This section describes how to configure an OpenGL context. For most applications the information described here is far too low-level to be of any concern, however more advanced applications can take advantage of the complete control pyglet provides.
When you draw on a window in pyglet, you are drawing to an OpenGL context. Every window has its own context, which is created when the window is created. You can access the window’s context via its context attribute.
The context is created from an OpenGL configuration (or “config”), which describes various properties of the context such as what color format to use, how many buffers are available, and so on. You can access the config that was used to create a context via the context’s config attribute.
For example, here we create a window using the default config and examine some of its properties:
>>> import pyglet >>> window = pyglet.window.Window() >>> context = window.context >>> config = context.config >>> config.double_buffer c_int(1) >>> config.stereo c_int(0) >>> config.sample_buffers c_int(0)
Note that the values of the config’s attributes are all ctypes instances. This is because the config was not specified by pyglet. Rather, it has been selected by pyglet from a list of configs supported by the system. You can make no guarantee that a given config is valid on a system unless it was provided to you by the system.
pyglet simplifies the process of selecting one of the system’s configs by allowing you to create a “template” config which specifies only the values you are interested in. See Simple context configuration for details.
The system may actually support several different sets of configs, depending on which display device is being used. For example, a computer with two video cards would have not support the same configs on each card. Another example is using X11 remotely: the display device will support different configurations than the local driver. Even a single video card on the local computer may support different configs for the two monitors plugged in.
In pyglet, a “display” is a collection of “screens” attached to a single display device. On Linux, the display device corresponds to the X11 display being used. On Windows and Mac OS X, there is only one display (as these operating systems present multiple video cards as a single virtual device).
There is a singleton class Platform which provides access to the display(s); this represents the computer on which your application is running. It is usually sufficient to use the default display:
>>> platform = pyglet.window.get_platform() >>> display = platform.get_default_display()
On X11, you can specify the display string to use, for example to use a remotely connected display. The display string is in the same format as used by the DISPLAY environment variable:
>>> display = platform.get_display('remote:1.0')
You use the same string to specify a separate X11 screen :
>>> display = platform.get_display(':0.1')
Once you have obtained a display, you can enumerate the screens that are connected. A screen is the physical display medium connected to the display device; for example a computer monitor, TV or projector. Most computers will have a single screen, however dual-head workstations and laptops connected to a projector are common cases where more than one screen will be present.
In the following example the screens of a dual-head workstation are listed:
>>> for screen in display.get_screens(): ... print screen ... XlibScreen(screen=0, x=1280, y=0, width=1280, height=1024, xinerama=1) XlibScreen(screen=0, x=0, y=0, width=1280, height=1024, xinerama=1)
Because this workstation is running Linux, the returned screens are XlibScreen, a subclass of Screen. The screen and xinerama attributes are specific to Linux, but the x, y, width and height attributes are present on all screens, and describe the screen’s geometry, as shown below.
There is always a “default” screen, which is the first screen returned by get_screens. Depending on the operating system, the default screen is usually the one that contains the taskbar (on Windows) or menu bar (on OS X). You can access this screen directly using get_default_screen.
|||Assuming Xinerama is not being used to combine the screens. If Xinerama is enabled, use screen 0 in the display string, and select a screen in the same manner as for Windows and Mac OS X.|
When configuring or selecting a Config, you do so based on the properties of that config. pyglet supports a fixed subset of the options provided by AGL, GLX, WGL and their extensions. In particular, these constraints are placed on all OpenGL configs:
The visible portion of the buffer, sometimes called the color buffer, is configured with the following attributes:
Number of bits per sample. Common values are 24 and 32, which each dedicate 8 bits per color component. A buffer size of 16 is also possible, which usually corresponds to 5, 6, and 5 bits of red, green and blue, respectively.
Usually there is no need to set this property, as the device driver will select a buffer size compatible with the current display mode by default.
- red_size, blue_size, green_size, alpha_size
These each give the number of bits dedicated to their respective color component. You should avoid setting any of the red, green or blue sizes, as these are determined by the driver based on the buffer_size property.
If you require an alpha channel in your color buffer (for example, if you are compositing in multiple passes) you should specify alpha_size=8 to ensure that this channel is created.
- sample_buffers and samples
Configures the buffer for multisampling, in which more than one color sample is used to determine the color of each pixel, leading to a higher quality, antialiased image.
