Inter-Client Communication Conventions Manual

At the conceptual level, atoms are unique names that clients can use to communicate information to each other. They can be thought of as a bundle of octets, like a string but without an encoding being specified. The elements are not necessarily ASCII characters, and no case folding happens. ^[1]

The protocol designers felt that passing these sequences of bytes back and forth across the wire would be too costly. Further, they thought it important that events as they appear on the wire have a fixed size (in fact, 32 bytes) and that because some events contain atoms, a fixed-size representation for them was needed.

To allow a fixed-size representation, a protocol request ( InternAtom ) was provided to register a byte sequence with the server, which returns a 32-bit value (with the top three bits zero) that maps to the byte sequence. The inverse operator is also available ( GetAtomName ).

The protocol specifies a number of atoms as being predefined:

Predefined atoms are an implementation trick to avoid the cost of interning many of the atoms that are expected to be used during the startup phase of all applications. The results of the InternAtom requests, which require a handshake, can be assumed a priori.

Language interfaces should probably cache the atom-name mappings and get them only when required. The CLX interface, for instance, makes no distinction between predefined atoms and other atoms; all atoms are viewed as symbols at the interface. However, a CLX implementation will typically keep a symbol or atom cache and will typically initialize this cache with the predefined atoms.

The built-in atoms are composed of uppercase ASCII characters with the logical words separated by an underscore character (_), for example, WM_ICON_NAME. The protocol specification recommends that atoms used for private vendor-specific reasons should begin with an underscore. To prevent conflicts among organizations, additional prefixes should be chosen (for example, _DEC_WM_DECORATION_GEOMETRY).

The names were chosen in this fashion to make it easy to use them in a natural way within LISP. Keyword constructors allow the programmer to specify the atoms as LISP atoms. If the atoms were not all uppercase, special quoting conventions would have to be used.

The core protocol imposes no semantics on atoms except as they are used in FONTPROP structures. For further information on FONTPROP semantics, see the X Logical Font Description Conventions.

The protocol defines six distinct spaces in which atoms are interpreted. Any particular atom may or may not have some valid interpretation with respect to each of these name spaces.

Sometimes a protocol requires an arbitrary number of similar objects that need unique names (usually because the objects are created dynamically, so that names cannot be invented in advance). For example, a colormap-generating program might use the selection mechanism to offer colormaps for each screen and so needs a selection name for each screen. Such names are called "discriminated names" and are discriminated by some entity. This entity can be:

    A screen
    An X resource (a window, a colormap, a visual, etc.)
    A client

If it is only necessary to generate a fixed set of names for each value of the discriminating entity, then the discriminated names are formed by suffixing an ordinary name according to the value of the entity.

If name is a descriptive portion for the name, d is a decimal number with no leading zeroes, and x is a hexadecimal number with exactly 8 digits, and using uppercase letters, then such discriminated names shall have the form:

To discriminate a name by client, use an X resource ID created by that client. This resource can be of any type.

Sometimes it is simply necessary to generate a unique set of names (for example, for the properties on a window used by a MULTIPLE selection). These names should have the form:

Ud     (e.g.,  U0  U1  U2  U3  ...)

if the names stand totally alone, and the form:

name_Ud     (e.g.,  FOO_U0  BAR_U0  FOO_U1  BAR_U1  ...)

if they come in sets (here there are two sets, named "FOO" and "BAR"). The stand-alone Ud form should be used only if it is clear that the module using it has complete control over the relevant namespace or has the active cooperation of all other entities that might also use these names. (Naming properties on a window created specifically for a particular selection is such a use; naming properties on the root window is almost certainly not.)

In a particularly difficult case, it might be necessary to combine both forms of discrimination. If this happens, the U form should come after the other form, thus:

    FOO_R12345678_U23

To relinquish ownership of a selection voluntarily, a client should execute a SetSelectionOwner request for that selection atom, with owner specified as None and the time specified as the timestamp that was used to acquire the selection.

