The first hypothesis about the UI that is supposed to be verified in GnomeShell usability tests (http://live.gnome.org/GnomeShell/Design/UsabilityTesting/PhaseI) is the following:

1. A persistent window list is distracting to the user. A user will be less distracted if the window list is not always visible on the screen.

How about another hypothesis:

1a. A user would prefer to be able to switch between running applications with just one click instead of two

IMHO the taskbar is one of the best achievements for desktop UI design, I remember how much of a difference it made when moving from Windows 3.11 to 95.
The decision to get rid of the taskbar is similar only to Microsoft’s decision to get rid of the application menus and replace them with a loosely structured collection of icons (the ribbon UI)

Here’s a sample of what it might be like:
http://farm4.static.flickr.com/3550/3495515137_47b1d16603_o.png

(But please don’t start complaining about it, this is just one suggestion out of many :). Check the mailing list’s archives if you want more information about them all.).

I’d like to echo this comment.

There’s an add-on for removing tabs from Firefox (Tab Killer), at which several people have commented, “Why would you want to do that?!”. There’s an assumption—largely valid—that window managers are all broken.

But with a decent (or even *good*) window manager, in-application tabs would become a hindrance.

Whoah, you said it so much better than I ever could. I don’t think that the user wants a list of windows. The user wants to switch between tasks, not between windows (or tabs). This is a really hard problem to solve, but if it is done right, there’s a lot of potential to make Gnome better (than any current desktop environment).

I can think of a few reasons I prefer tab bars to most window lists. Tab bars have close buttons, and they can do so safely because any accident is easy enough to undo, websites being good about autosaving their state. They display emoticons, which KDE’s window list, when combined with Konqueror, does too. They dispense with the modality mentioned by Greg Nicholson, and with window management futzing generally, by always maximizing their contents, which on any monitor I’ve ever used is very rarely a problem. (On a larger monitor, it might be cool to have predefined spaces, i.e. adjacent virtual desktops, always fully filled by their contents.) Chrome’s tab bar has nice animations too, and smart rules for grouping related tabs, and a cute little new tab button that when combined with their blank tab page and the flashing cursor in the “omnibox” is effectively a task launcher.

For what it’s worth, in the larger debate over the future of the graphical shell, I think tab bars and window lists are both great, and both are unfairly maligned (though I do agree that internal tab bars represent a failure of window managment, which should be done right at the right level). To an eye accustomed to reading they’re much easier to scan than an arrangement of miniaturized pages.

Mr. Odd: that’s a feature I hadn’t known about. Cool!

An old version of the Ubuntu Netbook Remix window picker got this right (although the latest version gets it wrong again).

2. New items appear at the end, instead of alongside similar items. (For example, new Firefox windows don’t appear next to other Firefox windows.) Windows XP got this right.

3. The window list doesn’t use all the available space—this makes Fitts’s Law cry. In the simple case, when there are n buttons, each should occupy 1/n of the available space.

Safari 4 beta’s tab/title bar gets this right. (And there’s a Windows XP registry tweak to accomplish the same.)

(3a. Cleverer would be to make more-relevant buttons wider, whilst still occupying all of the available space overall.)

I know some people say the window-list is broken. I’m rather used to it, but I don’t mind innovation – if it works. I do, however, find the window-selector panel plugin a useful addition to the window list (I do wish there was some hotkey to bring it up).

Take a look at the different ways applications function with regards to windows. Some have no main windows, and so with today’s environment have no way of “existing” to the user unless they ‘abuse’ the notification area or make use of deprecated component frameworks. (And don’t get me started on panel applets. That is a prime example of consistent user interface being pushed over by implementation details. Why should the user have to care whether a timer application appears in the panel or a window? With panel applets, we end up with two methods of launching applications with the only functional difference being that panel applet applications never present themselves beyond their launchers). If we are going to have “open window = application,” why don’t we come up with a framework to represent /applications/ instead of all the guessing?

An application window can become urgent, triggering a fancy glow effect in the window list applet. The notification daemon (libnotify) stuff has made that look redundant. Canonical’s “Messages” panel applet coming in Jaunty has the same problem.
The window list provides us with notifications in the same style as libnotify. (I hate to call them notifications, but it’s become the trend. It’s really more like a shelf). In this case, the window list gives us rather passive notifications that things are happening in the background. Each notification has a variety of actions through the right click menu (all provided by libwnck) and a primary one through a single left click. They can also be dragged and dropped since they’re all linked to relevant content types.

The user is driven to look to the window list when windows appear and start yelling for attention, so why not when other objects like emails and people and downloads start appearing? What, other than feeble implementation details (no windows exist for the objects) say that these things should be panel applets or should operate through actioney notification bubbles?

If notifications are meant to be quick and easy and helpful, they should not be happening in six places at once. It’s like having six over-zealous assistants all competing for attention as you try to get work done. (The notification bubbles are winning, by the way). One becomes more interested in strangling them than in actually listening to what they have to say.