XMonad.Doc.Extending

This module documents the xmonad-contrib library and guides you through some more advanced parts of extending the capabilities of xmonad. If you're new to xmonad, you should first check out the tutorial and treat this document as supplemental reading.

Knowing Haskell is by no means a prerequisite for configuring xmonad and the tutorial emphasizes this. This document, however, does assume a basic familiarity with the language. This is so that we can dive a bit deeper into what the different hooks do, or how to write our own little functions to configure xmonad.

Those wishing to be totally hardcore and develop their own xmonad extensions (it's easier than it sounds, we promise!) should read the documentation in XMonad.Doc.Developing.

More configuration examples can be found here.

The xmonad-contrib library

The xmonad-contrib library is a set of extension modules contributed by xmonad hackers and users that provide additional features to xmonad. Examples include various layout modes (tabbed, spiral, three-column...), prompts, program launchers, the ability to manipulate windows and workspaces in various ways, alternate navigation modes, and much more. There are also "meta-modules" which make it easier to write new modules and extensions.

This is a description of the different namespaces in xmonad-contrib. For more information about any particular module, go to the root of the documentation and just click on its name to view its Haddock documentation; each module should come with extensive documentation. If you find a module that could be better documented, or has incorrect documentation, please report it as a bug!

First and foremost, xmonad defines its own prelude for commonly used functions, as well as re-exports from base.

• XMonad.Prelude: Utility functions and re-exports for a more ergonomic developing experience.

There are also other documentation modules, showing you around individual parts of xmonad:

• XMonad.Doc.Configuring: Brief tutorial that will teach you how to create a basic xmonad configuration.
• XMonad.Doc.Developing: A brief overview of xmonad's internals.

A list of the contrib modules can be found at https://xmonad.github.io/xmonad-docs/xmonad-contrib-0.18.0/

Actions

In the XMonad.Actions namespace you can find modules exporting various functions that are usually intended to be bound to key combinations or mouse actions, in order to provide functionality beyond the standard keybindings offered by xmonad.

Hooks

In the XMonad.Hooks namespace you can find modules exporting hooks—actions that xmonad performs when certain events occur. The three most important hooks are:

• manageHook: this hook is called when a new window that xmonad must take care of is created. This is a very powerful hook, since it lets us examine the new window's properties and act accordingly. For instance, we can configure xmonad to put windows belonging to a given application in the float layer, not to manage dock applications, or open them in a given workspace. See Editing the manage hook for more information on customizing manageHook.
• logHook: this hook is called when the stack of windows managed by xmonad changes; for example, this is invoked at the end of the windows function. A big application for this is to display some information about xmonad in a status bar. The aptly named XMonad.Hooks.StatusBar will produce a string (whose format can be configured) to be written, for example, to an X11 property.
• handleEventHook: this hook is called on all events handled by xmonad, thus it is extremely powerful. See Graphics.X11.Xlib.Extras and xmonad source and development documentation for more details.

Layouts

In the XMonad.Layout namespace you can find modules exporting contributed layout algorithms, such as a tabbed layout, a circle, a spiral, three columns, and so on.

You will also find modules which provide facilities for combining different layouts, such as XMonad.Layout.Combo, XMonad.Layout.ComboP, XMonad.Layout.LayoutBuilder, XMonad.Layout.SubLayouts, or XMonad.Layout.LayoutCombinators.

Layouts can be also modified with layout modifiers. A general interface for writing layout modifiers is implemented in XMonad.Layout.LayoutModifier.

For more information on using those modules for customizing your layoutHook see Editing the layout hook.

Prompts

In the XMonad.Prompt namespace you can find modules providing graphical prompts for getting user input and using it to perform various actions.

The XMonad.Prompt module provides a library for easily writing new prompts.

Utilities

In the XMonad.Util namespace you can find modules exporting various utility functions that are used by the other modules of the xmonad-contrib library.

There are also utilities for helping in configuring xmonad or using external utilities.

Extending xmonad

Since the xmonad.hs file is just another Haskell program, you may import and use any Haskell code or libraries you wish, such as extensions from the xmonad-contrib library, or other code you write yourself.

Adding key bindings

In the customization section of the tutorial we have seen how to add new keys to xmonad with the help of the additionalKeysP function. But how does that work? Assuming that library didn't exist yet, could we write it ourselves?

