some open source enthusiasts will talk about composable ui components and views in different contexts, meanwhile i find if you slightly change the line spacing on a poem it can change how it reads
while open source has always promised extra trust because you can inspect the source to see whats its doing, its a stronger claim now that ai has made that more practical
so cool you managed it on your own laptop
your post is making me imagine a "stubbornly resilient client" whose goal is just basically to get posts from wherever it possibly can (firehose, microcosm, pds)
@jackvalinsky.com aspects like you mentioned is why I said "gracefully degrades". in the case of certain infra being down, I wouldn't expect full backlinks and like counts, but it would be nice for things to not "failwhale" that don't need to
in the end what i really would like is scuttlebutt, but that considers bluesky as a useful relay whenever it happens to be running
generally interested in atproto tooling that gracefully degrades to accessing pds-es directly when other infra is down
spoke with an iranian friend at length yesterday who basically said "disliking the islamic republic and also not wanting to be bombed by the us has led me to something close to anarchism and dislike of nation states in general"
seems remarkable/silly that sshing into a homeserver via tailscale on the airplane wifi works
thought about releasing notplants 3.8 today but decided its too powerful, so you will continue getting normal notplants content for now
this explains a lot
pds stands for penguin data storage
enjoyed learning more about paul erdős today
(not particularly related to the news that initiated it)
the firehose if everyone was just DMing you
lol love this
love the feeling of how decentralized systems can at times dissolve the sense of data/code existing in a particular place
having visual indicators of sync happening is delightful, the same way getting the arduino to turn on an led was always delightful too
seems consistent in tests, but still some weird timing issues with triggering syncs, like you need to hit the tv to make it sync sometimes
current protocol for syncing git repos over iroh with iroh-blobs: tangled.org/notplants.bs...
using yrs merge driver for text conflicts
last-write-wins for binary files
if filetrees are identical, lower sha commit is selected to avoid infinite loops
a screenshot of two editors with the same filetree shown
we believe the file trees and git commit histories are now converging with arbitrary amounts of peers and edits
> Passed. The result is "The geelue cat sat on the mat." — "green" and "blue" interleaved at character level. Weird, but no conflict markers, and deterministic. Both sides would produce the same "geelue" result.
current status
learning more about gits internal architecture feels a bit to me like visiting one of the modern wonders of the world
alternatively it could follow in the non-dual thread and ditch the small self all together
like if there was YrsFolder sync, for a folder of files (some of which are Ydoc and some of which are binary files), and it handled it, as simple as a Ydoc, would be cool
but there are also some nice things about the approach im trying now with git since it plays well with other git stuff
not impossible challenges, but not trivial (at least from my first attempts)
for this reason, I have been exploring using git, since it already handles a lot of that quite well
although I would love if someone built a YrsFolder thing was a robust and easy to use thing for a whole folder
@nickthesick.com I think something along the lines of what you are describing sounds right for a folder (a manifest of files as a Map), but with edge cases around create/delete/rename, and orchestrating when to sync manifest and when to sync files, in such a way that minimizes network trips
definitely seems similar using the pds as a way to take a snapshot
my general thought was live syncing when people are both online, and snapshots to pds for syncing/coldstart when people are not online at the same time
We can just work together
does anyone know any prior art on a strategy for syncing two git repos that is eventually consistent?
it seems to me like default git workflows are not that, but maybe by adding some additional assumptions/constraints such a strategy could exist?
