4 ms·

Hi HN! I wanted to research more elegant ways to enable document sync and collaboration in my apps sometime last year, and ended up discovering a new class of d

Hi HN! I wanted to research more elegant ways to enable document sync and collaboration in my apps sometime last year, and ended up discovering a new class of data structure that made it possible to build collaboration into documents right on the data level, completely separate from the network architecture. (Maybe you're already familiar with CRDTs. Well, these aren't just ordinary CRDTs, but almost meta-CRDTs that can represent a variety of convergent data types in a generic way. They also make it super easy to implement things like past revision viewing and delta patches.) As a proof of concept, I built a basic real-time collaborative text editor for iOS that works equally well over regular iCloud sync, CloudKit Sharing, and offline, as well as a simple mesh network simulator for macOS that has support for collaborative text editing and simple Bézier editing. The code is strictly educational for the time being, but I'm working on a Swift library that can hopefully put this stuff to use in real apps. Hope to start dog-fooding in release software by the end of the year!

urbit 8y ago
Unfortunately I wouldn't expect a lot of substantive technical discussion for a post like this on HN -- this post will expire before anyone has time to read it properly and produce thoughtful responses. (Which is broadly the problem with posting enormous applied CS-research braindumps to HN. Don't know a better place to do it, though. And I see people are upvoting you. Also, the absence of thoughtless responses is itself a great response.) The key point here, as far as I can tell, is really building a CRDT system as a true platform layer, rather than as a one-off solution for a special-purpose app. I think it's fairly clear that generic sync is a pretty essential part of a modern decentralized computing environment, and I don't think it's clear at all what the best way to solve it is. But... it seems to me that you've solved a large piece of the problem but not the whole thing. Because the most important document sync and collaboration platform is, of course, source code control. If you have a document sync and collaboration model that doesn't at least generalize to classic revision control, why not? Now, there's a sensible reason to separate these problems -- CRDT and OT solutions tend to specialize in the zero-maintenance case where a user-resolved merge is just impossible. Whereas if it was possible to build a zero-maintenance revision-control system, which automatically resolved all merges and conflicts, someone would have done so already. This certainly suggests that the two are different problems. But generalizing across slight differences is what system software does. Maybe one is a special case of the other? A generalized layer for lightweight collaboration is pretty powerful. But it certainly seems like the case that if you could generalize across lightweight collaboration and heavyweight revision control, you would have something that would be incredibly powerful. Or is this too ambitious? I don't know so I'm asking you.
- kd5bjo 8y ago
  I've had some thoughts along the same lines, but haven't taken the opportunity to flesh them out nearly as far. A git repository is nearly a CRDT that represents a file system hierarchy -- the only thing that's missing is a deterministic merge algorithm. If you allow files to be committed in a conflicted state that gets resolved by a later edit, you have a full CRDT that automated systems can work freely with. The catch, of course, is that despite pushing the resolution to an arbitrary point in the future, these conflicts still have to be resolved by hand eventually.
- blake8086 8y ago
  I suspect you could build such a thing if you could more rigidly define edit operations on source code files. The way we edit code now is roughly "insert/edit/delete string <x> at position <y>", which doesn't really carry enough context to auto-merge. If edits carried all the semantic information of what they were doing to the source: "rename symbol <x> to <y> everywhere" "perform step <x> after step <y>", then we could probably build a zero-maintenance revision-control system.
aboodman 8y ago
Hi archagon, Have you checked out Noms (https://github.com/attic-labs/noms https://github.com/attic-labs/noms)? > But studying Differential Sync further, I realized that a number of details made it a non-starter. First, though the approach seems simple on the surface, its true complexity is concealed by the implementation of diff and patch. This class of functions works well for strings, but you basically need to be a seasoned algorithms expert to design a set for a new data type. (Worse: the inherent fuzziness depends on non-objective metrics, so you’d only be able to figure out the effectiveness of your algorithms after prolonged use and testing instead of formal analysis.) Second, diff and patch as they currently exist are really meant for loosely-structured data such as strings and images. Barring conversion to text-based intermediary formats, tightly structured objects would be very difficult to diff and patch while maintaining consistency. Noms is a structured database in the spirit of Git. It includes efficient (cost proportional to size of diff, not size of input) diff and merge for sets, lists, maps, and user-defined structs. The built-in merge strategies are commutative and idempotent, so if your operations can be merged using it, then your schema is a (state-based) crdt. We didn't initially build Noms for this type of p2p, masterless architecture, but more and more recently, I find myself thinking it's a great fit. > Next, there are some issues with using Differential Sync in an offline-first environment. Clients have to store their entire diff history while offline, and then, on reconnection, send the whole batch to their peers for a very expensive merge. Assuming that other sites had been editing away in the meantime, distantly-divergent versions would very likely fail to merge on account of out-of-date context info and lose much of the data for the reconnected peer. Finally, Differential Sync only allows one packet at a time to be in flight between two peers. If there are network issues, the whole thing grinds to a halt. I may be misunderstanding here, but I think these issues do not afflict Noms: 1/ Noms can compute the diff between any two states efficiently. It doesn't matter how many local changes you make while offline - what is transmitted is just the diff between two states. 2/ Merge performance is relative to the size of the diff being merged, not the number of changes. 3/ I don't think one-packet-in-flight-at-a-time applies to Neil Fraser's differential sync. It definitely doesn't apply to Noms. Given your interest in CRDTs, I thought you might be interested to take a look. There's still more work that needs to be done to make Noms work really well for this kind of use case, but I think it's one that makes a lot of sense. As somebody who's just spent a ton of time looking at all the options, I'd love to hear your thoughts on how Noms fits into your thinking.
phyzome 8y ago
Thank you for posting this -- it was a great walkthrough, and very clear. Cool stuff. :-)

