Double-chance function

In software engineering, a double-chance function is a software design pattern with a strong application in cross-platform and scalable development.

Computer graphics
Consider a graphics API with functions to,  , and. It is easy to see that  can be implemented solely in terms of , and   can in turn be implemented through four calls to. If you were porting this API to a new architecture you would have a choice: implement three different functions natively (taking more time to implement, but likely resulting in faster code), or write  natively, and implement the others as described above using common, cross-platform, code. An important example of this approach is the X11 graphics system, which can be ported to new graphics hardware by providing a very small number of device-dependent primitives, leaving higher level functions to a hardware-independent layer.

The double-chance function is an optimal method of creating such an implementation, whereby the first draft of the port can use the "fast to market, slow to run" version with a common  function, while later versions can be modified as "slow to market, fast to run". Where the double-chance pattern scores high is that the base API includes the self-supporting implementation given here as part of the null driver, and all other implementations are extensions of this. Consequently, the first port is, in fact, the first usable implementation.

One typical implementation in C++ could be: Note that the  function is never used, per se, as any graphics call goes through one of its derived classes. So a call to  would have its first chance to render a square by the   class. If no native implementation exists, then the base class is called, at which point the virtualization takes over and means that  is called. This gives the  class a “second chance” to use native code, if any is available.

With this method it is, theoretically, possible to build an entire 3D engine (applying software rasterizing) using only one native function in the form of DrawPoint, with other functions being implemented as and when time permits. In practice this would be hopelessly slow, but it does demonstrate the possibilities for double-chance functions.