Hierarchical organization [LWN.net]

Hierarchical organization

Posted Jan 30, 2013 13:48 UTC (Wed) by dskoll (subscriber, #1630)
In reply to: Seigo: Plasma.next()? by aseigo
Parent article: Seigo: Plasma.next()?

Let me expand on why hierarchical organization is incredibly useful.

Take a collection of photos. Organizing it with tags assumes that we'll actually take the trouble to tag the photos. Raise your hand if you've ever sat down to 500 holiday snaps and relished the thought of classifying them. Thought so.

On the other hand, I can pretty quickly find the photo I need using my file system hierarchy. I know that it's likely to be in Albums/2012/08-Trip-to-Zurich. I also know that by changing into that directory, I'm not going to see extraneous photos... I can shut out the rest of the world from my visibility, reducing clutter.

In effect, by dumping my holiday snaps in that directory, I have mass-tagged them all at once, without any need for assistance from the UI. Merely copying the files serves to categorize them.

Now, take backups. If I don't know where all my files are, how do I know what needs backing up? Yes, the UI can help with a dedicated backup tool, but suppose some random app drops files in a location the UI tool doesn't know about... what then? If I know where my data is, I can feel confident in my backups. If I don't, then I have no idea.

And suppose I lose a particular file and want to restore it from backup. How do I do that? I don't know where it is, either originally or on the backup volume. I'm drowning in a sea of tags with no location information.

The human brain seems to be hard-wired for hierarchical classification systems. Try to convince biologists to abandon their classification system for species. Sure, there may be disagreements about where on the hierarchy a particular species belongs or even small disagreements about how the hierarchy should be organized, but no-one seriously proposes abandoning it. It has proved incredibly useful. Similar classification systems abound in most human endeavours.

design is hard.

Yes, it is. And if you plan on bucking the trend established by decades of computer science and hundreds of years of human progress, you'd better have a pretty good flame-proof suit. :)

(Log in to post comments)

Hierarchical organization

Posted Jan 30, 2013 16:55 UTC (Wed) by aseigo (guest, #18394) [Link]

> Organizing it with tags assumes that we'll actually take
> the trouble to tag the photos

I think you're assuming tagging is harder than dragging it to a folder. In the Files app, it isn't. It's the exact same motion: select, drag, drop. Voila. (There's a "New" entry at the top to create new tags as well)

> I can shut out the rest of the world from my visibility, reducing clutter.

exactly how tags work. press on the tag(s) you care about and voila, the rest of the world is shut out from your visibility

> Merely copying the files serves to categorize them.

"default tag as part of the copy process" is one of the workflows we've sketched out but haven't yet fully implemented. it's pretty easy already, but once that is completed it will be as easy as your "copy 'n dump"

> Now, take backups. If I don't know where all my files are, how do I know
> what needs backing up?

files remain in your home directory. they aren't magically hidden in a wormhole between the sectors on the disk ;)

and for such backups, i'd recommend taking the metadata with the files as well. (well, for the metadata that isn't already in them, of course)

> UI can help with a dedicated backup tool,

all backup tools are "dedicated backup tools" in the sense you describe. except they are dedicated to raw fs hierarchy. i agree we need some spiffy backup/restore software (first syncing tools are landing in PA4, btw), but that will be no different than any backup software used.

the caveat indeed is that if you are using the metadata system, and then you want to go around and micromanage manual backups it won't be as smooth as it would be. that's a definite trade off.

the goal / assumption is that in the case of Active by using the system that, um, you're using you won't need to go to those lengths. which leaves your use case in the "edge case" category.

i will be the first to admit that the tablet interface in Active may not be for everyone. we're not trying to make something acceptable for everyone, because that usually means great for, statistically speaking, nobody.

if you find the semantic metadata system so inconvenient on your tablet (though i still suggest you may be surprised!), then i can offer a few options: don't use Plasma Active on it ;) .. or use (or make!) a more "normal" touch friendly file manager (i already know of one being written with QML, btw) ..

