3.1 Programming Paradigms

Programming paradigms provide fundamental styles for implementing software. In the following, a short overview of the individual programming paradigms utilized in this work is given. Extensive comparisons are available in the literature [77][78].

• Imperative programming focuses on executing a sequence of instructions, operating on a state. Imperative programming is the primary utilized programming style in scientific implementations, due to its simplicity and intuitive approach. This paradigm is supported by all major languages relevant to CSE, such as Fortran, C, and C++.
• Object-oriented programming aims for modularity by wrapping functionality into reusable classes, offering defined interfaces for external utilization [79]. Polymorphism is implemented via so-called virtual class hierarchies, allowing to specialize implementations according to run-time information. The object-oriented approach is the most widely taught paradigm at universities, especially with mainstream languages like Java and C++. The paradigm has also some basic support by C and Fortran, although the features are not as extensive as with typical object-oriented languages, such as C++.
• Functional programming uses mathematical functions for formulating algorithms and programs by simultaneously avoiding states of objects and mutable data [80]. Functional programming supports so-called lambda expressions, based on the lambda calculus [81]. Pure functional programming languages are available, such as Haskell [82]. Since the new C++ standard 2011, C++ natively supports lambda functions, thus enabling functional style programming. However, previous C++ generations required external libraries to enable functional-style programming, such as Boost Lambda [83] and Boost Phoenix [84].
• Generic programming fosters code reuse by lifting algorithms or data structures from concrete implementations to their most general form [85][86][87]. For instance, in C++ generic programming approaches polymorphism from a different angle than the object-oriented paradigm. Where the latter relies on run-time decisions - introducing additional execution overhead - the first uses compile-time dispatches, omitting run-time dispatch costs. C++ supports generic programming, due to its type system and template mechanism. Also, C++’s standard template library makes extensive use of this paradigm, as the algorithms, such as copy, are separated from the datastructures, like vector, allowing to use an algorithm with different datastructures and vice versa.
• Meta programming enables programs to generate or manipulate themselves or other programs [88][89]. Meta programming techniques are considered to be rather exotic, and are typically not taught in engineering curricula. Template meta programming, as supported by C++, is a type of meta-programming, where the template system is used to generate source code at compile-time. Consequently, compilation times are increased significantly, impeding the development process due to the potentially increased idle time. In C++, template meta programming is used to implement the generic programming paradigm. Although meta programming in general promises significant reduction in run-time costs, it requires advanced programming skills to develop, debug, and maintain implementations. Closely connected to this issue is the term leaky abstraction, referring to the fact that intentionally hidden implementation details are exposed to the developers in case of errors, further complicating debugging. This exposure is triggered by the compile-time nature of template meta programming.