Talking Point: Could Linux Abandon Directories In Favour Of Tagging?

Hoarders: folks who pile things all in one room (or house) until there is no room to walk around. They mentally tagged where each item was but after the passage of time they no longer can find a specific item. When a hoarder's stash is cleared up and a long lost item is discovered the response often is "Oh, I wondered where that was".

Other problems with abandoning directories is that of ownership and security, one of the hallmarks of Linux. It would put a huge burden on a file browser or a shell to show only those files a particular user has the right to see, modify or delete, each file having to be examined in order to decide.

I just used "find . -type f | wc -l" as root to discover that my Kubuntu 10.4 system has a total of 502,862 files! A SOHO or corporate LAN might contain hundreds or thousands of times as many files. Microsoft's "Active Directory" is already a slug when trying to show the files in a directory with only a fraction of that total. Having a Linux file browser scan/sort/display who knows how many millions of files residing in one "directory" (or on one HD) would make AD look like lightening. Obviously a heavily indexed database with fast response times would be necessary.

My "tags" are my sub-directory names, and that paradigm works quite well for me, so I'll pass on what the author is proposing. Besides, as one comment already pointed out, there are apps that already allow for file tagging. No need to change the Linux file hierarchy structure.

I don't think that Tags are that useful, using them would imply to have to memorize lots of tags. Since entropy is a fact and you cannot remember your old ideas forever, after 2 years you may also wonder "why did I tag that file that way". Of course the same can happen with a hierarchical structure but the advantage there is that you only need to remember the broader subject and then go into more specific sub-categories by choosing among the options that you created, without having to remember them at all times. We all need to define our own standard about how to categorize things, most OSs have implemented pre-stablished directories like Documents, Videos, Pictures, tmp, etc. But we still need to define by ourselves (perhaps with some expert's advice) how to categorize our files beyond those main categories. One useful tool could be that in addition to the well know Ctrl-C Ctrl-V to copy and paste, an easy Ctrl-C Ctrl-L standar option were made also available in order to copy and paste symbolic links of the original files in many different places if those files fall under many different categories.

Okay so you're talking of desktop so it's pretty much a given that it will be GUI-based but tagging and associated searching/finding of files is heavily biased towards the GUI. How do you backup or copy only the 1000 files that are tagged with "client name" using rsync, tar or equivalent?

Also you have a heavy presumption on documents here - productivity ones at that. That's fine but there are a host of other files I have under /home. Finally with a directory structure it's easier to implement permissions based on groups etc. How would you easily restrict access to one bunch of files by group if they were all stored in one directory and tagged only?

I think extended attributes will answer some of your problems, check them out. Software like Beagle uses them, but you can use the user namespace for anything you like.

Software is/will be ported to use them. For example, see wget:
https://github.com/wertarbyte/wget/tree/xattrurl

Cheers,
Kalin.

I realized the strength and flexibility of tagging when first using labels within gmail. I love the idea of having attributes on a file and then searching or sorting by attributes. The current linux file system architecture is has lots of mileage and is very powerful. I love it. Merging these two in an elegant way would be the challenge. Maybe just keeping them separate and adding this type of functionality to an arbitrary folder, say /home/{user}/taggable-data, would be a decent trade off. If I want version control in a directory, I use git (git init....) and use the tool from there. Why not have a similar tool that uses the existing filesystem yet presents a DB view of the directories contents. That tool could be command line app or a file manager plugin that allows you to add a file, set/modify/delete its attributes, and search for a file. Apple has been doing this for some time with they way iPod songs are presented to users either on the computer or the iPod.

Isn't this what the "Semantic Desktop" is for? In KDE you can tag, rate, comment on your files.

Personally, I see it that tagging too much effort. I have to do more work saving files. Then I have to remember what tags that I used (more work). Keep a good hierarchy keeps my search time low. If I need a broader search, I just use general search tools (find/grep type).

Awesome ideas, I like them! Faster, sensible, and more like how we actually work. The traditional FS hierarchy is very limited and restrictive. The notion of file metadata with tags that cross multiple categories is closer to the paper-and-filing-cabinet world-- paper files can live in only one physical location, but can logically be in multiple categories. This is handled with cross-indexes, which assign multiple categories to single files.

You can put tags in the filename (seperated by underscores) and use find to retrieve them.
I wrote a script which prompts for some tags and then fills a directory with symlinks to files matching those tags.
I only use this for certain files which I store in a single flat directory.