Programming Paradigms

(Lectures on High-performance Computing for Economists VII)

Jesús Fernández-Villaverde1 and Pablo Guerrón2

• A paradigm is the preferred approach to programming that a language supports.

• Main paradigms in scientific computation (many others for other fields):

1. Imperative.

2. Structured.

3. Procedural.

4. Object-Oriented.

5. Functional.

1
Paradigms II

• Multi-paradigm languages: C++, introduction a few years go of λ-calculus features.

• Different problems are better suited to different paradigms.

• You can always “speak” with an accent.

• Idiomatic programming.

2
Imperative, structured, and
procedural
Imperative

• Oldest approach.

• Closest to the actual mechanical behavior of a computer⇒ original imperative languages were
abstractions of assembly language.

• A program is a list of instructions that change a memory state until the desired end state is achieved.

• Useful for quite simple programs.

• Difficult to scale.

• Soon, it led to spaghetti code.

3
Structured

• Go To Statement Considered Harmful, by Edsger Dijkstra in 1968.

• Structured program theorem (Böhm-Jacopini): sequencing, selection, and iteration are sufficient to
express any computable function.

• Hence, structured: subroutines/functions, block structures, and loops, and tests.

• This is the paradigm you are likely to be most familiar with.

4
Procedural

• Evolution of structured programming.

• Divide the code into procedures: routines, subroutines, modules, methods, or functions.

• Advantages:

1. Division of work.

2. Debugging and testing.

3. Maintenance.

4. Reusability.

5
OOP
Object-oriented programming I

• Predecesors in the late 1950s and 1960s in the LISP and Simula communities.

• 1970s: Smalltalk from the Xerox PARC.

• Large impact on software industry.

• Complemented with other tools such as design patterns or UML.

• Partial support in several languages: structures in C (and structs in older versions of Matlab).

• Slower adoption in scientific and HPC.

• But now even Fortran has OO support.

6
Object-oriented programming II

• Object: a composition of nouns (numbers, strings, or variables) and verbs (functions).

• Class: a definition of an object.

• Analogy with functional analysis in math.

• Object receives messages, processes data, and sends messages to other objects.

7
Object-orientated programming: basic properties

• Encapsulation.

• Inheritance.

• Polymorphis.

• Overloading.

• Abstraction penalty.

8
Example in Matlab

• Class household.

• We create the file household.m.

• We run Example Use Class.m.

• Public, private, and protected properties and methods.

9
Functional Programming
Functional programming

• Nearly as old as imperative programming.

• Created by John McCarthy with LISP (list processing) in the late 1950s.

• Many important innovations that have been deeply influential.

• Always admired in academia but with little practical use (except in Artificial Intelligence).

10
11
Theoretical foundation

• Inspired by Alonzo Church’s λ-calculus from the 1930s.

• Minimal construction of “abstractions” (functions) and substitutions (applications).

• Lambda Calculus is Turing Complete: we can write a solution to any problem that a computer can
solve.

• John McCarthy is able to implement it in a practical way.

• Robin Milner creates ML in the early 1970’ s.

12
Why functional programming?

• Recent revival of interest.

• Often functional programs are:

1. Easier to read.

2. Easier to debug and maintain.

3. Easier to parallelize.

• Useful features:

1. Hindley–Milner type system.

2. Lazy evaluation.

3. Closures.

13
Main idea

• All computations are implemented through functions: functions are first-class citizens.

• Main building blocks:

1. Immutability: no variable gets changed (no side effects). In some sense, there are no variables.

2. Recursions.

3. Curried functions.

4. Higher-order functions: compositions (≃operators in functional analysis).

14
Interactions

• How do we interact then?

1. Pure functional languages (like Haskell): only limited side changes allowed (for example, I/O) and
tightly enforced to prevent leakage.

2. Impure functional languages (like OCalm or F#): side changes are allowed at the discretion of the
programmer.

• Loops get substituted by recursion.

• We can implement many insights from functional programming even in standard languages such as
C++ or Matlab.

15
Functional languages

• Main languages:

1. Mathematica.

2. Common Lisp/Scheme/Clojure.

3. Standard ML/Calm/OCalm/F#.

4. Haskell.

5. Erlang/Elixir.

6. Scala.

16