Wednesday, September 07, 2005

File browser

What I want in a File Browser(tm). Or what I don't want, and I think would be a mistake.

Some obvious things are speed. Instantaneous display of folder contents is an absolute. If it takes as long to display a moderately sized folder as it does to open Konsole, Konsole wins.

I don't want a gui designer to tell me how I should organize my files. Design decisions force usage. If a small number of files in a folder work best, then the designer is telling me I should work that way. No.

The current thinking is to somehow model real objects to help us understand simple concepts (We spent a few years learning to read, but we really don't want to use that). In the past, any large amount of data that needed to be sorted, stored or retrieved had some intermediary who did that for a living. Think executive secretary, librarian, the ancient scribes and secretaries. For what I have in my home folder and descendants, I would probably have needed two or three people to keep it for me. The objects I can keep track of in spatial memory are few. The rest are thrown in closets or piled on my desk waiting for me or someone to sort through them. Any screenies of 'spatial file manager' conveniently shows 3-6 objects in each folder. Spare me. What we need to model is the parsing, sorting and retrieving capabilities of a good librarian/secretary.

A good example of this is iTunes, although I ran into a flaw within 15 seconds which renders it unusable for me. It models a very narrow minded librarian who cannot fathom anyone being interested in anything but music. For music it works well, with a few hints allows you to find what you want. But it only works with a very narrow data type. Let me explain.

I heard that iTunes had audiobooks. And indeed they do. So I installed iTunes for Windows using Crossover Office. It is an older version, so maybe things have changed a little, but it illustrates the point. I found the audiobooks section, clicked on it and got a bunch of categories; non-fiction, history, mysteries, etc. Great. So I click on Non-Fiction and what do I get? A list of authors. Bah. You have to select each one individually to get the book title. If you search on something, it returns music selections and audiobooks. There may be a way of fine tuning this, but frankly I gave up. Life is too short to try to shoehorn a piece of software vision into my reality.

So a file representation that works well with one type of data may work horribly with another.

Another very irritating thing, this time about Konqueror, is the difficulty in narrowing the displayed files. In my home folder, if I'm looking for a pdf file, I can conveniently type /home/derek/*.pdf. I end up with half the screen showing subfolders, with my pdf files at the bottom. That is full sized screen. If I want to split the screen in two for easy moving, all I see are folders. This is where I look for the little X button and open Konsole.

By the way, is there any way using the mouse to do a wild card sort on a particular metadata? In list view there a limited way, but icon view?

I want an easy way to drill down into non-hierarchical data, with an obvious and very easy way to back out. I want to be able to use applications to handle their data effectively, without knackering the file manager. I want the file manager to recognize and respect the hierarchical structures that I build, at the same time ignoring them when I'm trying to find something. I want the file manager to recognize my skills of pattern matching and filtering out unimportant data.

If you have ever worked with a good secretary, you know what I want. Remember that job we did over at so and so's? I want to see the engineering calculations I did, or didn't we have a mechanical failure similar to the one we had yesterday? Who did we talk to about that? A good secretary will have the answers for you in a very short time. I want my computer to do the same thing for me. Let me throw things on her desk, and have it at my fingertips. But don't touch this pile. I'm working on it right now.

I don't want to have to enter a bunch of data to have this happen. If I download a document, create a spreadsheet or KWord document, I don't want to have to fill in a bunch of metadata. I would prefer not to have to create a file name either. The computer, like the secretary, knows (or should know) what I'm working on, what the file is about, where it came from, and where it should go. Sometimes I may want to be specific, but most of the time not. The computer can read, can't it?

Being fussy and miserable, damn the computer that does things without telling me, and doesn't allow me to override it's decisions.

I haven't found a graphical filemanager interface that matches the power and flexibility of the command line. I suspect that most of the audience here hasn't either. Shouldn't a good gui filemanager be far better than the command line?


