Hacker News Reader

Comment by dexen | Hacker News Reader

1990's called, they want their technology back. [1]

It's been tried times and times again. I've seen it on Symbian on SE P910 (virtual directories per file sense, like video, audio etc), Windows 95 & later(`My Computer'), KDE, N900. You have seen it on other software, and probably will see a few more times.

The `no filesystem' is a leaky abstraction [2]; sooner or later programs start exposing files and directories, because that's what people understand and want to use. `No filesystem' is flimsy magic, files feel like physical objects you actually own. `No filesystem' is the limited menu of your DVD, files and directories are your house with rooms with drawers with shelves with boxes with items.

Your way.

[1] http://en.wikipedia.org/wiki/Symbian

[2] http://c2.com/cgi/wiki?LeakyAbstraction

View on HN

"No filesystem" is only a leaky abstraction when people rooted in the 1980s write filesystem software and try to wrap it in a filesystem-less interface.

Files and directories are what you--and people like you--like and want to use. The rest of us don't give a damn about them, especially the way you build them for us.

Now I will return to your snarky quip. Your contention is that if something failed in the 1990s, it will necessarily fail in 201x? This makes no sense to me whatsoever. Almost everything that succeeds today failed in some form in the past, it's just that times have changed and things previously impossible or impractical are suddenly reasonable.

The whole point of $500 thingummies is that they exist in a very different world than $2,500 whatchamacallits.

> Your contention is that if something failed in the 1990s, it will necessarily fail in 201x?

Nope; the reasoning for the inadequacy is based on the fact you can't please users by including just 80% of features. Sure they use 80% at most, but everybody has different 80% [1].

A collection of objects (here: files) can only (reasonably) be a: 1) set, 2) directed acyclic graph - a.k.a. tree, 3) relational database (in the general sense)

1) is pretty much pointless,

2) is what my mom and little sister understand and use as it requires little querying,

3) requires a powerful query language (that's what right & comfy for us nerds, right?).

So far the filesystem-less software has been two-prong: mounting filesystems at certain pre-designated mountpoints (Windows & friends), nevermind that; and presenting a querable database with a bunch of predefined queries -- AND NO WAY OF CREATING ANY MORE QUERIES. That's the 80% your users aren't looking for.

Now if you come up with a query language that's quick and easy to use from GUI, I'll buy you a week-long supply of your $FAV_BEWERAGE, cross my heart. (drop me a line at dexen.devries + the usual gmail suffix).

But it will not `look like' -- semantics-wise -- as the curret breed of Symbians and iOSes. QED.

[1] http://www.joelonsoftware.com/items/2006/12/09.html

Well, how I could I possibly argue with a guy who worked at Microsoft and by extension, with the argument that produced the design of Microsoft Word? Is that what you are selling me, a future where every tablet includes every feature of a PC right down to having a filesystem I can format? I don't want one. You go ahead, be my guest.

This argument attempts to bludgeon me with logic into believing that filesystems are somehow natural, when nearly forty years of hands on experience with computers tells me otherwise. They aren't natural for a guy like me, only a mathematician would suggest that a DAG is the right answer for storing all the world's information. Your argument falls into the Turing Tar Pit of embracing a single design where everything is possible but nothing of interest is easy.

Now about my sister. Consider iTunes: She can type the name of a song and listen to it right away, that's my own Joel Spolsky quote. She can look at her albums. She can look at her artists. She can look at the songs she's purchased. She can look at the songs she has added to iTunes in the last few weeks.

If you want to call it a database, fine. If you want to call it a hierarchal filesystem, fine. If you want to call it a set, fine. I don't care what you call it under the covers, from my sister's POV, iTunes is NOTHING like looking at song files in the Finder.

My sister also likes her third generation iPod. She has never once complained that it doesn't have music files on it. If your sister wants a filesystem, I'm sure somebody makes music players with visible filesystems where she can be comfortable, and I'm ok with that.

Now I think I am going to go my own way and wish you much happiness as you go yours. If you would like something pretentious to ponder, here is my post on what's wrong with OOP:

https://github.com/raganwald/homoiconic/blob/master/2010/12/...

I suggest to you--with absolutely no justification whatsoever--that organizing programs as towers of classes all inheriting from each other is isomorphic to organizing deeply nested hierarchies of files.

Another poster [1] in this thread made better argument than mine: it's not that non-hierachical file organization is bad. It's good (especially for small, homogenous data sets, I guess).

For other data sets it can be improved, and it will be: by adding support for somewhat hierarchical structure. They did it a few decades ago and it will happen again and again.

The example of music is very, very specific [2] -- all files come with diligently filled-in metadata, and all files have same metadata fields. Each song can be uniquely addressed by (artist, title, version) and mostly uniquely by some other tuples.

But that won't work very well for average Excel and Word files, at least not in the way they are created right now; they largely miss metadata or it differs from file to file.

MS Word? Creeping featurism, bullet-point engineering, yea. They painted themselves into a corner. They broke out of it with Ribbon UI. I like it :D

raganwald, I'm terribly sorry for being irritating in the above posts; I'll try to social up :-)

[1] http://news.ycombinator.com/item?id=2301548

[2] http://news.ycombinator.com/item?id=2301942

I don’t buy it. I haven’t touched any of my 4,822 audio files in the filesystem in years. They are in my music library and I couldn’t care less about where on my drive they are or how they are organized there. The same is true of all my photos. The abstraction works perfectly for me in those two cases, I’m missing nothing.

(My parents would be delighted if Word or Excel had a similar library of documents.)

The media library presents you the files pretty much as an OLAP cube. That's possible only due to metadata diligently assigned to music and photos (location, geotags and date and time). It makes much sense to filter by one or two attributes to pick the music for to-day. Try testing your media library with all the metadata removed...

Since the Word and Excel documents often have no such metadata, and the possible attributes are numberous, it's no wonder no `document library' got popular as of yet. The best you get -- the state of the art of desktop search -- is full-text-search with some semantic analysis, right?

My parents (the only, well, “casual” users I know really well) have all their documents in one folder with no hierarchy whatsoever, any sort of library would improve their experience. There is enough metadata to make it useful, I think. There is the time of creation, the time a document was last edited, its name – that’s more or less just as much metadata as I use for browsing my photos. (I don’t spend much time organizing my photos and my photo library is not very powerful – I cannot, for example, only display all photos taken with a certain shutter speed. Despite that I would really never want to go back to organizing my photos in the filesystem.)

A sort of document library could be just as good or better than the filesystem for managing documents, I don’t think that you have to end up with a leaky abstraction. (Browsing documents on Google Docs, for example, doesn’t feel wrong but you are not using a filesystem to access your documents, you are using a sort of library view.)

I think it’s kind of astonishing that no one has really tried to do just that. There are half-hearted attempts but one of the things about abstractions is that they have to be good to be successful. It’s certainly not easy but I think it’s possible.

Reply on news.ycombinator.com