Package org.apfloat.spi


package org.apfloat.spi
The apfloat Service Provider Interface (SPI).

The apfloat API is a high-level API that defines algorithms on the level of e.g. the Newton iteration for the inverse of a number. Behind this high-level API there is a lot of low-level functionality that makes all the arbitrary precision arithmetic happen. The digits of a large number are stored in an array of ints, for example. In fact, an Apfloat is structurally just a pointer to an ApfloatImpl, and most of the functionality of the Apfloat class is simply delegated to the underlying ApfloatImpl.

The apfloat SPI defines the general interface for the low-level things that must happen behind the scenes of the high-level API. An actual implementation of the SPI can be optimized for different things, for example:

  • Size of numbers: different algorithms are efficient for numbers with with 1000 or 1000000 digits. This applies both to the actual storage method of the data, and the mathematical algorithms used for e.g. multiplying numbers.
  • Memory consumption vs. performance: different types of Fast Fourier Transform (FFT) based algorithms can be used to find a suitable trade-off between memory consumption and performance.
  • Hardware architecture: 32-bit and 64-bit systems handle int and long type elements correspondingly most efficiently, for example. Some systems perform floating-point operations (with float or double type elements) faster than integer operations (int or long).
  • Complexity: a more complex implementation may be optimized for different cases, however more code will take more space and use more memory. This may be a concern on some systems (e.g. mobile devices).
A Service Provider is only required to implement the BuilderFactory interface, and actually only the BuilderFactory.getApfloatBuilder() method in this interface. All apfloat implementations (ApfloatImpl) are created through the ApfloatBuilder interface's methods. The rest of the interfaces in the SPI exist only for the convenience of the default apfloat SPI implementations (org.apfloat.internal).

The apfloat SPI suggests the usage of various patterns, as encouraged by the specification of all the interfaces in the SPI. These patterns include:

  • Abstract factory pattern, for getting instances of the various builders, as well as other types of components built by the different builders (ApfloatBuilder, DataStorageBuilder, ConvolutionBuilder, NTTBuilder).
  • Factory method pattern; obviously the abstract factories use factory methods to create instances of objects.
  • Builder pattern: an ApfloatImpl needs various "parts" for its structural construction (DataStorage) as well as its behavior (ConvolutionStrategy). Builders are used to build the different sub-parts needed, and the ApfloatImpl itself only knows the high-level algorithm for how the parts are used and related. The construction of the sub-part details is left for the builders, and the ApfloatImpl accesses the parts only via an interface.
  • Strategy pattern: for multiplying numbers, completely different algorithms are optimal, depending on the size of the numbers. The ConvolutionStrategy defines different convolution algorithms to be used in the multiplication. For very large numbers, a transform-based convolution can be used, and even a different transform strategy can be specified via the NTTStrategy interface.
  • Iterators are used for iterating through DataStorage elements in a highly flexible manner. The base class is DataStorage.Iterator. For example, a data storage that uses a simple array to store the entire data set in memory can return a simple iterator that goes through the array element by element. In comparison, a data storage that stores the data in a disk file, can have an iterator that reads blocks of data from the file to a memory array, and then iterates through the array, one block at a time.
  • Singleton pattern, assumed to be used in the BuilderFactory class, as there should be no need to have more than one instance of each builder class. Also the BuilderFactory instance itself is a singleton, within an ApfloatContext.
  • Bridge pattern: the SPI itself is the bridge pattern. An Apfloat provides a simple high-level programming interface and the complex technical implementation details are delegated to an ApfloatImpl. The Apfloat class can be subclassed for additional functionality, and independent of that, different subclasses of an ApfloatImpl can be used to optimize the implementation.
Associations of the SPI classes are shown in a class diagram format below:

Class diagram

The class implementing BuilderFactory that is used in creating apfloat implementations is defined in the ApfloatContext. You can set the BuilderFactory instance programmatically by calling ApfloatContext.setBuilderFactory(BuilderFactory), for example:

BuilderFactory builderFactory = new MyBuilderFactory();
ApfloatContext.getContext().setBuilderFactory(builderFactory);
It's a lot easier to specify this to happen automatically whenever your program starts. To do this just specify the BuilderFactory class name in the apfloat.properties file (or the apfloat ResourceBundle if you use one). For example, the apfloat.properties file might contain the line:
builderFactory=org.mycompany.MyBuilderFactory
For more details about configuring the apfloat BuilderFactory, see the documentation for ApfloatContext.
See Also: