file-systems as databases

[Andy Davidoff]

As you may have guessed, I'm a big fan of file-systems as databases.
More than a few of my tools use the file-system for metadata, because
it's such a portable format and it lets me leverage so much existing
infrastructure for rapid and portable prototyping, not to mention the
ease of debugging the database itself...

Many people don't realize you can do this[0]:
	$ ln -s "some short value" key
	$ readlink key
	some short value
	$
This is nice because it forces key/value pairs to conform to a more
rigid interface and therefor simplifies the back-end significantly.

It's also possible (though not necessarily recommended!) to make
hard-links to directories, which lets you create equivalent (and
alternate) views into the same file-system.

Anyone else use file-systems for perverse purposes?[1]  Such as what?

[0] readlink is not present in Slugaris; at least not in versions <=7.
[1] pr0n is not an appreciated response, here. ;-D


#if John Chambers /* Oct 02, 17:45 */
> Well, yeah, but it also describes the usual unix file system.  The
> single file is called something like "/dev/hda3", and within that
> file, a file system is implemented.  That file system has names
> that are a string of keywords, which looks a whole lot like a DBMS
> key.  Lots of people have observed that you can easily implement
> many common DB "features" simply by playing games with the way you
> build pathnames and create directories.  Thus, the shell globbing
> that does things like '*' and '?' implement cross-sections or
> projections.  And so on.
>
> This is not an accident ...
#endif /* jc at trillian.mit.edu */

-- 
Andy Davidoff
Sen. Unix SysAdmin
Tufts University