General-purpose programming languages typically define literal notation for only a small number of common data structures, like lists. This is unsatisfying because there are many other data structures for which literal notation might be useful, e.g. finite maps, regular expressions, HTML elements, SQL queries, syntax trees for various languages and chemical structures. There may also be different implementations of each of these data structures behind a common interface that could all benefit from common literal notation. This paper introducestyped literal macros (TLMs), which allow library providers to define new literal notation of nearly arbitrary design at any specified type or parameterized family of types. Compared to existing approaches, TLMs are uniquelyreasonable. TLM clients can reason abstractly, i.e. without examining grammars or generated expansions, about types and binding. The system only needs to convey to clients, via secondary notation, the inferredsegmentationof each literal body, which gives the locations and types of spliced subterms. TLM providers can reason modularly about syntactic ambiguity and expansion correctness according to clear criteria. This paper incorporates TLMs into Reason, an emerging alternative front-end for OCaml, and demonstrates, through several non-trivial case studies, how TLMs integrate with the advanced features of OCaml, including pattern matching and the module system. We also discuss optional integration with MetaOCaml, which allows TLM providers to be more confident about type correctness. Finally, we establish these abstract reasoning principles formally with a detailed type-theoretic account of expression and pattern TLMs for “core ML”.

@article{paperbot269,
  title = {Reasonably programmable literal notation},
  author = {Cyrus Omar and Jonathan Aldrich},
  journal = {Proceedings of the ACM on Programming Languages},
  volume = {2},
  number = {ICFP},
  year = {2018},
  doi = {10.1145/3236801}
}