Alternatively, the client may destroy the window used as the owner value of the SetSelectionOwner request, or the client may terminate. In both cases, the ownership of the selection involved will revert to None.

If a client gives up ownership of a selection or if some other client executes a SetSelectionOwner for it and thus reassigns it forcibly, the previous owner will receive a SelectionClear event. For the definition of a SelectionClear event, see the section called “Responsibilities of the Selection Owner”

The timestamp is the time the selection changed hands. The specified owner is the window that was specified by the current owner in its SetSelectionOwner request.

There can be an arbitrary number of selections, each named by an atom. To conform with the inter-client conventions, however, clients need deal with only these three selections:

Other selections may be used freely for private communication among related groups of clients.

The atom that a requestor supplies as the target of a ConvertSelection request determines the form of the data supplied. The set of such atoms is extensible, but a generally accepted base set of target atoms is needed. As a starting point for this, the following table contains those that have been suggested so far.

References:

It is expected that this table will grow over time.

Selection owners are required to support the following targets. All other targets are optional.

Some targets (for example, DELETE) have side effects. To render these targets unambiguous, the entries in a MULTIPLE property must be processed in the order that they appear in the property.

In general, targets with side effects will return no information, that is, they will return a zero length property of type NULL. (Type NULL means the result of InternAtom on the string "NULL", not the value zero.) In all cases, the requested side effect must be performed before the conversion is accepted. If the requested side effect cannot be performed, the corresponding conversion request must be refused.

These side-effect targets are used to implement operations such as "exchange PRIMARY and SECONDARY selections."

In general, the encoding for the characters in a text string property is specified by its type. It is highly desirable for there to be a simple, invertible mapping between string property types and any character set names embedded within font names in any font naming standard adopted by the Consortium.

The atom TEXT is a polymorphic target. Requesting conversion into TEXT will convert into whatever encoding is convenient for the owner. The encoding chosen will be indicated by the type of the property returned. TEXT is not defined as a type; it will never be the returned type from a selection conversion request.

If the requestor wants the owner to return the contents of the selection in a specific encoding, it should request conversion into the name of that encoding.

In the table in the section called “Target Atoms”, the word TEXT (in the Type column) is used to indicate one of the registered encoding names. The type would not actually be TEXT; it would be STRING or some other ATOM naming the encoding chosen by the owner.

STRING as a type or a target specifies the ISO Latin-1 character set plus the control characters TAB (octal 11) and NEWLINE (octal 12). The spacing interpretation of TAB is context dependent. Other ASCII control characters are explicitly not included in STRING at the present time.

COMPOUND_TEXT as a type or a target specifies the Compound Text interchange format; see the Compound Text Encoding.

There are some text objects where the source or intended user, as the case may be, does not have a specific character set for the text, but instead merely requires a zero-terminated sequence of bytes with no other restriction; no element of the selection mechanism may assume that any byte value is forbidden or that any two differing sequences are equivalent. ^[5] For these objects, the type C_STRING should be used.

Type STRING, COMPOUND_TEXT, and C_STRING properties will consist of a list of elements separated by null characters; other encodings will need to specify an appropriate list format.

Requestors may receive a property of type INCR ^[6] in response to any target that results in selection data.

This indicates that the owner will send the actual data incrementally. The contents of the INCR property will be an integer, which represents a lower bound on the number of bytes of data in the selection. The requestor and the selection owner transfer the data in the selection in the following manner.

The selection requestor starts the transfer process by deleting the (type==INCR) property forming the reply to the selection.

The selection owner then:

The selection requestor:

The type of the converted selection is the type of the first partial property. The remaining partial properties must have the same type.