Let's concentrate on the easier case of trying to write our own additionalKeys. This works exactly like its almost-namesake, but requires you to specify the keys in the "default" style—that is:

main :: IO ()
main = xmonad $ def
  `additionalKeys`
    [ ((mod1Mask, xK_m        ), spawn "echo 'Hi, mom!' | dzen2 -p 4")
    , ((mod1Mask, xK_BackSpace), spawn "xterm")
    ]

The extra work that additionalKeysP does is only in parsing the input string (turning "M1-m" into (mod1Mask, xK_m)). As we have seen in the tutorial, is also allows one to write M and have xmonad pick up on the correct modifier key to use—something which additionalKeys can't do.

Editing key bindings means changing the keys field of the XConfig record used by xmonad. For example, to override all of the default bindings with our own, we would write

import XMonad
import Data.Map (Map)
import qualified Data.Map as Map

main :: IO ()
main = xmonad $ def { keys = myKeys }
 where
  myKeys :: XConfig l -> Map (ButtonMask, KeySym) (X ())
  myKeys conf = Map.fromList
    [ ((mod1Mask    , xK_m        ), spawn "echo 'Hi, mom!' | dzen2 -p 4")
    , ((modMask conf, xK_BackSpace), spawn "xterm")
    ]

Now, obviously we don't want to do that; we only want to add to existing bindings (or, perhaps, override some of them with our own). Let's break myKeys down a little. You can think of the type signature of myKeys (and hence also of keys) like this:

   myKeys :: UserConfig -> Map KeyPress Action

It takes some user config and, from that, produces a map that associates certain keypresses with actions to execute. The reason why it might take the user config may seem a bit mysterious at first, but it is for the simple reason that some keybindings (like the workspace switching ones) need access to the user config. We have already seen this above when we queried modMask conf. If it helps, think of this as a Reader monad with the config being the read-only state.

This means that, as a first guess, the type signature of our version of additionalKeys might look like

myAdditionalKeys :: XConfig l
                    -- ^ Base config with xmonad's default keybindings
                 -> (XConfig l -> Map (ButtonMask, KeySym) (X ()))
                    -- ^ User supplied keybindings
                 -> XConfig l
                    -- ^ Resulting config with everything merged together

However, even assuming a correct implementation, using this is not very ergonomic:

main = xmonad $ def
 `myAdditionalKeys`
   (\conf -> Map.fromList
     [ ((mod1Mask    , xK_m        ), spawn "echo 'Hi, mom!' | dzen2 -p 4")
     , ((modMask conf, xK_BackSpace), spawn "xterm")
     ])

Having to specify a lambda with parentheses and call fromList does not make for a good user experience. Since one always has to call that function anyways, we may well just accept a list from the user and transform it to a map ourselves. As an additional simplification, how about we don't care about the config argument at all and simply ask the user for a list? The resulting signature

myAdditionalKeys :: XConfig l
                 -> [(ButtonMask, KeySym), (X ())]
                 -> XConfig l

looks exactly like what we want! Note that this is also the time we lose the ability to automagically fill in the correct modifier key, since the input to myAdditionalKeys is already structured data (as opposed to just some strings that need to be parsed).

Now that we know what kind of data structure—that is, maps—we are dealing with, the implementation of this function simply merges the two together, preferring the user config to xmonad's defaults in case of any conflicts. Thankfully, someone else has already done the hard work and written the merging function for us; it's called union.

What's left is essentially playing "type tetris":

myAdditionalKeys baseConf keyList =
  let mergeKeylist conf = Map.fromList keyList `Map.union` (keys baseConf) conf
   in baseConf { keys = mergeKeylist }

The function mergeKeyList take some user config, transforms the custom keybindings into a map (Map.fromList keyList), gets the keys from the base config (remember keys baseConf is again a function, morally of type UserConfig -> Map KeyPress Action, and so we have to apply conf to it in order to get a map!), and then merges these two maps together. Since mergeKeylist now has exactly the right type signature, we can just set that as the keys.

If you like operators, <> (or xmonad's alias for it, <+>) does exactly the same as the explicit usage of union because that's the specified binary operation in the Monoid instance for Map. Note that the function works as expected (preferring user defined keys) because union is left biased, which means that if the same key is present in both maps it will prefer the associated value of the left map.

Our function now works as expected:

main :: IO ()
main = xmonad $ def
  `myAdditionalKeys`
    [ ((mod1Mask, xK_m        ), spawn "echo 'Hi, mom!' | dzen2 -p 4")
    , ((mod1Mask, xK_BackSpace), spawn "xterm")
    ]

Lastly, if you want you can also emulate the automatic modifier detection by additionalKeysP by defining the bulk of your config as a separate function

myConfig = def { modMask = mod4Mask }

and then using that information

main :: IO ()
main = xmonad $ myConfig
  `myAdditionalKeys`
    [ ((mod, xK_m        ), spawn "echo 'Hi, mom!' | dzen2 -p 4")
    , ((mod, xK_BackSpace), spawn "xterm")
    ]
 where mod = modMask myConfig

Hopefully you now feel well equipped to write some small functions that extend xmonad an scratch a particular itch!

