Sync is a fundamentally broken.
The only solution is to switch to technologies where merging data is a primary core feature. Thus why I use text files and git to manage the vast majority of my data.
The sooner people realize that sync isn't fixable (or a good idea in many cases), the sooner real solutions appear.
Yep. It's weird that everything is built around just one button that says "Sync", when there are two different directions involved.
Either you want to save the state of whatever device you're currently using, or you want to revert it to the state stored in the cloud.
Export <--> Import, pretty much.
'sync' somewhat makes sense, but two sync options labelled easy to understand would go a long way to clearing this up, but... would likely introduce other issues as well.
'sync' works fine, except for when it doesn't. :/
The solution would have to be to store everything in ascii formats (probably taking a hit in terms of filesize and performance) and get everyone to understand the ascii representation each of these data formats and how to merge them effectively.
The alternative would be to build a diff tool for every binary based format out there.
The simplest solution would simply be to implement locking on everything and have one authoritative copy. So you can have your data on any device you want and share it with as many others as you want but only one person/device can have write access at any given time and can only have that once they have updated to the latest version of that rescource.
Does it do offline sync with multiple users whose changes may have different priority?
How about if time isn't set properly between multiple devices?
How about a multi-way modify/delete/rename where two records are given identical keys on more than one system, then both systems are reconciled?
I really need to write up a "Sync Gauntlet of Pain" that shows how sync implementations are incapable of making correct decisions, as they just don't have enough information, and giving them more information may not help.
Yes it does offline sync with multiple users. Changes aren't prioritized but you can configure what it does with conflicts (merge if possible, discard, or leave the conflict).
Time synchronization between multiple devices has been a solved problem for years. Not worth worrying about.
Creation of identical record keys — while theoretically possible — really never occurs in practice.
This looks like a really good system, and far more transparent in how it works vs. the "Black box" approach most systems use.
Many of the same methods can be applied to other types of synchronization, but it's more effort than most people can spare when they're in a hurry to get something released.
It's for this reason I've been keeping my eye on Rich Hickey's Datomic. It's not open source or free, but the idea of immutable data (everything is attached to a timestamp--yes, everything), seems intriguing.