For the past few weeks I've been learning a lot about the communication interface between the X11 server and its clients. As a devotee of Qtile, I have been using it as a testing ground for a number of tools that directly talk with the X server, using the Python X bindings provided by xcffib.
This exercise has seen me spend a fair bit of time and effort trying to 'port' C code using XCB or xlib to xcffib equivalents. At times this is tedious - as anybody who has written X client code will know - but I like the idea of extending Qtile to include more features typically outside of the scope of a standalone window-manager.
One such feature is the simple ability to set the background wallpaper. Sounds easy, right? The X server manages the drawing of windows as regions within the root window, which itself is handled in many ways like a normal window. The appearance of each window (including the root window) is stored within a pixmap, whose data can be rendered on-screen as pixels. This description is super simplified; for a fantastic and more in-depth exploration of how this works I recommend reading xplain by Jasper St. Pierre.
Setting the desktop wallpaper therefore means colouring and rendering the pixmap for the root window of each screen. Wrapping this behaviour as an X client: we need to open a connection to the X server, load our wallpaper image in a form that can be painted onto a pixmap, and then perform the painting. I've packaged what I've described here with a convenient interface on my github.
Opening an X client connection is straightforward and requires the
DISPLAY environmental variable. xcffib gives us a
Connection object from which we can get information from the X server such as screen setup.
import os import xcffib import cairocffi # needed later import cairocffi.pixbuf # needed later import xcffib.xproto # needed later conn = xcffib.Connection(display=os.environ.get("DISPLAY")) screens = conn.get_setup().roots
Next we need to load our image. The cairocffi library provides Python bindings for cairo, a 2D graphics library that supports rendering graphics to X pixmaps. We can load our image into what cairo calls a
with open('/path/to/image.png', 'rb') as fd: image, _ = cairocffi.pixbuf.decode_to_image_surface(fd.read())
As we need to paint the image to one screen at a time, we could load multiple images and use a different image for each painting operation.
The next part performs the painting of the root pixmap to set the wallpaper. First we must ask the server for a new resource ID and use this to create our pixmap for the current screen. Creating the pixmap requires the colour depth of the screen (
screen.root_depth) as well as its own ID (
screen.root), and its dimensions:
screen = screens pixmap = conn.generate_id() conn.core.CreatePixmap( screen.root_depth, pixmap, screen.root, screen.width_in_pixels, screen.height_in_pixels, )
We could iterate over
screens to paint each screen.
Next we need to extract from the screen its
visual, which contains information about how it manages colour maps and depths. We need this to create a cairo surface that is compatible with the root window, onto which we can paint our image:
for depth in screen.allowed_depths: for visual in depth.visuals: if visual.visual_id == screen.root_visual: root_visual = visual break surface = cairocffi.xcb.XCBSurface( conn, pixmap, root_visual, screen.width_in_pixels, screen.height_in_pixels, )
The cairocffi API for manipulating surfaces provides us a
Context in which to modify and use our
Surface objects (the image and pixmap surfaces). Surface manipulation is pretty nice, and it only takes one command to set our image as a data source and another to paint it to the pixmap:
with cairocffi.Context(surface) as context: context.set_source_surface(image) context.paint()
It is at this point where we could add more image manipulations to the source image before painting, such as stretching or tiling.
Root windows have two properties named
ESETROOT_PMAP_ID which it uses to publish the root pixmap so that other X clients can have access to the pixel data. This is used for effects such as the pseudo-transparency feature of urxvt. We therefore need to set these properties using our newly painted pixmap.
The xcffib API for this might look a bit cryptic; we are passing the property-setting mode
Replace, the root window concerned, the property we want to change, the type of data we are passing (
PIXMAP) and lastly the bit format, number of items and our list of items (just our pixmap):
conn.core.ChangeProperty( xcffib.xproto.PropMode.Replace, screen.root, conn.core.InternAtom(False, 13, '_XROOTPMAP_ID').reply().atom, xcffib.xproto.Atom.PIXMAP, 32, 1, [pixmap] ) conn.core.ChangeProperty( xcffib.xproto.PropMode.Replace, screen.root, conn.core.InternAtom(False, 16, 'ESETROOT_PMAP_ID').reply().atom, xcffib.xproto.Atom.PIXMAP, 32, 1, [pixmap] )
We can then change the root window's background pixmap to our pixmap and clear the area that contains it, which refreshes those pixels to display their new values:
conn.core.ChangeWindowAttributes( screen.root, xcffib.xproto.CW.BackPixmap, [pixmap] ) conn.core.ClearArea( 0, screen.root, 0, 0, # x and y position screen.width_in_pixels, screen.height_in_pixels )
ClearArea call background pixels will only refresh when you move a window over them, which can be a cool effect.
Lastly we should set our X client's
CloseDown mode to
RetainPermanent to make the our changes to the root window persist after the client closes, and then disconnect.
The logic we've looked at so far is sufficient to set the X wallpaper, and can easily be extended to apply wallpapers to multiple screens and to manipulate our desired image before painting it the pixmap.
For example, if the dimensions of our image and screen might differ or if we want to use only a subregion of an image, we can use cairocffi's
Context API to change how we paint to our pixmap. The library exposes
Context.translate() methods which can be used right before the paint command to change how the image will map onto the pixmap.
If we want to paint the wallpaper with a single colour instead of an image, we can replace the call to
context.set_source_surface() with the following, where the three arguments correspond to red, green and blue values:
context.set_source_rgb(1, 1, 1)