Removing key bindings

As we've learned, XMonad stores keybindings inside of a Map, which means that removing keybindings requires modifying it. This can be done with difference or with delete.

For example, suppose you want to entirely rid yourself of "M-q" and "M-s-q" (you just want to leave xmonad running forever). To do this with bare xmonad, you need to define newKeys as a difference between the default map and the map of the key bindings you want to remove. Like so:

newKeys :: XConfig l -> Map (KeyMask, KeySym) (X ())
newKeys x = keys def x `M.difference` keysToRemove x

keysToRemove :: XConfig l -> Map (KeyMask, KeySym) (X ())
keysToRemove x = M.fromList
         [ ((modm              , xK_q ), return ())
         , ((modm .|. shiftMask, xK_q ), return ())
         ]

As you can see, it doesn't matter what actions we associate with the keys listed in keysToRemove, so we just use return () (the "null" action). Since newKeys contains all of the default keys, you can simply pass it to XConfig as your map of keybindings:

main :: IO ()
main = xmonad $ def { keys = newKeys }

However, having to manually type return () every time seems like a drag, doesn't it? And this approach isn't at all compatible with adding custom keybindings via additionalKeysP! Well, good thing XMonad.Util.EZConfig also sports removeKeysP. You can use it as you would expect.

main :: IO ()
main = xmonad $ def
  { … }
 `removeKeysP` ["M-q", "M-S-q"]

Can you guess how removeKeysP works? It's almost the same code we wrote above, just accepting a list of keybindings. Try to see if you can come up with an implementation of

removeKeysP :: XConfig l -> [String] -> XConfig l

If you're done, just click on # Source when viewing the removeKeysP documentation (did you know that Haddock lets you do that for every function?) and compare.

By the way, one can conveniently combine additionalKeysP and removeKeysP by just intuitively chaining them:

main :: IO ()
main = xmonad $ def
  { … }
 `additionalKeysP myKeys
 `removeKeysP`    ["M-q", "M-S-q"]

If you don't use the P alternatives of EZConfig, there is also an aptly named removeKeys. Again, can you try to come up with an implementation yourself that has the correct signature?

removeKeys :: XConfig a -> [(KeyMask, KeySym)] -> XConfig a

In addition to delete, you will probably need to use foldr.

Editing mouse bindings

Most of the previous discussion of key bindings applies to mouse bindings as well. For example, you could configure button4 to close the window you click on like so:

   import qualified Data.Map as M

   myMouse x  = [ (0, button4), (\w -> focus w >> kill) ]

   newMouse x = M.union (mouseBindings def x) (M.fromList (myMouse x))

   main = xmonad $ def { ..., mouseBindings = newMouse, ... }

Overriding or deleting mouse bindings works similarly. You can also configure mouse bindings much more easily using the additionalMouseBindings and removeMouseBindings functions from the XMonad.Util.EZConfig module.

Editing the layout hook

When you start an application that opens a new window, when you change the focused window, or move it to another workspace, or change that workspace's layout, xmonad will use the layoutHook for reordering the visible windows on the visible workspace(s).

Since different layouts may be attached to different workspaces, and you can change them, xmonad needs to know which one to use. In this sense the layoutHook may be thought as the list of layouts that xmonad will use for laying out windows on the screen(s).

The problem is that the layout subsystem is implemented with an advanced feature of the Haskell programming language: type classes. This allows us to very easily write new layouts, combine or modify existing layouts, create layouts with internal state, etc. This means that we cannot simply have a list of layouts: a list requires every member to belong to the same type!

Instead the combination of layouts to be used by xmonad is created with a specific layout combinator: |||.

Suppose we want a list with the Full, tabbed and Accordion layouts. First we import, in our xmonad.hs, all the needed modules:

   import XMonad

   import XMonad.Layout.Tabbed
   import XMonad.Layout.Accordion

Then we create the combination of layouts we need:

   mylayoutHook = Full ||| tabbed shrinkText def ||| Accordion

Now, all we need to do is change the layoutHook field of the XConfig record, like so:

   main = xmonad $ def { layoutHook = mylayoutHook }

Thanks to the new combinator, we can apply a layout modifier to a whole combination of layouts, instead of applying it to each one. For example, suppose we want to use the noBorders layout modifier, from the XMonad.Layout.NoBorders module (which must be imported):

   mylayoutHook = noBorders (Full ||| tabbed shrinkText def ||| Accordion)

If we want only the tabbed layout without borders, then we may write:

   mylayoutHook = Full ||| noBorders (tabbed shrinkText def) ||| Accordion

Our xmonad.hs will now look like this:

   import XMonad

   import XMonad.Layout.Tabbed
   import XMonad.Layout.Accordion
   import XMonad.Layout.NoBorders

   mylayoutHook = Full ||| noBorders (tabbed shrinkText def) ||| Accordion

   main = xmonad $ def { layoutHook = mylayoutHook }

That's it!