Requestors may receive properties of type PIXMAP, BITMAP, DRAWABLE, or WINDOW, which contain an appropriate ID. While information about these drawables is available from the server by means of the GetGeometry request, the following items are not:

In general, requestors converting into targets whose returned type in the table in the section called “Target Atoms” is one of the DRAWABLE types should expect to convert also into the following targets (using the MULTIPLE mechanism):

Properties with type SPAN contain a list of cardinal-pairs with the length of the cardinals determined by the format. The first specifies the starting position, and the second specifies the ending position plus one. The base is zero. If they are the same, the span is zero-length and is before the specified position. The units are implied by the target atom, such as LINE_NUMBER or CHARACTER_POSITION.

A client's top-level window is a window whose override-redirect attribute is False. It must either be a child of a root window, or it must have been a child of a root window immediately prior to having been reparented by the window manager. If the client reparents the window away from the root, the window is no longer a top-level window; but it can become a top-level window again if the client reparents it back to the root.

A client usually would expect to create its top-level windows as children of one or more of the root windows by using some boilerplate like the following:

win = XCreateSimpleWindow(dpy, DefaultRootWindow(dpy), xsh.x, xsh.y,
     xsh.width, xsh.height, bw, bd, bg);

If a particular one of the root windows was required, however, it could use something like the following:

win = XCreateSimpleWindow(dpy, RootWindow(dpy, screen), xsh.x, xsh.y,
     xsh.width, xsh.height, bw, bd, bg);

Ideally, it should be possible to override the choice of a root window and allow clients (including window managers) to treat a nonroot window as a pseudo-root. This would allow, for example, the testing of window managers and the use of application-specific window managers to control the subwindows owned by the members of a related suite of clients. Doing so properly requires an extension, the design of which is under study.

From the client's point of view, the window manager will regard its top-level window as being in one of three states:

Newly created windows start in the Withdrawn state. Transitions between states happen when the top-level window is mapped and unmapped and when the window manager receives certain messages. For further details, see the section called “WM_HINTS Property”. and the section called “Changing Window State”.

Once the client has one or more top-level windows, it should place properties on those windows to inform the window manager of the behavior that the client desires. Window managers will assume values they find convenient for any of these properties that are not supplied; clients that depend on particular values must explicitly supply them. The window manager will not change properties written by the client.

The window manager will examine the contents of these properties when the window makes the transition from the Withdrawn state and will monitor some properties for changes while the window is in the Iconic or Normal state. When the client changes one of these properties, it must use Replace mode to overwrite the entire property with new data; the window manager will retain no memory of the old value of the property. All fields of the property must be set to suitable values in a single Replace mode ChangeProperty request. This ensures that the full contents of the property will be available to a new window manager if the existing one crashes, if it is shut down and restarted, or if the session needs to be shut down and restarted by the session manager.

Some of these properties may contain the IDs of resources, such as windows or pixmaps. Clients should ensure that these resources exist for at least as long as the window on which the property resides.

If these properties are longer than expected, clients should ignore the remainder of the property. Extending these properties is reserved to the X Consortium; private extensions to them are forbidden. Private additional communication between clients and window managers should take place using separate properties. The only exception to this rule is the WM_PROTOCOLS property, which may be of arbitrary length and which may contain atoms representing private protocols (see the section called “WM_PROTOCOLS Property” ).

The next sections describe each of the properties the clients need to set, in turn. They are summarized in the table in the section called “Summary of Window Manager Property Types”

width = base_width + ( i x width_inc )

height = base_height + ( j x height_inc )

The properties that were described in the previous section are those that the client is responsible for maintaining on its top-level windows. This section describes the properties that the window manager places on client's top-level windows and on the root.

From the client's point of view, the window manager will regard each of the client's top-level windows as being in one of three states, whose semantics are as follows:

In fact, the window manager may implement states with semantics other than those described above. For example, a window manager might implement a concept of an "inactive" state in which an infrequently used client's window would be represented as a string in a menu. But this state is invisible to the client, which would see itself merely as being in the Iconic state.

