[ge-talk] Search, Don't Browse

[Waldemar Kornewald]

Ville Koskinen wrote:
> I find the idea of a pure database as the filesystem thrilling, but
> also a bit scary. I'll be eager to try it out some day.

This FS-as-DB suggestion comes up every few months on this list. :)
The idea is not new, but as you already said, it's very difficult to get 
right. We will need new ways of information representation. I'm not 
seeing this as simple lists of files matching some query. There are 
cases where simple categorization (music files by artist or genre) is 
sufficient, but we need better ways to display complex data relations.

I'm not suggesting a solution (very difficult). This is just a list of 
ideas and TODOs that should be solved before we can seriously consider 
dropping the directory-based metaphor:

1. Automating Categorization (directories are too manual)
How do automate data categorization (text, audio, video, ...) in such a 
way that a query can reliably find what you're searching for and at the 
same time without requiring the user to enter all this information. This 
is a very big problem with non-textual data.

2. Browsing Data (directories are too statically structured)
What is the fastest and easiest way to create queries (possibly based on 
existing queries) and get from one query/file to another query/file with 
related data? This must at least be comparably fast&easy as browsing 
folders!

3. Visualizing Query Results (directories are restricted to lists)
Is it sufficient to display a list of file icons?
When are ZUIs better (e.g.: summarize content based on zoom level)?
Which visualization methods from knowledge systems could we reuse?
Would 3D help? What about 2+1/2D solutions (ZUI "hybrids")?

One desired property is the concept of static (at least approximate) 
"location". Preferably, this "location" should be spatial instead of 
textual, but at the same time it should not make it more difficult to 
reorganize data (use queries to change the visualization).

-----
possible additions that might become important for this solution:
-----

4. Data Validation
Files are stupid and static structures. Without an application it is 
impossible to validate the correctness of meta-data. For example, when 
edited from within Tracker, your Person files (contacts) can't check if 
the email address you entered is a valid email address and whether the 
phone number you entered is bogus. Also, we need a good way to enter 
(partial) date, time, and time range (and other complex) information. 
All this requires application logic to be part of your data if it should 
be done correctly and reliably.

5. Persistence
When you close a window (file/data) and open it again the previous state 
is preserved. You don't save changes made to your data, anymore. Use 
checkpoints to mark important stages/states of your data. Use the 
"unlimited undo" to revert changes you do not like. The state of the 
whole system is automatically saved and preserved between reboots 
without requiring the user to do anything.

-------------------

Maybe someone has enough time to think about a good solution? There are 
more problems than I can mention here. Some use-cases:
* big development projects (*many* files...)
* burn a CD/DVD
* synchronize contacts and appointments with your mobile/PDA
* files belonging to multiple projects
* pick one or multiple files (e.g.: backup, copy, send via mail)
There are very complex use-cases that can't be solved with simple 
queries and even simple categorization/tagging. Directories can handle 
nearly anything (even if it gets ugly), but queries can make effective 
work impossible.

Let's stop thinking in terms of simple lists of data/files (queries, 
Spotlight, Google). This is not enough to drop directories and the 
desktop metaphor.

Somehow I have the impression that I wrote a lot of text, but not a lot 
of information...maybe next time. :)

Bye,
Waldemar