Extending Intrinsic Classes

You can extend an intrinsic class's programmatic interface by adding new methods and static (class-level) properties. This capability allows you to add new functionality to an intrinsic class, and then access the new functionality using the same syntax that you would use to access native features of the class.

To add a method to an intrinsic class, use the modify statement. The modify syntax for intrinsic classes is almost exactly the same as it is for regular objects:

modify intrinsicClassName
   method1 ( paramList ) { code }
   method2 ...

When you modify an intrinsic class, the intrinsic class must be defined before the modify statement. In most cases, this simply means that you must #include the system header file that defines the intrinsic class before the modify statement in the same compilation unit.

The following example adds a base-2 logarithm function to the BigNumber intrinsic class.

#include <bignum.h>

modify BigNumber
     *   cache ln(2)  use slightly greater precision than we
     *   actually need, to avoid rounding error
    if (BigNumber.cacheLn2_ == nil
        || BigNumber.cacheLn2_.getPrecision() < getPrecision() + 3)
      BigNumber.cacheLn2_ = 
        new BigNumber(2, getPrecision() + 3).logE();

     *   Calculate ln(self), then divide by ln(2) to get the
     *   result (note that ln-base-B of x for any B is equal to
     *   ln(x)/ln(B)).  Reduce the precision of the result back
     *   to our own precision before returning.
    return (self.setPrecision(getPrecision() + 3).logE()
            / BigNumber.cacheLn2_).setPrecision(getPrecision());

  // our caches ln(2) value  we don't have any value initially
  cacheLn2_ = nil

This example illustrates several aspects of intrinsic class extensions.

First, note that we can refer to self within the method we add to the class. We're modifying BigNumber, so self is always a BigNumber object; we can thus refer to its methods, such as getPrecision() and setPrecision(), and we can also use the value in arithmetic.

Second, note that we can add a data property to the class. In the example, we add a property called cacheLn2, which contains a cached value of the natural logarithm of 2. (In the example, we have chosen to cache the value rather than recalculate it each time we call the ln2() method as a performance optimization; calculating ln(2) is fairly expensive, so it makes sense to save the value when we first calculate it, and re-use the same value rather than calculating it on each new call to the method.)

Note, though, that we can only add class-level properties to the class. We cannot add instance properties. In other words, we can only add a property to BigNumber itself, not to each individual BigNumber value. So, we cannot write something like this within an intrinsic class extension method:

self.val1 = 5;  // ILLEGAL!

The statement above is illegal in an intrinsic class extension method because it attempts to store a property value with the object itself, rather than with the intrinsic class.


There are some restrictions on modifying intrinsic classes:

Using Aggregation

There might be times when the intrinsic class extension mechanism is too restrictive for a particular application you have in mind. In particular, you might sometimes find it necessary to add new properties to individual instances of a class; since you can't do this by extending an intrinsic class, you'll have to find an alternative approach in such cases.

One approach that you might consider is aggregation. Aggregation is a common technique in object-oriented programming in which you create a "wrapper" object that contains an instance of a class you want to extend. In other words, rather than subclassing, you create a new, independent class, and store an instance of the class you wish to extend as a property of the new class.

For example, suppose you wanted to create a complex number class. (A complex number is a mathematical entity with two components, a "real" part and an "imaginary" part. The real part is an ordinary real number, and the imaginary part is a real number multiplied by i, the square root of -1, known as the imaginary unit. The two parts are added together to form the complex number.) You can't create a complex number class by extending BigNumber, since there'd be no way to store the two separate numbers making up a complex value. Instead, you could use aggregation.

To create a complex number class using aggregation, we'd start with something like this:

class Complex: object
  construct(r, i) { r_ = r; i_ = i; }
  r_ = nil // the real part
  i_ = nil // the imaginary part

Here, the properties r_ and i_ are simply BigNumber values. We have thus aggregated two BigNumber values into this new class that we call Complex.

Starting in TADS 3.1, you can use take this idea further using operator overloading, which lets you define mathematical operators (such as + and *) on a custom class. This makes it possible to define a Complex class that you can use as though it were a built-in numeric type.

  // adding to the Complex class definition...
  operator+(v) { return new Complex(r_ + v.r_, i_ + v.i_); }
  operator-(v) { return new Complex(r_ - v.r_, i_ - v.i_); }

  // perform some arithmetic on some complex values
  local a, b, c;
  a = new Complex(1.0, 2.0);
  b = new Complex(-3.1, -2.0);
  c = a + b;

With a little more work, we could add the other standard math operators to round out the class. Note that a really robust Complex class would also want to deal with integer and BigNumber values as arguments to the math operators; the example code above is a little too simplistic in that it'll only work with other Complex values.