Newly created top-level windows are in the Withdrawn state. Once the window has been provided with suitable properties, the client is free to change its state as follows:

Only the client can effect a transition into or out of the Withdrawn state. Once a client's window has left the Withdrawn state, the window will be mapped if it is in the Normal state and the window will be unmapped if it is in the Iconic state. Reparenting window managers must unmap the client's window when it is in the Iconic state, even if an ancestor window being unmapped renders the client's window unviewable. Conversely, if a reparenting window manager renders the client's window unviewable by unmapping an ancestor, the client's window is by definition in the Iconic state and must also be unmapped.

When changing the state of the window to Withdrawn, the client must (in addition to unmapping the window) send a synthetic UnmapNotify event by using a SendEvent request with the following arguments:

When a client withdraws a window, the window manager will then update or remove the WM_STATE property as described in the section called “WM_STATE Property”. Clients that want to re-use a client window (e.g., by mapping it again or reparenting it elsewhere) after withdrawing it must wait for the withdrawal to be complete before proceeding. The preferred method for doing this is for clients to wait for the window manager to update or remove the WM_STATE property. ^[8]

If the transition is from the Normal to the Iconic state, the client should send a ClientMessage event to the root with:

Other values of data[0] are reserved for future extensions to these conventions. The parameters of the SendEvent request should be those described for the synthetic UnmapNotify event.

Clients can resize and reposition their top-level windows by using the ConfigureWindow request. The attributes of the window that can be altered with this request are as follows:

The coordinate system in which the location is expressed is that of the root (irrespective of any reparenting that may have occurred). The border width to be used and win_gravity position hint to be used are those most recently requested by the client. Client configure requests are interpreted by the window manager in the same manner as the initial window geometry mapped from the Withdrawn state, as described in the section called “WM_NORMAL_HINTS Property” Clients must be aware that there is no guarantee that the window manager will allocate them the requested size or location and must be prepared to deal with any size and location. If the window manager decides to respond to a ConfigureRequest request by:

The general rule is that coordinates in real ConfigureNotify events are in the parent's space; in synthetic events, they are in the root space.

Clients should be aware that their borders may not be visible. Window managers are free to use reparenting techniques to decorate client's top-level windows with borders containing titles, controls, and other details to maintain a consistent look-and-feel. If they do, they are likely to override the client's attempts to set the border width and set it to zero. Clients, therefore, should not depend on the top-level window's border being visible or use it to display any critical information. Other window managers will allow the top-level windows border to be visible.

Clients that change their position in the stack must be aware that they may have been reparented, which means that windows that used to be siblings no longer are. Using a nonsibling as the sibling parameter on a ConfigureWindow request will cause an error.

Clients that must position themselves in the stack relative to some window that was originally a sibling must do the ConfigureWindow request (in case they are running under a nonreparenting window manager), be prepared to deal with a resulting error, and then follow with a synthetic ConfigureRequest event by invoking a SendEvent request with the following arguments:

Window managers are in any case free to position windows in the stack as they see fit, and so clients should not rely on receiving the stacking order they have requested. Clients should ignore the above-sibling field of both real and synthetic ConfigureNotify events received on their top-level windows because this field may not contain useful information.

The attributes that may be supplied when a window is created may be changed by using the ChangeWindowAttributes request. The window attributes are listed in the following table:

Most attributes are private to the client and will never be interfered with by the window manager. For the attributes that are not private to the client:

There are four models of input handling:

The four input models and the corresponding values of the input field and the presence or absence of the WM_TAKE_FOCUS atom in the WM_PROTOCOLS property are listed in the following table:

Passive and Locally Active clients set the input field of WM_HINTS to True, which indicates that they require window manager assistance in acquiring the input focus. No Input and Globally Active clients set the input field to False, which requests that the window manager not set the input focus to their top-level window.

Clients that use a SetInputFocus request must set the time field to the timestamp of the event that caused them to make the attempt. This cannot be a FocusIn event because they do not have timestamps. Clients may also acquire the focus without a corresponding EnterNotify. Note that clients must not use CurrentTime in the time field.

