Here you can find a solution to a well known problem of multimethods.
It will be demonstrated using famous "crossing shapes example".
The main merits of the proposed solution are:
- no use of type casts of any kind (dynamic, static, reinterpret, const or C-style);
- no use of RTTI;
- no use of preprocessor;
- strong type safety;
- constant time of multimethod execution;
- no dynamic memory allocation (via new or
malloc) is performed during multimethod call;
- compiler doing all the work;
- use of only standard C++ features.
This approach makes it possible for user to write quite readable code.
Suppose, we have the following hierarchy of shapes:
---- Shape_ ------
| | |
Circle_ Rect_ Triangle_
Then user can write code like this:
Shape_* circle = new Circle_( Point_( 5, 5 ), 3 );
Shape_* rect = new Rect_( Point_( 5, 5 ), Point_( 10, 10 ) );
bool cross1 = circle->Cross( rect );
bool cross2 = rect->Cross( circle );
Multimethod is called in following form:
(arg0)->Cross( arg1 )
where types of both arg0 and arg1 (TArg0 and TArg1, respectively) are unknown at compile time.
At runtime the TArg0 type is detected via virtual method call (method Cross) and the TArg1 -
via use of 'deferred dispatching' pattern (Visitor-like pattern, where nodes can be declared
independently of visitor).
User is responsible for providing implementation of crossing detection code
for situations where types TArg0 and TArg1 are known.
Implementations are provided in the form of template class specialization.
struct Crossing_< Circle_, Rect_ >
One can provide some (or all) of the following:
- fully specialized versions of Crossing_ (circles against rectangles, for example),
- symmetric versions of Crossing_ (rectangles against circles via circles against rectangles crossing, for example),
- hierarchy depended versions (crossing of two rectangular shapes, for example),
- fully specialized homogeneous versions (crossing of two circles, for example),
The details are provided in the form of well-commented code in demo project (in file mmvdd_initial.cpp).
- standard compliant C++ compiler,
Keep in mind that some obvious improvements are missed to keep readability.
Some improvements of the initial solution will be provided below.
Deferred dispatcher generation
Someone thinks that manual writing of dispatcher is awkward.
But we can generate a dispatcher from repository typelist, that
contains types of all used concrete shapes.
Suppose there is two libraries of shapes and types provided by them
are joined in two typelists: Shapes0 and Shapes1. One can merge this typelists
and use this libraries together.
Typelists provided by libraries can be easily modified. One can add or remove
Compilation time may grows.
Number of shape types is limited by standard (by template recursion limitation).
The details of this improvement are provided in the form of well-commented
code in demo project (in file mmvdd_generated.cpp).
- Andrei Alexanderscu "Modern C++ Design"
- David Vandevoorde, Nicolai M. Josuttis "C++ Templates: The Complete Guide"
- Erich Gamma, Richard Helm, Ralph Johnson, John Vlissides "Design Patterns: Elements of Reusable Object-Oriented Software"