Myths and mythconceptions: What does it mean to be a programming language, anyhow?

Abstract

Modern software does not stand alone; it is embedded in complex physical and sociotechnical systems. It relies on computational support from interdependent subsystems as well as non-code resources such as data, communications, sensors, and interactions with humans. Both general-purpose programming languages and mainstream programming language research focus on symbolic notations with well-defined abstractions that are intended for use by professionals to write programs that solve precisely specified problems. There is a strong emphasis on correctness of the resulting programs, preferably by formal reasoning. However, these languages, despite their careful design and formal foundations, address only a modest portion of modern software and only a minority of software developers. Several persistent myths reinforce this focus. These myths express an idealized model of software and software development. They provide a lens for examining modern software and software development practice: highly trained professionals are outnumbered by vernacular developers. Writing new code is dominated by composition of ill-specified software and non-software components. General-purpose languages may be less appropriate for a task than domain-specific languages, and functional correctness is often a less appropriate goal than overall fitness for task. Support for programming to meet a specification is of little help to people who are programming in order to understand their problems. Reasoning about software is challenged by uncertainty and nondeterminism in the execution environment and by the increasingly dominant role of data, especially with the advent of systems that rely on machine learning. The lens of our persistent myths illuminates the dissonance between our idealized view of software development and common practice, which enables us to identify emerging opportunities and challenges for programming language research.