This library provides a generic and modular way to prove the correctness, including stability, of several mergesort functions. The correctness lemmas in this library are overloaded using a canonical structure (StableSort.function). This structure characterizes stable mergesort functions, say sort, by its abstract version asort parameterized over the type of sorted lists and its operators such as merge, the relational parametricity of asort, and two equational properties that asort instantiated with merge and concatenation are sort and the identity function, respectively, which intuitively mean that replacing merge with concatenation turns sort into the identity function. From this characterization, we derive an induction principle over traces—binary trees reflecting the underlying divide-and-conquer structure of mergesort—to reason about the relation between the input and output of sort, and the naturality of sort. These two properties are sufficient to deduce several correctness results of mergesort, including stability. Thus, one may prove the stability of a new sorting function by defining its abstract version, proving the relational parametricity of the latter using the parametricity translation (the Paramcoq plugin), and manually providing the equational properties.

As an application of the above proof methodology, this library provides two kinds of optimized mergesorts. The first kind is non-tail-recursive mergesort. In call-by-need evaluation, they allow us to compute the first k smallest elements of a list of length n in the optimal O(n + k log k) time complexity of the partial and incremental sorting problems. However, the non-tail-recursive merge function linearly consumes the call stack and triggers a stack overflow in call-by-value evaluation. The second kind is tail-recursive mergesorts and thus solves the above issue in call-by-value evaluation. However, it does not allow us to compute the output incrementally regardless of the evaluation strategy. Therefore, there is a performance trade-off between non-tail-recursive and tail-recursive mergesorts, and the best mergesort function for lists depends on the situation, in particular, the evaluation strategy and whether it should be incremental or not. In addition, each of the above two kinds of mergesort functions has a smooth (also called natural) variant of mergesort, which takes advantage of sorted slices in the input.

• Author(s):
  • Kazuhiko Sakaguchi (initial)
  • Cyril Cohen
• License: CeCILL-B Free Software License Agreement
• Compatible Coq versions: 8.13 or later
• Additional dependencies:
  • MathComp 1.13.0 or later
  • Paramcoq 1.1.3 or later
  • Mczify (required only for the test suite)
  • Equations (required only for the test suite)
• Coq namespace: stablesort
• Related publication(s):
  • A bargain for mergesorts (functional pearl) — How to prove your mergesort correct and stable, almost for free doi:10.48550/arXiv.2403.08173

The easiest way to install the development version of Stable sort algorithms in Coq is via OPAM:

git clone https://github.com/pi8027/stablesort.git
cd stablesort
opam repo add coq-released https://coq.inria.fr/opam/released
opam install ./coq-stablesort.opam

The mergesort functions and the stability proofs provided in this library are mostly based on ones in the path library of Mathematical Components.