1. The Desktop
1.0.1. Making the Desktop Better
JonoBacon: I think we really need to focus on the integration of the desktop. Currently, the desktop is not really integrated at all. I originally wrote much of this up in this article (http://www.oreillynet.com/pub/wlg/6911). We really need to allow the GNOME desktop platform to hook these different applications together in a way that truly optimises what users need to do with their desktop. This also hooks in with first-level objects. I created a page to discuss this on GnomeProjectDesktop
1.0.2. You don't want to switch desktops. You want to switch tasks, don't you?
JörnDreyer: When do you switch the (virtual) desktop? When you are interrupted and have to do something else: you switch to another free desktop to start working on a new task. So, why not support the concept of tasks? Sure, you could create a desktops for each task you are working on (as someone on this page already desribed, the panel switcher could grow quite large...), but it won't prevent your mailprogram to open an html link in an already running browser window on another desktop. Thats not what you want. You don't even want to see all of your emails in the context for the task you are working on. And you don't want to see all the people online in Gaim. And you don't want to see all your notes and todos. Maybe, you even want to shade out appointments in your calendar that are not related to the task at hand. You can very well organize on one desktop when working on a single task ... running an application and switching to the documentation with good old alt-tab or even let the windowmanager automatically split/tile multiselected windows (just the way you multiselect files witch ctrl + left click - then rightclick on the titlebar to open a popupmenu). Now, boom, your buddy pops in with his great idea and he needs some input so you start googling, reading and searching, starting an editor and hack at it. If you were lucky you switched to another desktop. Basically, you started a new task. Imagine the task remembering the searches you did (beagle), locallly storing any pages you read the emails you sent and the people you talked to on IM. Maybe even the documents you opened. Imagine sending the whole task to your buddy after you managed to convince him he should carry on alone on his OWN computer because you'd like to get on with your own ramblings ... RDF might cone in handy: create relations between two objects. Or maybe even easier: tagging two objects creates a new relation with that tag as relation between them. Or a context menu / shortcut with an "Add to task ..." command. Whatever is selected + added will be remembered by the task. Switching between tasks and context would be a simple desktop switch. Gnome has a lot of features that already support this. the tasks could, early on, simply remember the complete desktop layout with panel location, started applets, application window positions etc. The first thing could be a panel applet that re-/stores, backups and sends the configuration of the currently active (current desktop, currently selected) task (via gconf?). Of course, it would have to be possible to have different panel / applet layouts on different desktops ... not sure if that is currently possible. I'm just trying to give some inspiration here. People want to get things done, and they use a lot of different tools for it. At least mail, instant-messaging, a calendar and some main application. Imagine moving tasks to someone else: simply by sending him a "taskfile" he will see the important contacts in gaim, have the contacts added to evolotion / a more general contacts database, have the relevant applications starters in the top panel, with the needed documents transferred and updated as recently used in the task. The atoms that Gnome should be able to relate to each other would be contacts, emails, desktop configuration settings and documents / files (well if everything was its own file we would only need that ...). That way applications could try to fetch some kind of "context/task environment" and act accordingly, by hiding unuseful information and storing actions in a local repository. Another approach might be to make everything taggable and then let the user define filters based on these tags, though I think having to tag each applet would be overkill ... anyway, its late and I needed to get this of my head.
DanBallard: Reading this some warning bells were going off. Not to say it's wrong, you just have to be very careful with this. MacOSX sort of took this a bit too far and went with a 1 app to 1 task thing and made it not practical to work on tasks that involved more than one app. I found working with MacOSX for any period of time to be very frustrating and limited. Still, that doesn't mean this can't be done right. Just don't assume you know all the use cases of your users and leave it open and extensible (fixable ;)).
JörnDreyer: Absolutely. The whole idea is to make applications task aware and link them to these tasks. Ideally mail and IM apps would show different mails, depending on the task being worked on. Spying on haystack, they implemented some kind of semantic PIM: email, im, calendar, tasks, photo gallery etc. everything groupable. Just take the tour on how you could work with a little more context awareness. This was pre 2003 btw! We could grab some very fine ideas here. With dbus we could e.g. achieve the announcement of incoming mails and im events and show tham in a unified view like here. The idea of working with information rather than applications is interesting too. Why bother with finding the right application when you want to get something done.
JohnNilsson: I just had to add that one of my most frequent use of desktop switching is to find som free space on the desktop. Either for acces to the desktop context menu, or to open some object on the desktop. Writing this I just realized that there is another function for this, the "hide all windows" button, but it has the same signs of trouble. It is a sign of something beeing wrong when you have to hide what you have in front of you to get to the most important objects and functions.
