Overview

unifier is a Rust crate for constraint satisfaction problems (CSP) and constraint optimization problems (COP): integer variables with domains, constraints between them and, for optimization, weighted objectives. Typical instances are scheduling, timetabling and resource allocation.

These problems are generally NP-hard, so there is no single best algorithm. unifier ships several interchangeable solver strategies and is designed as an anytime solver: find a valid solution fast, improve it, and stop whenever a time limit, node budget or cancellation says so.

What it covers

• A constraint graph of variables, domains, constraints and objectives, validated before any solver sees it.
• Sixteen built-in constraints in 0.3.2, including the global constraints AllDifferent, NoOverlap and Cumulative with dedicated propagation.
• Hard/soft scoring: hard constraints decide feasibility, weighted soft terms on three lexicographic levels rank feasible solutions.
• Five solvers: Backtracking, Branch & Bound, Local Search, Large Neighbourhood Search and a parallel portfolio.
• Scheduling primitives (Interval, Activity, Resource, Group) that compile into constraints.
• An incremental feasibility check that evaluates only the constraints touched by a change.

Where it stops

The crate is early (0.3.x). Not covered yet: general unsat cores, serde-based model or solution serialization, a command-line interface, and independent verification against production-scale scheduling scenarios. Evaluate it accordingly before relying on it for production planning.

How the docs are organised

• Getting started: add the crate and solve a first model.
• Guides: modelling, solvers and how unifier relates to pathwise.
• Reference: an overview of the public modules. Item-level documentation lives on docs.rs.