> but suppose some random app drops files in
> a location the UI tool doesn't know about

no such thing. all user-writable files are indexed.

> And if you plan on bucking the trend established by decades of computer
> science and hundreds of years of human progress,

ah, but you see here's where you're rather wrong. hierarchical file systems were designed for technical uses and users and within the limitations of computers (both in terms of hardware performance *and* in terms of data volume and variety). computer science has moved on some time ago from hierarchical systems. we merely need to point to google.com to find an example.

what hasn't changed is the implementation facing the user. this is particularly noticeable on mobile devices where there have been two approaches: ignore it and expose hierarchies all over the place; try to hide the file system and eviscerate the applications in the process (which you can get away with on systems that don't have high file management demands such as phones, at least for most people). mobile is a distinct use case from the desktop and the hierarchical systems just become more and more awkward there.

besides a couple decades of accumulated usability research to touch upon, we actually do user testing. at times that testing comes back with negative results and we adjust accordingly until we get positive results. so we're not shooting entirely in the dark, nor creating a self-serving echo chamber. we're probably more critical of what we do than most on the outside ... which leads to results that we are relatively confident about once they get this far :)

> you'd better have a pretty good flame-proof suit.

yeah, no problem there.

Hierarchical organization

Posted Jan 30, 2013 18:09 UTC (Wed) by dskoll (subscriber, #1630) [Link]

I think you're assuming tagging is harder than dragging it to a folder.

In my workflow, it is. Here's how I copy files:

cp /media/disk/*.jpg /target

I don't use "motion", "select", "drag" or "drop". And I use "directories", not "folders", so get off my lawn! :)

I think the rest of your post is simply your opinion (as, I suppose, are my points.) So we will have to agree to disagree.

Hierarchical organization

Posted Jan 30, 2013 20:11 UTC (Wed) by jospoortvliet (subscriber, #33164) [Link]

seriously, you use only command line on your tablets? Oh, you're not talking about a touch-only interface? Then what are you arguing about, as you must have noticed that Aaron was very clear on the fact that Plasma Active is for touch devices, while Plasma Desktop is for - surprise - desktops...

Hierarchical organization

Posted Jan 30, 2013 23:38 UTC (Wed) by dskoll (subscriber, #1630) [Link]

seriously, you use only command line on your tablets?

I don't own a tablet. I do own a Nokia N-900 touchscreen phone, and while I don't use only the command line on that device, I do use it quite a bit.

Plasma Active is for touch devices

The N-900 is a touch device, but still has a hierarchical file browser for which I'm grateful.

Hierarchical organization

Posted Jan 30, 2013 18:11 UTC (Wed) by dskoll (subscriber, #1630) [Link]

I see you did not reply to one of my questions:

How do I restore a specific file from backup without knowing where in the directory hierarchy it is?

In my experience, this is an uncommon thing to want to do, but not so rare as to be labelled an "edge-case".

Hierarchical organization

Posted Jan 30, 2013 18:32 UTC (Wed) by raven667 (subscriber, #5198) [Link]

Actually I would say this is the most common case. In my experience, accidentally deleting an important file is the leading cause of restores from backup.

Hierarchical organization

Posted Jan 30, 2013 18:54 UTC (Wed) by hummassa (subscriber, #307) [Link]

IMHO the answer to this question is: "the backup/restore tool must offer the same view of the backed-up data as the viewing tool offers for the original data".

Hierarchical organization

Posted Jan 30, 2013 19:21 UTC (Wed) by dskoll (subscriber, #1630) [Link]

IMHO the answer to this question is: "the backup/restore tool must offer the same view of the backed-up data as the viewing tool offers for the original data".

That's fine if all the systems you might conceivably connect the backup device to run the same software. But once you move beyond hierarchical directories, that is unlikely to be the case.