Merging concepts
EtienneBersac : now i read this page, some clear ideas come to mind. Nevertheless, the following line are a raw braindump 
. We clearly intend to manage data object (mail, chat, contacts, document, video, etc.) instead of application (mail viewer, chat client, document writer, video player, etc.).
Merging taskbar, launcher and notification bar : An intersteing point is the dock : the dock merge launchbar and taskbar (with the tiny arrow). When i launch etheir gajim, gaim or liferea, i have at least three time an representation of this app : a launcher (an entry in the menu, an icon in the panel or the desktop), an icon in the notification area and at least one button in the taskbar. This should be very simple and clean to merge all this apparition of the same apps in One representation. the dock from OSX is near this vision but, we still need Exposé to switch between app. Consider a dock with big icons. Each icon is a launcher. If the app is launched, clicking ot the icon just focus on the runing window of the app. This icons have emblem which notify user like in the notification area. If the user drop the mail icon but receive a mail, the mail icon pop in the dock like in the notification area. This imply an application runing in the background and then a lower startup (see update-notifier, etc).
Merging workspace, desk and perspective : Actually metacity implement workspace as 'named group of windows'. this is quite usefull, but not so intuitive for every people and not so easy to manage. Perspective change the way the screen/app is arranged to fit another activity. When i name my workspace, i have one workspace for devel, several for internet (web, mail, news, chat, etc.), one for media (rhythmbox). In the real life, i have three desktop (i'm not so lucky) : one for computer, one for writing, drawing, etc, and another for droping mail,and a bunch of other things. Each desktop has its own tools (computer, CDs, printer, etc./paper and pens, etc./kniffe for letter,etc.). Applying thos two concept (worskpace and desk), We can implement perspective as virtual place where different tools are presented for the user. exemple : the dock does not have the same launchers. The devel desk has emacs, anjuta, eclipse and glade in the doc (replace which Source code, projects, interface in order to manage data, not application). The "discussion" desk has Mail, Web, Chat, etc. in the dock. The "Work" desk has Text, Presentation, Diagrams, Picture in the dock. Switching between desk change tool you want to use. currently, when i switch to workspace "devel", i don't care neither about synaptic nor gimp. These tool are useless in this perspective (develop). Merging workspace and persective should allow the interface to fit the needs and wish of the user. Workspace does not become coercitives, user should be able to launch every app in every workspace. this is why the dock does not replace an application menu even if, imho, the application menu should become a data oriented menu. People do not want to launch firefox, but view web page. I like the naming scheme of the gnome menu "firefox web browser", etc.
Merging this concepts make interface simpler. User do not see applications splitted in several place in the screen. just dock and window. I must recognize that a lot of improvement are inspired by NeXT and Apple. The goal is not to clone, but to go further.
1.0.3. Re-thinking the problem
Chiv: Butting in here, I can offer a unique perspective on what the desktop is trying to achieve. Before anything else, the desktop is a means to organise many different types of information and do many tasks on a very limited screen size. The problem is that you want any given app to use the full screen, but you also invariably need lots of other apps to get any work done. There are lots of good ideas above, but I can give a paradigm that has as far as I know not been considered - why not just ditch every other idea and make the desktop larger than the size of the screen?
It's actually quite simple. All windows take as much space as they need within a super large 'desktop map'. When you put focus on a window, the screen wraps around it. If you want to switch to another, you simply scroll the desktop to the position of the other window. This could be done using a kind of minimap (like the workspace switcher), or an OSX-like Expose (the entire screen zooms out to fit the desktop), or even just a good old taskbar. Do this and now you suddenly have practically unlimited 'work space', and several nifty ways to keep a keen eye over it all. You've seen this method before no doubt, its called 'Command and Conquer', and it was apparently quite popular
1.1. A catalog feature in nautilus
vince: I just thought it'd be very useful to integrate a CD/DVD catalog directly into Nautilus. In this way searches through files and contents would be very efficient, as it wouldn't be needed to open your cataloging software to know which disk you have to pick from your archive. For instance, an expanding icon could be added in the Computer view into Nautilus, saying myArchive, as well as in the Tree view into the Side Pane. File stored in the CD/DVD archive shouldn't be listed as belonging to the disk they were actually burnt into, but to the Archive itself, and maybe listed according to tags or mime. Then, in order to actually open that file, Nautilus would ask the user to insert the proper disk, maybe even suggesting the place that disk should be picked up from. Do you think it'd be possible to simply use code from existing cataloging software?