Abstract

Programs can be modularly decomposed in several dimensions. However, it has long been noted that existing programming languages typically suffer from ``the tyranny of the dominant decomposition’’, only supporting decomposition of programs well in one dimension. Bad support for other dimensions leads to crosscutting concerns: code that logically represents some separate, modular functionality of the program, but which is not easily modularized. The main problem lies in existing programming language structuring abstractions, such as algebraic/inductive datatypes in functional languages or object interfaces in object-oriented languages, which dictate the particular flavor of modularity supported by the language.

This talk suggests a form of algebraic signatures, which we generally refer to as algebras, as an alternative programming language structuring abstraction. Algebras do not dictate a particular modularity dimension on the programmer. Instead they support various composition operators which allow them to cater for several dimensions of modularity at once. Algebras have desirable properties of a programming abstraction: they support modular type-checking, separate-compilation and modular reasoning/proofs. I will show how algebras can already be encoded in existing programming languages and theorem provers, and how they can help dealing with several practical problems: from modularizing DSL components, to modularizing inductive proofs and meta-theory of programming languages. I’ll finish the talk by discussing some of the remaining challenges on creating truly algebra-oriented programming languages.