A Spatial File Browser

spatialBrowser is a file browser offering a dual view of files: a traditional tree list view, with expandable/collapsible nodes, and a 2D spatial view, where files can be freely arranged. The 2D spatial view shows the contents of the currently selected folder, and can be panned and zoomed with the mouse. The spatial layout of files within each folder is persistent across sessions. For some of the rationale behind the design, and references to relevant literature, go here [1].

You can download the source code [2] (about 9k, or 1200 lines of code) or a pre-compiled binary [3] for Linux/i386. The software was developed on Redhat 7, but should be easy to port to other GNU/Linux platforms, or even to Windows with a little more effort. Required libraries are Qt [4] (for most of the UI), STL (for templated lists), OpenGL (for the spatial view), and GLUT [5] (for the scaled fonts in the spatial view). I used Qt 2.2, however an older version may suffice. Mesa [6] can of course be used in lieu of OpenGL.

To unpackage the source, compile and run, try

   gzip -d spatialBrowser-0.1.tar.gz
   tar -xf spatialBrowser-0.1.tar
   cd spatialBrowser-0.1
   ln -s Makefile.linux Makefile
   make
   ./spatialBrowser [directory]

The argument to the executable is optional. If none is provided, the browser will start up with the current directory loaded by default.

In the spatial view, use the left-mouse-button (LMB) to move files, middle-mouse-button (MMB) to pan, and the LMB+MMB together to zoom. Folders are drawn in cyan, and regular files are drawn in white.

Note that if you arrange any of the files in the 2D spatial view of the browser, a hidden ._ (dot underscore) file will be saved containing the layout. One ._ file will be saved in each directory that is given a spatial layout by the user. Directories that initially contain no ._ file are given a default layout by the browser, and the layout is only saved if the user modifies it.

spatialBrowser was implemented during March and April of 2001 as part of CSC 2537, a computer science graduate course on hypermedia given at the University of Toronto. In the interest of rapidly generating a first prototype, the source code for v0.1 was based on a Qt example program called dirview which demonstrates how to use Qt list views. (As such, the source code for v0.1 is rather messy and should probably be rewritten from scratch.)

Things left to do:

• At the moment, the spatial view serves only to view and arrange files in space. The user cannot invoke any actions on files in the spatial view (such as going into directory, or launching an editor), which makes the spatial view nearly useless. The spatial view should be enhanced so that users can double click on files to launch appropriate applications, or double click on directories to go into them.
• Multiple selection should be implemented for the spatial view, so that users can shift-select or rectangle-select multiple files and move them within the 2D space as a group.
• The browser should support copying, deleting, and moving of files from one directory to another. It should also support the creation of new directories.

Possible future enhancements:

• When navigating from a folder to a subfolder or parent, it might prove pleasing for many users to have the spatial layout briefly but smoothly animate the transition from folder to folder (i.e. "zooming in" to a subfolder or "zooming out" to the parent). There is at least anecdotal evidence that, in 3D applications, smoothly animated transitions from orthogonal views to perspective views help novice users build a mental model of the 3D scene.

• The default spatial layouts that are now generated are very simplistic. Better default arrangements might be generated using heuristic rules, such as ordering by file name or by time stamp, and by trying to fill space more efficiently.

  It would also be useful for users to have access to snap-to grids or tools for cleaning up spatial arrangements. A user could approximately arrange a set of files, and then select them as a group and apply a "grid-ize" or "circle-ize" tool. Below are depictions of these two operations:

  
  

• It would be interesting to experiment with 3D spatial representations of the files. Perhaps the files could be arranged on a 2D plane (as they are now), seen askance. Files could be represented as boxes whose thickness in z indicates the size in bytes (or some other attribute) of the file. How much more useful information would a 3D representation convey, and would there be any reduction in usability due to the increased complexity of navigating such a space ? Perhaps the browser could allow users to switch between 2D and 3D views as desired.

• Can the UI for this browser be enhanced with elements taken from traditional command-line user interfaces ? Navigation and selection in UNIX shells can be very fast, if the user knows the names of directories or files of interest and can use tab-expansion to rapidly enter filenames. Perhaps a text-field could be added to spatialBrowser's UI where users could enter (with tab-expansion) the names of directories to browse or files to perform operations on. As the user enters directory or file names, the spatial view could dynamically update.

