@jgoerzen Funny timing; I just read about Syncthing's protocol yesterday: https://docs.syncthing.net/specs/bep-v1.html

If we treat the file tree as an associative array where keys are filenames and values are either associative arrays (subdirectories) or immutable values (files), then I guess Syncthing is a CRDT.

Of course, we can emulate a set by using empty files for values.

I guess one could implement some of the text CRDTs using file trees as, well, trees, storing data in directory names, but those names are limited by filesystems (e.g. 255 bytes on ext4).

Syncthing is still a CRDT if files are modified, as long as the modifications touch different blocks (which are 16 MiB at most). If one thinks of files as arrays whose elements are 16MiB values, then one can implement grow-only and positive-negative counters; although I think it's more space-efficient to implement them with sets (directories full of empty files, the number of files is the count).

I always thought that trees are a universal data structure that can emulate the interface of any other, so I also guess that Syncthing can implement other data structures too (but inefficiently).

This is all very impractical, but is a fun theoretical exercise to work through :)