Synthesis and Repair for Functional Programming: A Type- and Test-Driven Approach

Doctoral thesis, 2024

Modern programs in languages like Haskell include a lot of information beyond what is strictly required for compilation, such as unit tests, property-based tests, and type annotations more specific than those necessary to resolve ambiguity. This additional specification is usually only used for post-compilation verification, by running the tests to verify that the code-as-written matches the specification provided by the types and properties. In this thesis, we explore ways of going beyond verification, and how this additional information can aid the developer during development.

This can be done in multiple ways, for example by helping the programmer write an implementation that matches the specification, by helping the programmer track down the source of a bug in the implementation, and even automatically repairing the implementation when it doesn't match the specification.

In the first part, I explore the integration of program synthesis into GHC compiler error messages using typed-hole suggestions to aid completion of partial programs during development. In the second part, we present PropR, an automatic repair tool. PropR is based on type-driven synthesis, guided by property-based testing and fault localization in conjunction with genetic algorithms. A rich specification is required for these approaches to be effective. This motivates third part, where we present Spectacular, a tool that uses ECTA-based synthesis to automatically infer properties of programs, and lets us bootstrap specifications from previous versions. In the fourth and fifth part of this thesis, we explore how to localize program faults, and present CSI: Haskell and TastySpectrum, lightweight trace-based and spectrum-based fault localization tools that help programmers and repair tools understand likely sources of a bug.