  It would be interesting to study whether there is any difference in the ability of users of graphical and non-graphical systems to remember and type in pathnames. Perhaps users of UNIX shells, being forced to always type in pathnames, are much better at remembering them, and perhaps users of graphical browsers would not be able to remember or enter pathnames as quickly. Would users of a hybrid system eventually learn to type in pathnames very quickly, or would they be discouraged by the initial learning curve and fall back on the slower but easier graphical UI ?

• At the moment, the spatial view only differentiates files with two colours (for folders and regular files) and file names. Unless the user takes the time to give distinctive layouts to the files in each folder, many folder layouts end up looking very similar.

  File browsers such as those in Microsoft Windows or Mac OS represent files as icons or thumbnails to help distinguish them. However, unless icons are carefully chosen or are familiar to the user, having a multitude of different icon images or automatically-generated thumbnails can result in visual noise that is meaningless to the user.

  To reduce the uniform look of the spatial layouts, it would be worthwhile to experiment with having different colour schemes, landmarks (such as 2D or 3D icons, images, or textures), or other visual cues for each directory. Ideally, the visual cues should be noticeable enough to be remembered by the user, but not so obtrusive as to distract or displease. If the visual cues are generated automatically, they should be easy to change by the user on demand.

  (It is instructive to contrast the spatial views of spatialBrowser with webpages on the web: the latter have a very un-uniform look, and thus are more easily distinguished and remembered visually.)

• Along the same lines as landmarks, another way to enrich the spatial representation of files would be to allow the user to create annotations or text labels and arrange them in the 2D space with the files. It might be useful to allow the user to adjust the font size of the annotations, so that spatial groups and supergroups could have large labels that are still legible when the user zooms out.

  It is possible that, by using annotations and by grouping files at different scales, users would be able to manage a larger number of files in each folder, and would have less need to fragment subgroups of files into subfolders.

• To maximize the use of screen space in the spatial view, it would be useful to allow the user to "expand" folders and see their spatial contents nested in the space of the parent folder. This would be especially useful when there are subfolders that contain few files, such as here:
  

  Rather than representing the parent folder Stuff like this:

  

  It could be more usefully shown like this, after the user has expanded the x and y folders:

  

  Having subfolders expanded like this allows children and parent files to occupy a common 2D space, and changes the subfolders into explicit visual groupings. The subfolders essentially become light-weight windows, requiring very little management by the user since they are only visible when the user is browsing the corresponding parent folder.

• A further extension would allow text files (and other documents) to also be expanded within the spatial view, occupying light-weight windows for viewing (or even editing) their contents. This might prove very useful for small documents, notes, etc, or even for browsing the first few lines of a larger document. Having ten documents opened like this would allow the user to quickly scan the first few lines of each and be reminded of their contents, without having to explicitly launch a browser for each document. Users could also use light-weight text editors embedded in the 2D space to create "PostIt" notes of to-do lists or reminders. Since each folder has its own 2D space, and each 2D space extends infinitely and can be panned and zoom, there are no limitations on the amount of ephemeral information that could be stored and browsed in such a system. (Of course, it is an open question whether large amounts of ephemeral information would be manageable.)

  Another possibility, rather than having documents expanded into small windows, would be to have rollover browsing: a cursor that dwells over a file would cause a popup window to appear containing the first few lines (or a summary) of the document, and perhaps by hitting a modifier key, the user would switch into a scrolling mode where mouse motion would scroll the popped-up document up and down.

In current traditional systems, the only way for users to organize electronic documents is to group them into explicitly named subfolders, and/or arrange them on a virtual desktop of limited size. The principle goals in the above suggestions are to provide the user with means for rapid and visual browsing of folders, files, and their contents, and to ease the organization of information (ephemeral or otherwise) by enabling the use of large, spatial layouts that can be as informal or as explicitly hierarchical as the user wishes.

This document is available online at

    http://www.dgp.utoronto.ca/~mjmcguff/csc2537/project/

The links in the online document are lost when printed, and are hence repeated here:

[1]  http://www.dgp.utoronto.ca/~mjmcguff/csc2537/proposal/

[2]  http://www.dgp.utoronto.ca/~mjmcguff/csc2537/project/spatialBrowser-0.1.tar.gz

[3]  http://www.dgp.utoronto.ca/~mjmcguff/csc2537/project/spatialBrowser

[4]  http://www.trolltech.com/products/qt.html

[5]  http://reality.sgi.com/mjk/#glut

[6]  http://mesa3d.sourceforge.net

Michael McGuffin, April 23, 2001