Comments:
Correct me if I'm wrong here, but when you say "the power and flexibility of the command line", don't you mean "it does EXACTLY what I tell it, because I tell it EXACTLY what to do".

Surely that is the opposite of your secretary analogy, where you expect to give instructions/hints which the secretary which get parsed and processed into more complex/fully evolved instructions without your help.
 
"The computer can read, can't it?"

No, it can't, or not much yet, and that's the main problem. We barely have syntax and grammar checking for a bunch of latin-character-based languages; current AI implementations know nothing about your social and cultural environment and its conventions... Statistical analysis of language patterns and linked behavious has just started to be applied on an industrial basis with reliable results, mainly in limited contexts like email. The usual "automated secretary" utopia you describe is still far in the future, even though we'll probably get there one day. But good luck telling your secretary that XSLT files are not the same as OpenOffice spreadsheet documents...
 
"The computer can read, can't it?"

No, it can't, or not much yet, and that's the main problem. We barely have syntax and grammar checking for a bunch of latin-character-based languages; current AI implementations know nothing about your social and cultural environment and its conventions... Statistical analysis of language patterns and linked behavious has just started to be applied on an industrial basis with reliable results, mainly in limited contexts like email. The usual "automated secretary" utopia you describe is still far in the future, even though we'll probably get there one day. But good luck telling your secretary that XSLT files are not the same as OpenOffice spreadsheet documents...
 
"The computer can read, can't it?"

No, it can't, or not much yet, and that's the main problem. We barely have syntax and grammar checking for a bunch of latin-character-based languages; current AI implementations know nothing about your social and cultural environment and its conventions... Statistical analysis of language patterns and linked behavious has just started to be applied on an industrial basis with reliable results, mainly in limited contexts like email. The usual "automated secretary" utopia you describe is still far in the future, even though we'll probably get there one day. But good luck telling your secretary that XSLT files are not the same as OpenOffice spreadsheet documents...
 
halcyon:

No. I mean it is very quick and concise. Very easy to show me very narrow datasets quickly. An example comes to mind. Do you have a /dev/pilot file? ls /dev/pilot gives the answer in seconds. I wouldn't even try that with any graphical interface and expect to be done within 1/2 an hour. Another more common approach. I download a tar.gz to my home folder. I save the file to it's default, ie. whatever Konq suggests. What is that file name? So much for spatial skills, I don't remember. So I ls *.tar.gz. I find the name, go to Konq and use it's fantastic capability to open a tar like a file folder.

Using the command line I can quickly, with a few commands narrow the data set to something managable. I had one the other day. I've got a bunch of pdf's, manufacturer data sheets. I needed one particular sheet. I have 9 pdfs for that piece of equipment, as indicated by the filename, but didn't remember which one. I did a grep phrase *.pdf, narrowed it down to 4. grep another more precise phrase *.pdf and it narrowed it down to a couple. Then I opened kpdf, browsed the two and found what I wanted.

I really don't think that these are uncommon usage patterns. Nor am I a whiz at the command line. And none of these instances are examples of precision.

Derek
 
I think the first thing to do is make a distinction between searching and filtering. If we look at databases, a search is a query, while a filter is a view. There are significant differences, we might express them in the same way, but one is an action upon a search result, yes it's just a recursive search, but the distinction seems natural none the less.

For some of the stuffs you're mentioning, well that's tenor's job.

As for what can be done with konqueror now, well it'd be nice if it could search and filter.

I think spatial browsing is best left not for dirs but for search results. Each search being though of more as a context, browsing to a dir is setting up the context. Thus the user defined hierarchy is somewhat respected.

*tangent* FFS when are we going to make applications the second class citizens they should be, except for games where applications are the star, every other time they should be largely nameless data/document/information viewer and/or editors. *end tangent*
 
Post a Comment

Subscribe to Post Comments [Atom]





<< Home

This page is powered by Blogger. Isn't yours?

Subscribe to Posts [Atom]