Editing the manage hook

The manageHook is a very powerful tool for customizing the behavior of xmonad with regard to new windows. Whenever a new window is created, xmonad calls the manageHook, which can thus be used to perform certain actions on the new window, such as placing it in a specific workspace, ignoring it, or placing it in the float layer.

The default manageHook causes xmonad to float MPlayer and Gimp, and to ignore gnome-panel, desktop_window, kicker, and kdesktop.

The XMonad.ManageHook module provides some simple combinators that can be used to alter the manageHook by replacing or adding to the default actions.

Let's start by analyzing the default manageHook, defined in XMonad.Config:

   manageHook :: ManageHook
   manageHook = composeAll
                   [ className =? "MPlayer"        --> doFloat
                   , className =? "Gimp"           --> doFloat
                   , resource  =? "desktop_window" --> doIgnore
                   , resource  =? "kdesktop"       --> doIgnore ]

composeAll can be used to compose a list of different ManageHooks. In this example we have a list of ManageHooks formed by the following commands: the Mplayer's and the Gimp's windows, whose className are, respectively "Mplayer" and "Gimp", are to be placed in the float layer with the doFloat function; the windows whose resource names are respectively "desktop_window" and kdesktop" are to be ignored with the doIgnore function.

This is another example of manageHook, taken from XMonad.Config.Arossato:

   myManageHook  = composeAll [ resource =? "realplay.bin" --> doFloat
                              , resource =? "win"          --> doF (W.shift "doc") -- xpdf
                              , resource =? "firefox-bin"  --> doF (W.shift "web")
                              ]
   newManageHook = myManageHook <> manageHook def

Again we use composeAll to compose a list of different ManageHooks. The first one will put RealPlayer on the float layer, the second one will put the xpdf windows in the workspace named "doc", with doF and shift functions, and the third one will put all firefox windows on the workspace called "web". Then we use the <> combinator to compose myManageHook with the default manageHook to form newManageHook.

Each ManageHook has the form:

   property =? match --> action

Where property can be:

• title: the window's title
• resource: the resource name
• className: the resource class name.
• stringProperty somestring: the contents of the property somestring.

(You can retrieve the needed information using the X utility named xprop; for example, to find the resource class name, you can type

xprop | grep WM_CLASS

at a prompt, then click on the window whose resource class you want to know.)

match is the string that will match the property value (for instance the one you retrieved with xprop).

An action can be:

• doFloat: to place the window in the float layer;
• doIgnore: to ignore the window;
• doF: to execute a function with the window as argument.

For example, suppose we want to add a manageHook to float RealPlayer, which usually has a resource name of "realplay.bin".

First we need to import XMonad.ManageHook:

   import XMonad.ManageHook

Then we create our own manageHook:

   myManageHook = resource =? "realplay.bin" --> doFloat

We can now use the <> combinator to add our manageHook to the default one:

   newManageHook = myManageHook <> manageHook def

(Of course, if we wanted to completely replace the default manageHook, this step would not be necessary.) Now, all we need to do is change the manageHook field of the XConfig record, like so:

   main = xmonad def { ..., manageHook = newManageHook, ... }

And we are done.

Obviously, we may wish to add more then one manageHook. In this case we can use a list of hooks, compose them all with composeAll, and add the composed to the default one.

For instance, if we want RealPlayer to float and thunderbird always opened in the workspace named "mail", we can do so like this:

   myManageHook = composeAll [ resource =? "realplay.bin"    --> doFloat
                             , resource =? "thunderbird-bin" --> doF (W.shift "mail")
                             ]

Remember to import the module that defines the shift function, XMonad.StackSet, like this:

   import qualified XMonad.StackSet as W

And then we can add myManageHook to the default one to create newManageHook as we did in the previous example.

One more thing to note about this system is that if a window matches multiple rules in a manageHook, all of the corresponding actions will be run (in the order in which they are defined). An alternative version where only the first rule that matches is run is available as composeOne.

For additional rules and actions you can use in your manageHook, check out the contrib module XMonad.Hooks.ManageHelpers.