Clients using the Globally Active model can only use a SetInputFocus request to acquire the input focus when they do not already have it on receipt of one of the following events:

In general, clients should avoid using passive-grabbed key events for this purpose, except when they are unavoidable (as, for example, a selection tool that establishes a passive grab on the keys that cut, copy, or paste).

The method by which the user commands the window manager to set the focus to a window is up to the window manager. For example, clients cannot determine whether they will see the click that transfers the focus.

Windows with the atom WM_TAKE_FOCUS in their WM_PROTOCOLS property may receive a ClientMessage event from the window manager (as described in the section called “ClientMessage Events”. ) with WM_TAKE_FOCUS in its data[0] field and a valid timestamp (i.e., not CurrentTime) in its data[1] field. If they want the focus, they should respond with a SetInputFocus request with its window field set to the window of theirs that last had the input focus or to their default input window, and the time field set to the timestamp in the message. For further information, see the section called “Input Focus”

A client could receive WM_TAKE_FOCUS when opening from an icon or when the user has clicked outside the top-level window in an area that indicates to the window manager that it should assign the focus (for example, clicking in the headline bar can be used to assign the focus).

The goal is to support window managers that want to assign the input focus to a top-level window in such a way that the top-level window either can assign it to one of its subwindows or can decline the offer of the focus. For example, a clock or a text editor with no currently open frames might not want to take focus even though the window manager generally believes that clients should take the input focus after being deiconified or raised.

Clients that set the input focus need to decide a value for the revert-to field of the SetInputFocus request. This determines the behavior of the input focus if the window the focus has been set to becomes not viewable. The value can be any of the following:

Note that neither PointerRoot nor None is really safe to use.

A convention is also required for clients that want to give up the input focus. There is no safe value set for them to set the input focus to; therefore, they should ignore input material.

The window manager is responsible for installing and uninstalling colormaps on behalf of clients with top-level windows that the window manager manages.

Clients provide the window manager with hints as to which colormaps to install and uninstall. Clients must not install or uninstall colormaps themselves (except under the circumstances noted below). When a client's top-level window gets the colormap focus (as a result of whatever colormap focus policy is implemented by the window manager), the window manager will ensure that one or more of the client's colormaps are installed.

Clients whose top-level windows and subwindows all use the same colormap should set its ID in the colormap field of the top-level window's attributes. They should not set a WM_COLORMAP_WINDOWS property on the top-level window. If they want to change the colormap, they should change the top-level window's colormap attribute. The window manager will track changes to the window's colormap attribute and install colormaps as appropriate.

Clients that create windows can use the value CopyFromParent to inherit their parent's colormap. Window managers will ensure that the root window's colormap field contains a colormap that is suitable for clients to inherit. In particular, the colormap will provide distinguishable colors for BlackPixel and WhitePixel.

Top-level windows that have subwindows or override-redirect pop-up windows whose colormap requirements differ from the top-level window should have a WM_COLORMAP_WINDOWS property. This property contains a list of IDs for windows whose colormaps the window manager should attempt to have installed when, in the course of its individual colormap focus policy, it assigns the colormap focus to the top-level window (see the section called “WM_COLORMAP_WINDOWS Property” ). The list is ordered by the importance to the client of having the colormaps installed. The window manager will track changes to this property and will track changes to the colormap attribute of the windows in the property.

If the relative importance of colormaps changes, the client should update the WM_COLORMAP_WINDOWS property to reflect the new ordering. If the top-level window does not appear in the list, the window manager will assume it to be of higher priority than any window in the list.

WM_TRANSIENT_FOR windows can either have their own WM_COLORMAP_WINDOWS property or appear in the property of the window they are transient for, as appropriate.