urbit 8y ago

Unfortunately I wouldn't expect a lot of substantive technical discussion for a post like this on HN -- this post will expire before anyone has time to read it properly and produce thoughtful responses. (Which is broadly the problem with posting enormous applied CS-research braindumps to HN. Don't know a better place to do it, though. And I see people are upvoting you. Also, the absence of thoughtless responses is itself a great response.) The key point here, as far as I can tell, is really building a CRDT system as a true platform layer, rather than as a one-off solution for a special-purpose app. I think it's fairly clear that generic sync is a pretty essential part of a modern decentralized computing environment, and I don't think it's clear at all what the best way to solve it is. But... it seems to me that you've solved a large piece of the problem but not the whole thing. Because the most important document sync and collaboration platform is, of course, source code control. If you have a document sync and collaboration model that doesn't at least generalize to classic revision control, why not? Now, there's a sensible reason to separate these problems -- CRDT and OT solutions tend to specialize in the zero-maintenance case where a user-resolved merge is just impossible. Whereas if it was possible to build a zero-maintenance revision-control system, which automatically resolved all merges and conflicts, someone would have done so already. This certainly suggests that the two are different problems. But generalizing across slight differences is what system software does. Maybe one is a special case of the other? A generalized layer for lightweight collaboration is pretty powerful. But it certainly seems like the case that if you could generalize across lightweight collaboration and heavyweight revision control, you would have something that would be incredibly powerful. Or is this too ambitious? I don't know so I'm asking you.

kd5bjo 8y ago
I've had some thoughts along the same lines, but haven't taken the opportunity to flesh them out nearly as far. A git repository is nearly a CRDT that represents a file system hierarchy -- the only thing that's missing is a deterministic merge algorithm. If you allow files to be committed in a conflicted state that gets resolved by a later edit, you have a full CRDT that automated systems can work freely with. The catch, of course, is that despite pushing the resolution to an arbitrary point in the future, these conflicts still have to be resolved by hand eventually.
blake8086 8y ago
I suspect you could build such a thing if you could more rigidly define edit operations on source code files. The way we edit code now is roughly "insert/edit/delete string <x> at position <y>", which doesn't really carry enough context to auto-merge. If edits carried all the semantic information of what they were doing to the source: "rename symbol <x> to <y> everywhere" "perform step <x> after step <y>", then we could probably build a zero-maintenance revision-control system.

aboodman 8y ago

Hi archagon, Have you checked out Noms (https://github.com/attic-labs/noms https://github.com/attic-labs/noms)? > But studying Differential Sync further, I realized that a number of details made it a non-starter. First, though the approach seems simple on the surface, its true complexity is concealed by the implementation of diff and patch. This class of functions works well for strings, but you basically need to be a seasoned algorithms expert to design a set for a new data type. (Worse: the inherent fuzziness depends on non-objective metrics, so you’d only be able to figure out the effectiveness of your algorithms after prolonged use and testing instead of formal analysis.) Second, diff and patch as they currently exist are really meant for loosely-structured data such as strings and images. Barring conversion to text-based intermediary formats, tightly structured objects would be very difficult to diff and patch while maintaining consistency. Noms is a structured database in the spirit of Git. It includes efficient (cost proportional to size of diff, not size of input) diff and merge for sets, lists, maps, and user-defined structs. The built-in merge strategies are commutative and idempotent, so if your operations can be merged using it, then your schema is a (state-based) crdt. We didn't initially build Noms for this type of p2p, masterless architecture, but more and more recently, I find myself thinking it's a great fit. > Next, there are some issues with using Differential Sync in an offline-first environment. Clients have to store their entire diff history while offline, and then, on reconnection, send the whole batch to their peers for a very expensive merge. Assuming that other sites had been editing away in the meantime, distantly-divergent versions would very likely fail to merge on account of out-of-date context info and lose much of the data for the reconnected peer. Finally, Differential Sync only allows one packet at a time to be in flight between two peers. If there are network issues, the whole thing grinds to a halt. I may be misunderstanding here, but I think these issues do not afflict Noms: 1/ Noms can compute the diff between any two states efficiently. It doesn't matter how many local changes you make while offline - what is transmitted is just the diff between two states. 2/ Merge performance is relative to the size of the diff being merged, not the number of changes. 3/ I don't think one-packet-in-flight-at-a-time applies to Neil Fraser's differential sync. It definitely doesn't apply to Noms. Given your interest in CRDTs, I thought you might be interested to take a look. There's still more work that needs to be done to make Noms work really well for this kind of use case, but I think it's one that makes a lot of sense. As somebody who's just spent a ton of time looking at all the options, I'd love to hear your thoughts on how Noms fits into your thinking.

phyzome 8y ago

Thank you for posting this -- it was a great walkthrough, and very clear. Cool stuff. :-)