It would be a shame it there where only delegated clients this would leave the ability to post as you and move your home sever in the hands of the PDS. One of my goals is that the PDS can go out of business with no warning and you from your client can authorize a new PDS and upload a full copy of your repo with no assistance from the old/non-existent server.

This means that you need a client with a always up to date repo of your content. For something like the "my twitter data" archive or ,i think, the mastodon migrate your data archive it is expensive to export an archive everyday and infeasible to do it on every action.

For a Merkel Search Tree it is a persistent data structure with good structural sharing so the diffs are small. You can compose your change(posts) commit locally and then upload to the PDS. You have an up to date repo and the PDS lags only by what is committed locally but not pushed to the server.

Users could use only the delegate client but that like using GitHub but only ever the web editor would lose much of the value of repos.

Find here from IPFS Camp 👋

https://youtu.be/OxmndRBGuUM?t=3704

I'm trying to wrap my head around what options indexers/crawlers have and was wondering about a couple of things I couldn't find information on.

a. Can everyone methods that sync repos from a PDS? ie com.atproto.syncGetRepo?
b. Is a client allowed to rewrite their commits to the MST/history after the fact? If so will the PDS complain/warn the user?
c. Is there a plan for the PDSes to provide aggregates about their repos, or will the indexer/crawler have do deal with retrieving updates for individual repos from the same PDS? I'm wondering mainly about how actions such as follows/likes can be easily retrieved without having to pull all repos from all "relevant" PDSes.

ps. Although I tried searching in this channel as well as the github repo, I still might have missed something so I do apologise. In which case a link or pointer to where I should be looking at would be appreciated :D

B) The latest head of the repo is the one that the home server returns and is signed by the key from the DID Document.
If the server updated to a commit that had no previous pointer or one that did not point to the last commit that is its wright. The old stale head is still valid just stale but there is no obligation that it be reachable from the latest head. Useful if you need to purge data from the repo.
C) The problem with aggregates is that you are trusting whoever made the aggregates that they are not adding or removing data. The servers will likely public some materialized views like the list and number of likes but to validate any particular like you would need to pull the repo with that like in it. Fortunately the repos are trees so a server that wanted to validate a like could do a narrow checkout of the repo with only the one record and the path to it. log(n) in the number of records in the repo.

The value of an indexer is that it is doing the work of pulling from many repos and building the aggregates so that a PDS that wants to show big would data like the set of replies to a post.

Have you guys thought much about how ATP might be able to introduce users to the Semantic Web? I mean like how to perhaps let users enter Duck-Typed data into the system without even having to have a special type-specific GUI. One interesting concept is to just have a "standard" (really a convention) where people can just embed YAML directly into their markdown like this:

123name:
  first: Clay
  last: Ferguson

That would be a way to get the Semantic Web booted up. Apps could choose to strip out the YAMO for most displays, or have a "Display Properties" kind of feature to render it as text, and yes, in some cases even present the typed text as a GUI Editor. Regardless, the indexing engine would be a billion times more powerful if people could just arbitrarily put in typed info that can become searchable, if the DB is something like MongoDB and can hold arbitrary key/values on records, etc.

In 10 years from now, people can just be told, "Cast your vote by putting something like the following in your Gov't Record"

12potus:
     MrWhoever

And that can literally be how voting is done.

I think the query parameters syntax dose this with already existing parsers

1?arg1=value+1&arg2=value2&arg3=123abcABC-_&arg4=true&arg5=5

one could image a syntax like

1lex:app.bsky.rel_me?did=did:plc:abc&handle=alice.example.com

that rendered oddly

I think what pfrazee: was proposing is that the entire repo is shared publicly, but that the access control is not done via the decryption key. Instead the PDS manages access.

Lmk if I got that wrong

Hey guys, I'm trying to use the api to create a sessionClient. Currently running into this error when connecting sessionClient

123		const sessionClient: SessionClient = (await import('../atproto/packages/api/src'))
			.sessionClient;
		const client: SessionServiceClient= sessionClient.service('http://localhost:2583');

Results in:

1234567session.ts:37 
        
       Uncaught (in promise) TypeError: Cannot read properties of undefined (reading 'sessionManager')
    at new SessionServiceClient (session.ts:37:37)
    at SessionClient.service (session.ts:25:12)
    at +page.svelte:17:47

Odds of Sha256 collision:

1/(2^256) + p(algorithm broken) ≈ p(algorithm broken)

Hey everyone 👋

Was checking out the site and saw the matrix link. Super excited for blue sky.

At the moment I would bet there are work application attacks that would slow down a home server but those are fixable. I would worry more about ways to spoof someone else's repo.

The only one that comes to mind is games with the Handels. DNS unicode lookalikes, http interception, PDS login

If you can spoof a binding between a handle and a DID they will know you by the DID there after. Not clear how to recover from this

If a repo contains records from a single user, that means that replies are stored in the repo of the person replying?

If so, how does replying to a thread/user on a different pds work?

ie thread’s repo R1 is on pds S1, user with repo R2 from pds S2 replies. Does R2 now needs to be synced to S1 for let’s say app.bsky.getPostThread to work on S1?

Does that means S2 needs to announce to S1 that there is update to a repo it cares about? And does that happen with updateRepo or is it different method?

Perhaps I have spent slightly too much time researching all the previous content distribution protocols.

Usenet is much better than the BBS(Bulletin Bord System)s that preceded the internet.