Whenever possible, clients should use the mechanisms described above and let the window manager handle colormap installation. However, clients are permitted to perform colormap installation on their own while they have the pointer grabbed. A client performing colormap installation must notify the window manager prior to the first installation. When the client has finished its colormap installation, it must also notify the window manager. The client notifies the window manager by issuing a SendEvent request with the following arguments:

This feature was introduced in version 2.0 of this document, and there will be a significant period of time before all window managers can be expected to implement this feature. Before using this feature, clients must check the compliance level of the window manager (using the mechanism described in the section called “Communication with the Window Manager by Means of Selections” ) to verify that it supports this feature. This is necessary to prevent colormap installation conflicts between clients and older window managers.

Window managers should refrain from installing colormaps while a client has requested control of colormap installation. The window manager should continue to track the set of installed colormaps so that it can reinstate its colormap focus policy when the client has finished colormap installation.

This technique has race conditions that may result in the colormaps continuing to be installed even after a client has issued its notification message. For example, the window manager may have issued some InstallColormap requests that are not executed until after the client's SendEvent and InstallColormap requests, thus uninstalling the client's colormaps. If this occurs while the client still has the pointer grabbed and before the client has issued the "finished" message, the client may reinstall the desired colormaps.

A client can hint to the window manager about the desired appearance of its icon by setting:

In the Iconic state, the window manager usually will ensure that:

Clients should observe the following conventions about their icon windows:

Window managers will differ as to whether they support input events to client's icon windows; most will allow the client to receive some subset of the keys and buttons.

Window managers will ignore any WM_NAME, WM_ICON_NAME, WM_NORMAL_HINTS, WM_HINTS, WM_CLASS, WM_TRANSIENT_FOR, WM_PROTOCOLS, WM_COLORMAP_WINDOWS, WM_COMMAND, or WM_CLIENT_MACHINE properties they find on icon windows.

Clients that wish to pop up a window can do one of three things:

Window managers are free to decide if WM_TRANSIENT_FOR windows should be iconified when the window they are transient for is. Clients displaying WM_TRANSIENT_FOR windows that have (or request to have) the window they are transient for iconified do not need to request that the same operation be performed on the WM_TRANSIENT_FOR window; the window manager will change its state if that is the policy it wishes to enforce.

A set of top-level windows that should be treated from the user's point of view as related (even though they may belong to a number of clients) should be linked together using the window_group field of the WM_HINTS structure.

One of the windows (that is, the one the others point to) will be the group leader and will carry the group as opposed to the individual properties. Window managers may treat the group leader differently from other windows in the group. For example, group leaders may have the full set of decorations, and other group members may have a restricted set.

It is not necessary that the client ever map the group leader; it may be a window that exists solely as a placeholder.

It is up to the window manager to determine the policy for treating the windows in a group. At present, there is no way for a client to request a group, as opposed to an individual, operation.

Clients must be aware that some window managers will reparent their top-level windows so that a window that was created as a child of the root will be displayed as a child of some window belonging to the window manager. The effects that this reparenting will have on the client are as follows:

Clients that want to be notified when they are reparented can select for StructureNotify events on their top-level window. They will receive a ReparentNotify event if and when reparenting takes place. When a client withdraws a top-level window, the window manager will reparent it back to the root window if the window had been reparented elsewhere.

If the window manager reparents a client's window, the reparented window will be placed in the save-set of the parent window. This means that the reparented window will not be destroyed if the window manager terminates and will be remapped if it was unmapped. Note that this applies to all client windows the window manager reparents, including transient windows and client icon windows.

Clients must be aware that some window managers will arrange for some client requests to be intercepted and redirected. Redirected requests are not executed; they result instead in events being sent to the window manager, which may decide to do nothing, to alter the arguments, or to perform the request on behalf of the client.

The possibility that a request may be redirected means that a client cannot assume that any redirectable request is actually performed when the request is issued or is actually performed at all. The requests that may be redirected are MapWindow, ConfigureWindow, and CirculateWindow.

