Methodology
To address such questions, we propose to proceed from looking at two kinds of texts: program texts and meta-texts. The core of our corpus will consist of the two categories, with additional texts and tools involved.
Due to the intricate relationship between source code and digital communication networks, vast amounts of source code are available online natively or have been digitized. They range from a few lines to several thousands, date between 1969 and 2021, with a majority written by authors in Northern America or Western Europe. On one side, code snippets are short, meaningful extracts usually accompanied by a natural language comment in order to illustrate a point. On the other, extensive code bases are large ensembles of source files, often written in more than one language, and embedded in a build system3
. Both can be written in a variety of programming languages, as long as these languages are composed in unicode-encoded alphanumeric characters.
This lack of limitations on size, date or languages stems from our empirical approach. Since we intend to assess code conditionally, that is, based primarily on its own, intrinsic textual qualities, it would not follow that we should restrict to any specific genre of program text. As we carry on this study, distinctions will nonetheless arise in our corpus that align with some of the varieties amongst source—for instance, the aesthetic properties of a program text composed of one line of code might be different from those exhbited by a program text made up of thousands of lines code.
We also intend to use source code in both a deductive and an inductive manner. Through our close-reading of program texts, we will highlight some aesthetic features related to its textuality, taking existing source code as concrete proof of their existence. Conversely, we will also write our own source code snippets in order to illustrate the aesthetic features discussed in natural language. We will make use of this technique in order to illustrate some of our points. Rather than discussing complex code snippets, we will sometimes list translated, simplified versions in the Python programming language, and refer to the reader to the actual listings in the annex. This use of source code snippets is widely spread among communities of programmers in order to qualify and strengthen their points in online discussions, and we intend to follow this weaving in of machine language and natural language in order to support our argumentation. This approach will therefore oscillate between theory and practice, the concrete and the abstract, as it both extracts concepts from readings of source code and illustrates concepts by writing source code.
The case of programming languages is a particular one: they do not exclusively constitute program texts (unless they are considered strictly in their implementation details as lexers, interpreters and compilers, themselves described in program texts), but are a necessary condition for the existence of source code. They therefore have to be taken into account when assessing the aesthetic features of program text, as integral part of the affordances of source code. Rather than focusing on their context-free grammars or abstract notations, or on their implementation details, we will focus on the syntax and semantics that they allow the programmer to use. Programming languages are hybrid artefacts, and their intrinsic qualities are only assessed insofar as they relate to the aesthetic manifestations of source code written in those languages.
Meta-texts on source code make up our secondary corpus. Meta-texts are written by programmers, provide additional information, context, explanation and justification for a given extract of source code, and is a significant part of the software ecosystem. Even though they are written in natural langauge, this ability to write comments has been a core feature of any programming language very early on in the history of computing, linking any program text with a potential commentary, whether directly among the source code lines (inline commentary) or in a separate block (external commentary)4
. Examples of external commentaries include user manuals, textbooks, documentation, journal articles, forums discussions, blog posts or emails. The inclusion in our corpus of those meta-texts is due to two reasons: the practical reason of the high epistemological barrier to entry when it comes to assessing source code in unfamiliar linguistic or hardware environments, and the theoretical reason of including the aesthetic judgment of programmers as it supports our conditional, rather than constitutive, approach.
While we intend to look at source through close-reading, favoring the role and essence of each line as a meaningful, structural element, rather than that of the whole, our interpretation of meta-texts will take place via discourse analysis. Building on Dijk and Kintsch's work on discourse comprehension (
Dijk, 1983)
Strategies of Discourse Comprehension by T. A. Dijk, W. Kintsch, 1983. [link]
, we intend to approach these texts at a higher level, in terms of the lexical field they use, as a marker of the aesthetic field they refer to, as well as at a lower level, noting which specific syntactic aspects of the code they refer to. This focus on both the micro-level (e.g. local coherence and proposition analysis) and on the macro-level (e.g. socio-cultural context, intended aim and lexical field usage) will allow us to link specific instances of written code with the broader semantic field that they exist in. This connection between micro- and macro- relies on the hypothesis that there is something fundamentally similar between a source code construct, its meaning and use at the micro-level, and the aesthetic field to which it is attached at a macro-level, a hypothesis we will address further when investigating the role of metaphor in source code. In this aim, we will also mobilize metaphor theory from Lakoff to identifiy some of the properties of code as a target domain through some of the features of the aesthetic fields taken as source domains (
Lakoff, 1980)
Metaphors We Live By by George Lakoff, Mark Johnson, 1980. [link]
.In the end, this process will allow us to construct a framework from empirical observations. The last part of our methodology, after having completed this analysis of program-texts and their commentaries, is to cross-reference it with texts dealing with the manifestation of aesthetics in those peripheral fields. Literary theory, centered around the works of Mary-Laure Ryan, Roland Barthes and Paul Ricoeur can shed light on the attention to form, on the interplay of syntax and semantics, of open and closed texts, and suggest productive avenues through the context of metaphor. Architectural theory will be involved through the two main approaches mentioned by software developers: functionalism as illustrated by the credo form follows function and works by Vitruvius, Louis Sullivan and the Bauhaus on one side, and pattern languages as initiated by the work of Christopher Alexander on the other. Mathematical beauty will be considered in its capacity to communicate complex concepts as well as to act as a heuristic when developing proofs for complex theorems, as explicited by scholars such as Gian-Carlo Rota and Nathalie Sinclair. Throughout, we will see how an approach to craft, as the enactment of tacit knowledge in the creation of functional artefact can apply these domains.
This study therefore aims at weaving in empirical observations, discourse analysis and external framing, in order to propose systematic approaches to source code's textuality. However, these will not unfold in a strictly linear sequence; rather, there will be a constant movement between practice and theory and between code-specific aesthetic references and broader ones: this interdisciplinary approach intends to reflect the multifaceted nature of software.
