yfs and FileSystem needs to be changed to support rename file and directory.

[kifhan]

Description

MultiHack use this way to maintain files.

self.yfs.set(filePath, Y.Text)

but I found out this way you get really messy when you try to rename file or directory.

Because of this problem: Cannot change property's key name on y-map. You must delete obj from Y-map and insert again. But when you do that (delete and insert) you cannot use exist Y data type object (y-text, y-array, etc) because Y-Map set takes only TypeDefinition and creates new data type object.

Solution

My Solution is use two Y-map data.

filesystem_index: 'Map',
filesystem_nodes: 'Map'

Both filesystem_index and filesystem_nodes use randomly generated contentID as a key. One contentID can access both file meta information (filesystem_index) and file content (filesystem_nodes).

Meta data looks like this.

self.yFSIndex.set(contentID, {
  name: filename,
  type: filetype,
  contentID: contentID,
  replydbID: replydbID,
  parentPath: parentPath
})

Feel free to check my implementation on Rellat. I also fix few things on filesystem.js and main file (index.js). One tricky thing is when you use _buildPath you can't guarantee to get right meta data at right timing. So I made this sideway.

self.execWhenTargetSet = function (target, f) {
    function ft () {
      if (target) f()
      else setTimeout(ft, 50)
    }
    ft()
  }

I'm not sure this is best way but this works pretty well.

[t-mullen]

I've talked to the YJS author about initializing values before, and it doesn't seem to be a planned feature. I don't think the algorithm can support it.

The overhead of pushing a project into YJS is huge because every character is an insert. :/

Your solution will work great for renames, but I wish we didn't need to do thousands of inserts every time a file is created.

[kifhan]

I see the point. Like you said, it's a limit of OT and CRDT structure.

ex) Only google docs, sheet, drawing, etc support OT in google drive. If you have a docx file and want to collaborate with other, you must convert it to google docs type.

So I think inserting file string into YJS is a sort of converting. It's not the best but better then dropping CRDT.

I'm also thinking about using WebTorrent magnet address for large file (ex. over 100MB or Zip file) instead of Y data type. Like this one. P2P shared-filesystem (yjs)

[t-mullen]

YJS uses a modified version of WOOT. I think a block-based approach like LogootSplit (which supports multicharacter inserts) would suit this project better. I'm working on it.

[kifhan]

Dotted LogootSplit: A delta-based sequence CRDT

I found this, have you checked? To me, I can't tell... I need to see the working code sample to know this is actually working.

Anyway, It is cutting edge. How do you think?