pretty much the same thing. Unlike atproto records though, COBs are replicated via a gossip protocol, so IIUC they can't really be deleted. I kinda like this approach better though; user data seems more resilient

That’s sick, i use deploy-rs in my current setup but the building is remote