Yes, a hierarchical directory structure full of files is the "lowest common denominator". But importantly, it's the lowest common denominator and is likely to be useful on any device or desktop.

Hierarchical organization

Posted Jan 30, 2013 18:34 UTC (Wed) by dskoll (subscriber, #1630) [Link]

computer science has moved on some time ago from hierarchical systems. we merely need to point to google.com to find an example

I'm not sure what you mean by that. Could you elaborate?

If you mean Google as a search engine, then yes: Hierarchical classification systems are not useful for searching. But doing a Google search is in no way analogous to organizing my files and documents.

Hierarchical organization

Posted Jan 30, 2013 20:16 UTC (Wed) by jospoortvliet (subscriber, #33164) [Link]

It's 2013 and you claim to not know what Aaron means when he says we've bumped into the limits of hierarchical organization, pointing at Google... You must've noticed the demise of 'hierarchical search engines' on the web?

Search is search. You can use a hierarchy to find things, but this scales badly. That's why computer science has moved on to other search methods - indexing and analysis of links between pages is what google used to do. Full text indexing, metadata-extraction, semantic analysis, tracking of user habits and explicit tagging is what Plasma Active uses. These technologies might be a tad more complicated than a simple hierarchy but they scale beyond the 1.44 mb floppy disk.

And again, in case you still don't understand it - nothing gets lost, as 'folders' (hierarchical or not) are nothing else than exclusive labels.

Note how Gmail (which uses tagging) works just fine with IMAP (which uses folders): mails which have multiple tags just show up in all the 'folders' in traditional UI's. Works just fine.

Hierarchical organization

Posted Jan 30, 2013 21:45 UTC (Wed) by dlang (✭ supporter ✭, #313) [Link]

> Note how Gmail (which uses tagging) works just fine with IMAP (which uses folders): mails which have multiple tags just show up in all the 'folders' in traditional UI's. Works just fine.

however, you note that people still use the concept of folders (and sub-folders) to navigate the tags.

You could do exactly the same thing with traditional IMAP servers via single-instance-store of the file at multiple places in the filesystem.

Hierarchical organization

Posted Jan 31, 2013 3:18 UTC (Thu) by raven667 (subscriber, #5198) [Link]

I'm not sure what you are describing is functionally or technically different than tags, it's just semantically different. Well, trying to shoe-horn a filesystem (with hard-links for example) into a tagging system brings along baggage such as enforced parent/child relationships that artificially constrain and complicate the system and should probably be worked around by using something designed for tags.

Hierarchical organization

Posted Jan 31, 2013 4:58 UTC (Thu) by dlang (✭ supporter ✭, #313) [Link]

my point is that tags are not something magic that achieve things that could never have been done before, and they don't require you to throw away the filesystem info.

Tagging is extremely useful to supplement filesystem location info, but trying to have it substitute for that info is wrong.

Hierarchical organization

Posted Jan 30, 2013 22:43 UTC (Wed) by dskoll (subscriber, #1630) [Link]

Search is search

Search and categorization are two separate things. My collection of files is not a massive glob of data that I content-search very often.

I'm all for indexing, full-text searching, etc... bring it on!

But don't hide the hierarchy as if it's some ugly stepchild.

Hierarchical organization

Posted Feb 1, 2013 20:41 UTC (Fri) by pboddie (subscriber, #50784) [Link]

The funny thing is that the Internet as searched by Google is organised hierarchically. It's just that Google exposes it through terms produced by their indexing activity. And when someone wants to access a specific resource, and especially update such a resource, they typically don't navigate to it by trying to fish it out using a combination of search keywords. It isn't a matter of "scp article.html editor@[linux,news,lwn,latest]" and hoping that it connects to the right place.

One can argue that this hierarchical organisation is imposed by physical constraints and that beyond the first level, one could use tags, terms, categories or whatever just as successfully as a path to a particular resource, but the issue is convenience for the specific task being performed.