A window with the override-redirect bit set is immune from redirection, but the bit should be set on top-level windows only in cases where other windows should be prevented from processing input while the override-redirect window is mapped (see the section called “Pop-up Windows” ) and while responding to ResizeRequest events (see the section called “Redirecting Requests” ).

Clients that have no non-Withdrawn top-level windows and that map an override-redirect top-level window are taking over total responsibility for the state of the system. It is their responsibility to:

In effect, clients of this kind are acting as temporary window managers. Doing so is strongly discouraged because these clients will be unaware of the user interface policies the window manager is trying to maintain and because their user interface behavior is likely to conflict with that of less demanding clients.

If the window manager moves a top-level window without changing its size, the client will receive a synthetic ConfigureNotify event following the move that describes the new location in terms of the root coordinate space. Clients must not respond to being moved by attempting to move themselves to a better location.

Any real ConfigureNotify event on a top-level window implies that the window's position on the root may have changed, even though the event reports that the window's position in its parent is unchanged because the window may have been reparented. Note that the coordinates in the event will not, in this case, be directly useful.

The window manager will send these events by using a SendEvent request with the following arguments:

The client can elect to receive notification of being resized by selecting for StructureNotify events on its top-level windows. It will receive a ConfigureNotify event. The size information in the event will be correct, but the location will be in the parent window (which may not be the root).

The response of the client to being resized should be to accept the size it has been given and to do its best with it. Clients must not respond to being resized by attempting to resize themselves to a better size. If the size is impossible to work with, clients are free to request to change to the Iconic state.

A top-level window that is not Withdrawn will be in the Normal state if it is mapped and in the Iconic state if it is unmapped. This will be true even if the window has been reparented; the window manager will unmap the window as well as its parent when switching to the Iconic state.

The client can elect to be notified of these state changes by selecting for StructureNotify events on the top-level window. It will receive a UnmapNotify event when it goes Iconic and a MapNotify event when it goes Normal.

Clients that wish to be notified of their colormaps being installed or uninstalled should select for ColormapNotify events on their top-level windows and on any windows they have named in WM_COLORMAP_WINDOWS properties on their top-level windows. They will receive ColormapNotify events with the new field FALSE when the colormap for that window is installed or uninstalled.

Clients can request notification that they have the input focus by selecting for FocusChange events on their top-level windows; they will receive FocusIn and FocusOut events. Clients that need to set the input focus to one of their subwindows should not do so unless they have set WM_TAKE_FOCUS in their WM_PROTOCOLS property and have done one of the following:

Clients should not warp the pointer in an attempt to transfer the focus; they should set the focus and leave the pointer alone. For further information, see the section called “The Pointer”.

Once a client satisfies these conditions, it may transfer the focus to another of its windows by using the SetInputFocus request, which is defined as follows:

SetInputFocus

There is no way for clients to prevent themselves being sent ClientMessage events.

Top-level windows with a WM_PROTOCOLS property may be sent ClientMessage events specific to the protocols named by the atoms in the property (see the section called “WM_PROTOCOLS Property” ). For all protocols, the ClientMessage events have the following:

The remaining fields of the event, including the window field, are determined by the protocol.

These events will be sent by using a SendEvent request with the following arguments:

Normal clients can use the redirection mechanism just as window managers do by selecting for SubstructureRedirect events on a parent window or ResizeRedirect events on a window itself. However, at most, one client per window can select for these events, and a convention is needed to avoid clashes.

In particular, clients that need to take some special action if they are resized can select for ResizeRedirect events on their top-level windows. They will receive a ResizeRequest event if the window manager resizes their window, and the resize will not actually take place. Clients are free to make what use they like of the information that the window manager wants to change their size, but they must configure the window to the width and height specified in the event in a timely fashion. To ensure that the resize will actually happen at this stage instead of being intercepted and executed by the window manager (and thus restarting the process), the client needs temporarily to set override-redirect on the window.

If a window manager detects that a client is not obeying this convention, it is free to take whatever measures it deems appropriate to deal with the client.