I remember reading a USENIX paper on doing just this, by mapping each filename-related syscall to different semantics. Essentially 'ls /foo/bar/' would list the

I remember reading a USENIX paper on doing just this, by mapping each filename-related syscall to different semantics. Essentially 'ls /foo/bar/' would list the same files as 'ls /bar/foo/'. This paper on 'TagFS' is not it, but close enough: http://semfs.ontoware.org/pubs/2006-09-iknow2006-tagfs.pdf http://semfs.ontoware.org/pubs/2006-09-iknow2006-tagfs.pdf If I am going to be opinionated this is not done to OSes because of all the applications written for the current semantics. If I was given funding to make file tagging happen, then I would forget retrofitting the OS-level file system. My attack angle would be in the GUI making it appear as files are tagged. Perhaps by changing the popular applications to connect to a daemon that handles persistent storage and tagging of blobs. The regular user wants to know as little as possible about the file system anyway. Their concept of stored files are those that can be seen as icons in their desktop. Pure hiearchical paths are a thing for programmers and power users (sadly?). Oh also, while going for tagging things, RDF is even more generalized, allowing you to classify tags them self. Each application might want to tag files they create or modify with an application-class of tag. (date-class tags, format-class tags, author-class tags,...) Each application should implicitly add as many tags as it can when saving things (application name, resource kind, content-type, author, date, icon, etc). This would allow you to get files by recalling "that image i shopped up in gimp last month".