Not quite sure if you really want that: open your favorite shell, and count the number of files in your data directory; then multiply it by, say, 2 seconds each -would you like to spend that time on organization, or rather making money? (obviously you can apply the same label to multiple files, but you still need to figure out what belongs to where).

Personally, I gave up on manual organization of my file system last year, and learned how to love built-in search. This works great, as the rough index is in your head anyways, so you can just take a single token of the filename, type it in, and open it up; massively reducing both access time, and eliminating organization entirely.

The fact that file systems force us to organize data a into a set of named files in a fairly rigid hierarchy has been lamented since the 1970s by Ted Nelson (the man who, among other things, coined the term "hypertext") and others. However, it goes even further back than that with the critique of library indexing systems by Vannevar Bush in his seminal 1945 article As We May Think [1] that described the "Memex" as an alternative: a system that to today's eyes might look something like a PDF viewer with chained bidirectional links (think '90s web rings, only for book and magazine pages), which were called "trails".

There have been attempts to remedy those problems by both individuals like Nelson [2] and corporations like Be, Inc. [3] and Microsoft (which may or may not be related to the fact that Nelson's book Computer Lib / Dream Machines was very influential at early Microsoft and Microsoft Press even republished it). It truly is lamentable that Microsoft didn't bring one into the mainstream with WinFS [4].

[1] http://www.theatlantic.com/magazine/archive/1945/07/as-we-ma...

[2] http://www.google.com/patents?vid=6262736, http://www.xanadu.com/

[3] http://www.nobius.org/~dbg/practical-file-system-design.pdf, Chapter 5

[4] https://en.wikipedia.org/wiki/WinFS

I wonder how an implementation of this using extended attributes (xattr) would scale? I couldn't find much information on anyone doing this before, except one command line/emacs tool:

  https://github.com/chujoii/xattr-tag/

  I suppose adding a gtk-gui on top of this would be possible.

As well as an explanation (of sorts) for why there doesn't appear to be support for (x)attrs in GNU find:

  http://web.archiveorange.com/archive/v/wNWnxu100AzH1ItS2Dk5

"find" can filter based on -exec -- but there doesn't appear to be a simple command that will take a filename, an attribute (and optional value) and simply return 0 if found, other if not found (not that it should be hard to write, based off the attr-utilities code, for instance).

Well, it isn't really a matter of file system since it is possible to implement "tags" on any FS using links. (i.e. by creating directories named after tags and by populating these directories with links to the actual files.)

Therefore what you need isn't a new file system but a file system browser that allows you to specify a (set of) file(s) using tags.

Considering how easy it would be to do just this using fuse, I'd guess some Linux folks already tried this, does anyone know about projects in this line?

In BFS (Be(OS?) File System) you can add any kind of attribute to a file, for what I know.

Unfortunally, this isn't very helpful - since you need to run either BeOS or Haiku. If I'm not mistaken, there's a FUSE port of BFS to Linux as well.

You can do this on a mac (Finder labels are stored in HFS+ metadata).

Google Docs used to do that (maybe it still does). It was absolute chaos and very confusing.

OSX supports this today with metadata and spotlight driven smart folders.