Web Commenting Redirect + Resource IDs

This week I've done a few additional experiments trying to use the Channel Messaging API with the idea of getting rid of the redirect flow when commenting on the web. In the end it didn't work. Apparently Safari is only able to give access to cross-origin cookies, but not other state like IndexedDB. I then started hashing out ideas on how to implement cross-origin signing of comments by using key delegations. The main concern I have is that I don't want the additional keys start sprawling around with full power over the account, like the agent keys we currently use for hyper.media. I want session keys to be scoped for a particular target. I haven't finished cross-origin signing, because  brought up the fact that addressing comments is a pressing concern right now, which I agree with. Exposing comment URLs is surprisingly tricky, because we need to change the format of a comment blob a little bit, to include additional information.

Universal Addressability + Updated Contacts

This week Eric brought up a few problems with the Contacts project that I wrote last week. Mainly it's about addressability of them. It also surfaced a few problems with mutability of Comments, which we still don't support, just like we don't support linking Comments. Both Contacts and Comments seem to be similar in terms of how they behave, namely they are state-based objects, unlike operation-based Documents. So we'd better have a generalized way of addressing and mutating those kinds of objects. I wrote a few more documents in Linear — unfortunately they are not publicly available yet. I guess we will publish them as Seed documents as soon as they are more settled.

Contacts v0 and Authenticated Data Structures

This week I spent a lot of time researching the various authenticated indexing data structures (Prolly Trees, Merkle Search Trees, HAMTs, etc.) that I wanted to use for implementing Contacts, because I also wanted to use them to solve other issues like deleting and moderating comments and so on. I've implemented a bunch of those data structures, made a lot of experiments, read a lot, but in the end abandoned the idea of doing it as the main approach for handling data — too many small but important tradeoffs to settle, and it's a bit disruptive for our current architecture. They still can be very helpful for designing overlay indexes on top of the blob data, that could be shared between peers in a verifiable ways (we've talked about that before), but for now it's not worth spending time on that. So, I finished writing up the Contacts project, using an approach that is closer to what we are currently doing, without introducing many major disruptions. There're still open questions there, so we should discuss them in Linear. Linear Project: https://linear.app/seedhypermedia/project/contacts-v0-6fb28d8b822b/overview.