Enable multisampling by setting sample_buffers=1, then give the number of samples per pixel to use in samples. For example, samples=2 is the fastest, lowest-quality multisample configuration. A higher-quality buffer (with a compromise in performance) is possible with samples=4.
Not all video hardware supports multisampling; you may need to make this a user-selectable option, or be prepared to automatically downgrade the configuration if the requested one is not available.
- Creates separate left and right buffers, for use with stereo hardware. Only specialised video hardware such as stereoscopic glasses will support this option. When used, you will need to manually render to each buffer, for example using glDrawBuffers.
Create separate front and back buffers. Without double-buffering, drawing commands are immediately visible on the screen, and the user will notice a visible flicker as the image is redrawn in front of them.
It is recommended to set double_buffer=True, which creates a separate hidden buffer to which drawing is performed. When the Window.flip is called, the buffers are swapped, making the new drawing visible virtually instantaneously.
In addition to the color buffer, several other buffers can optionally be created based on the values of these properties:
- A depth buffer is usually required for 3D rendering. The typical depth size is 24 bits. Specify 0 if you do not require a depth buffer.
- The stencil buffer is required for masking the other buffers and implementing certain volumetric shadowing algorithms. The typical stencil size is 8 bits; or specify 0 if you do not require it.
- accum_red_size, accum_blue_size, accum_green_size, accum_alpha_size
The accumulation buffer can be used for simple antialiasing, depth-of-field, motion blur and other compositing operations. Its use nowadays is being superceded by the use of floating-point textures, however it is still a practical solution for implementing these effects on older hardware.
If you require an accumulation buffer, specify 8 for each of these attributes (the alpha component is optional, of course).
Each auxilliary buffer is configured the same as the colour buffer. Up to four auxilliary buffers can typically be created. Specify 0 if you do not require any auxilliary buffers.
Like the accumulation buffer, auxilliary buffers are used less often nowadays as more efficient techniques such as render-to-texture are available. They are almost universally available on older hardware, though, where the newer techniques are not possible.
A context can only be created from a config that was provided by the system. Enumerating and comparing the attributes of all the possible configs is a complicated process, so pyglet provides a simpler interface based on “template” configs.
To get the config with the attributes you need, construct a Config and set only the attributes you are interested in. You can then supply this config to the Window constructor to create the context.
For example, to create a window with an alpha channel:
config = pyglet.gl.Config(alpha_size=8) window = pyglet.window.Window(config=config)
It is sometimes necessary to create the context yourself, rather than letting the Window constructor do this for you. In this case use Screen.get_best_config to obtain a “complete” config, which you can then use to create the context:
platform = pyglet.window.get_platform() display = platform.get_default_display() screen = display.get_default_screen() template = pyglet.gl.Config(alpha_size=8) config = screen.get_best_config(template) context = config.create_context(None) window = pyglet.window.Window(context=context)
Note that you cannot create a context directly from a template (any Config you constructed yourself). The Window constructor performs a similar process to the above to create the context if a template config is given.
Not all configs will be possible on all machines. The call to get_best_config will raise NoSuchConfigException if the hardware does not support the requested attributes. It will never return a config that does not meet or exceed the attributes you specify in the template.
You can use this to support newer hardware features where available, but also accept a lesser config if necessary. For example, the following code creates a window with multisampling if possible, otherwise leaves multisampling off:
template = gl.Config(sample_buffers=1, samples=4) try: config = screen.get_best_config(template) except pyglet.window.NoSuchConfigException: template = gl.Config() config = screen.get_best_config(template) window = pyglet.window.Window(config=config)
Allowing pyglet to select the best configuration based on a template is sufficient for most applications, however some complex programs may want to specify their own algorithm for selecting a set of OpenGL attributes.
You can enumerate a screen’s configs using the get_matching_configs method. You must supply a template as a minimum specification, but you can supply an “empty” template (one with no attributes set) to get a list of all configurations supported by the screen.
In the following example, all configurations with either an auxilliary buffer or an accumulation buffer are printed:
platform = pyglet.window.get_platform() display = platform.get_default_display() screen = display.get_default_screen() for config in screen.get_matching_configs(gl.Config()): if config.aux_buffers or config.accum_red_size: print config
As well as supporting more complex configuration selection algorithms, enumeration allows you to efficiently find the maximum value of an attribute (for example, the maximum samples per pixel), or present a list of possible configurations to the user.