There is a number of window management features or behaviors which are not specified in the ICCCM, but are commonly met in modern window managers and desktop environments.
The ICCCM allows window managers to implement additional window states, which will appear to clients as substates of NormalState and IconicState. Two commonly met examples are Maximized and Shaded. A window manager may implement these as proper substates of NormalState and IconicState, or it may treat them as independent flags, allowing e.g. a maximized window to be iconified and to re-appear as maximized upon de-iconification.
Maximization is a very old feature of window managers. There was even a ZoomedState in early ICCCM drafts. Maximizing a window should give it as much of the screen area as possible (this may not be the full screen area, but only a smaller 'workarea', since the window manager may have reserved certain areas for other windows). A window manager is expected to remember the geometry of a maximized window and restore it upon de-maximization. Modern window managers typically allow separate horizontal and vertical maximization.
With the introduction of the Xinerama extension in X11 R6.4, maximization has become more involved. Xinerama allows a screen to span multiple monitors in a freely configurable geometry. In such a setting, maximizing a window would ideally not grow it to fill the whole screen, but only the monitor it is shown on. There are of course borderline cases for windows crossing monitor boundaries, and 'real' maximization to the full screen may sometimes be useful.
The WM_TRANSIENT_FOR hint of the ICCCM allows clients to specify that a toplevel window may be closed before the client finishes. A typical example of a transient window is a dialog. Some dialogs can be open for a long time, while the user continues to work in the main window. Other dialogs have to be closed before the user can continue to work in the main window. This property is called modality. While clients can implement modal windows in an ICCCM compliant way using the globally active input model, some window managers offer support for handling modality.
The window manager may offer to arrange the managed windows on a desktop that is larger than the root window. The screen functions as a viewport on this large desktop. Different policies regarding the positioning of the viewport on the desktop can be implemented: The window manager may only allow the viewport position to change in increments of the screen size (paging) or it may allow arbitrary positions (scrolling).
To fulfill the ICCCM principle that clients should behave the same regardless whether a window manager is running or not, window managers which implement large desktops must interpret all client-provided geometries with respect to the current viewport.
There are two options for implementing a large desktop: The first is to keep the managed windows (or, if reparenting, their frames) as children of the root window. Moving the viewport is achieved by moving all managed windows in the opposite direction.
The second alternative is to reparent all managed windows to a dedicated large window (somewhat inappropriately called a 'virtual root'). Moving the viewport is then achieved by moving the virtual root in the opposite direction.
Both alternatives are completely ICCCM compliant, although the second one may be somewhat problematic for clients trying to figure out the window manager decorations around their toplevel windows and for clients trying to draw background images on the root window.
A window manager which implements a large desktop typically offers a way for the user to make certain windows 'stick to the glass', i.e. these windows will stay at the same position on the screen when the viewport is moved.
Most X servers have only a single screen. The window manager may virtualize this resource and offer multiple so-called 'virtual desktops', of which only one can be shown on the screen at a time. There is some variation among the features of virtual desktop implementations. There may be a fixed number of desktops, or new ones may be created dynamically. The size of the desktops may be fixed or variable. If the desktops are larger than the root window, their viewports (see the section called “Large Desktops”) may be independent or forced to be at the same position.
A window manager which implements virtual desktops generally offers a way for the user to move clients between desktops. Clients may be allowed to occupy more than one desktop simultaneously.
There are at least two options for implementing virtual desktops. The first is to use multiple virtual roots (see the section called “Implementation note”) and change the current desktop by manipulating the stacking order of the virtual roots. This is completely ICCCM compliant, but has the issues outlined in the section called “Implementation note”
The second option is to keep all managed windows as children of the root window and unmap the frames of those which are not on the current desktop. Unmapped windows should be placed in IconicState, according to the ICCCM. Windows which are actually iconified or minimized should have the _NET_WM_STATE_HIDDEN property set, to communicate to pagers that the window should not be represented as "onscreen."
A pager offers a different UI for window management tasks. It shows a miniature view of the desktop(s) representing managed windows by small rectangles and allows the user to initiate various window manager actions by manipulating these representations. Typically offered actions are activation (see the section called “Activation”), moving, restacking, iconification, maximization and closing. On a large desktop, the pager may offer a way to move the viewport. On virtual desktops, the pager may offer ways to move windows between desktops and to change the current desktop.
A taskbar offers another UI for window management tasks. It typically represents client windows as a list of buttons labelled with the window titles and possibly icons. Pressing a button initiates a window manager action on the represented window, typical actions being activation and iconification. In environments with a taskbar, icons are often considered inappropriate, since the iconified windows are already represented in the taskbar.
In the X world, activating a window means to give it the input focus. This may not be possible if the window is unmapped, because it is on a different desktop. Thus, activating a window may involve additional steps like moving it to the current desktop (or changing to the desktop the window is on), deiconifying it or raising it.
Some window managers display some form of animation when (de-)iconifying a window. This may be a line drawing connecting the corners of the window with the corners of the icon or the window may be opaquely moved and resized on some trajectory joining the window location and the icon location.
Window-in-window MDI is a multiple document interface known from MS Windows platforms. Programs employing it have a single top-level window which contains a workspace which contains the subwindows for the open documents. These subwindows are decorated with window manager frames and can be manipulated within their parent window just like ordinary top-level windows on the root window.
Some window managers keep the toplevel windows not in a single linear stack, but subdivide the stack into several layers. There is a lot of variation among the features of layered stacking order implementations. The number of layers may or may not be fixed. The layer of a toplevel window may be explicit and directly modifiable or derived from other properties of the window, e.g. the type of the window. The stacking order may or may not be strict, i.e. not allow the user to raise or lower windows beyond their layer.
This spec tries to address the following issues:
Allow clients to influence their initial state with respect to maximization, shading, stickiness, desktop, stacking order.
Improve the window managers ability to vary window decorations and maintain the stacking order by allowing clients to hint the window manager about the type of their windows.
Enable pagers and taskbars to be implemented as separate clients and allow them to work with any compliant window manager.
This spec doesn't cover any of the following:
Other IPC mechanisms like ICE or Corba.
Window manager configuration.
Window manager documentation.
Clients appearing on a proper subset of desktops.
The window manager is supposed to be in charge of window management policy, so that there is consistent behavior on the user's screen no matter who wrote the clients.
The spec offers a lot of external control about window manager actions. This is intended mainly to allow pagers, taskbars and similar window manager UIs to be implemented as separate clients. "Ordinary" clients shouldn't use these except maybe in response to a direct user request (i.e. setting a config option to start maximized or specifying a -desk n command line argument).