This is probably the most direct statement I’ve seen about AT-URIs….it’s appreciated but a bit disappointing.
It feels like you are describing URLs. atproto URIs are a bit different. They are structured references to records within a global namespace.
Not exactly. There’s been a number of conversations seeking solutions for deterministic non-”atproto_pds” structured AT-URI references.
Here a few (spaces may solve some of these):
-
https://discourse.atmosphere.community/t/proposal-at-uri-for-cross-service-records/676
-
https://ngerakines.leaflet.pub/3mf5wu6fs6225 (https://discourse.atmosphere.community/t/permissioned-data-is-a-love-triangle/621)
Putting the DID service endpoint identifier in the URI would be re-centering the hosting location in the URI itself, which is exactly what the protocol removed in the first place.
That implies that AT-URIs can only ever be served by a predetermined set of services defined by the AT Protocol itself leaving no latitude for 3rd-party PDS-like services to provide functionality that is unavailable in a user’s primary hosted PDS.
Atproto spaces already demonstrate this challenge where a user’s “primary” PDS may not support space functionality, thereby adding a new “service” to the user’s DID document to identify the capability.
While I’ve been following the design from afar, I’m curious to understand if the space’s functionality is required to be hosted by the user’s primary PDS?
I’m suspecting based on your comments this might be the case; therefore, to start using apps designed on spaces users would needs to:
- wait for their provider to upgrade to a space-compliant PDS, or
- migrate to a provider that offers a space-compliant PDS, or
- create a separate Atmosphere account for spaces
None of these options sounds appealing. What might I be missing?
I think it also implies that there is a protocol extension mechanism where folks can invent new service endpoint types, stick them in DID documents, create URI types, and call all that atproto. I don’t think that is realistic or good for interoperability.
That’s unfortunate. This implies that innovation must always be embedded into a singluare “primary” PDS? That seems impractical, non-scalable, not fault-tolerant, and not very decentralized as AT Protocol matures over time. The PDS becomes the single-point of failure (network outage, take-down, throttled, capacity constraints, etc) impacting every single app the user uses. (less we forget that authentication is also tied to the PDS…that’s another whole discussion)
I think the protocol could clearly specify ways of structuring namespaces for data in the network. Then it should separately define synchronization mechanisms. Knowing which service endpoint to use from the DID document is a very small detail once you have defined all of that.
Can you elaborate a little more? This seems to contradict the preceding statements, no?