Architecture

PathWyse is organized in four layers: entry points, data, algorithms, and utilities.

PathWyse
├── Entry Points
│   ├── Solver
│   └── PWSolver
├── Data
│   ├── Problem
│   ├── Graph
│   ├── Resource
│   │   ├── DefaultCost
│   │   ├── Capacity
│   │   ├── Time
│   │   ├── NodeLimit
│   │   └── TimeWindows
│   ├── ResourceData
│   ├── Path
│   └── BoundLabels
├── Algorithms
│   ├── Algorithm
│   ├── Preprocessing
│   ├── Dijkstra
│   ├── PWDefault  ·  LMDefault
│   ├── PWBucket   ·  LMBucket
│   ├── PWAcyclic  ·  LMacyclic
│   └── Label  ·  LabelAdv
└── Utilities
    ├── Parameters
    ├── DataCollector
    ├── Logger
    ├── Bitset
    └── constants

Entry Points

Class	Description
`Solver`	Main C++ interface. Owns the problem and the algorithm pipeline; exposes all setup, solve, and query operations.
`PWSolver`	Cython wrapper around `Solver`. Provides the same interface to Python.

Data

Class	Description
`Problem`	Central data object. Holds the graph, the objective, all resources, and the current solution pool.
`Graph`	Adjacency structure. Stores nodes, directed arcs, and active-node bookkeeping used during preprocessing.
`Resource`	Base class for all resource types. Defines bounds, consumption data, and the extension interface.
`ResourceData`	Raw numerical data for a resource: per-node and per-arc costs, stored in dense or sparse form depending on problem size.
`Path`	A solution: tour sequence, objective value, and resource consumption vector.
`BoundLabels`	Forward and backward bound labels per node, produced by preprocessing and used to prune the search.

Resource types

Class	Description
`DefaultCost`	Objective function. Accumulates arc and node costs along the path.
`Capacity`	Capacity resource. Consumption is collected on nodes only.
`Time`	Time resource. Supports both arc travel time and node service time.
`NodeLimit`	Limits the number of nodes in a path.
`TimeWindows`	Enforces time window constraints at individual nodes.

Algorithms

Class	Description
`Algorithm`	Abstract base class for all algorithms. Defines the `solve` / `reset` interface and holds a reference to the problem.
`Preprocessing`	Runs Dijkstra-based rounds to compute dual bounds and prune nodes unreachable from origin or destination.
`Dijkstra`	Single-source shortest path used internally by `Preprocessing`.
`PWDefault`	Main bidirectional DP labeling algorithm. Supports DSSR, NG-path relaxations, multiple candidate and join strategies.
`LMDefault`	Label manager for `PWDefault`. Maintains forward and backward label pools, handles dominance checks, candidate selection, and join operations.
`PWBucket`	Bucket-based DP algorithm for cyclic networks. Organizes labels into resource-indexed buckets for a label-setting extension.
`LMBucket`	Label manager for `PWBucket`. Manages bucket structure, insertion, and dominance checks.
`PWAcyclic`	A*-based algorithm for single-resource acyclic networks.
`LMacyclic`	Bucket-based label manager for `PWAcyclic`.
`Label`	Base label. Stores resource values, objective, current node, direction, and a pointer to the predecessor.
`LabelAdv`	Extended label with a visited-nodes bitset, required for NG-path relaxations and DSSR.

Utilities

Class	Description
`Parameters`	Global configuration. Reads and exposes all settings from `pathwyse.set`.
`DataCollector`	Collects algorithm performance data (times, iterations, bounds) and writes it to disk.
`Logger`	Verbosity-controlled console output with color support.
`Bitset`	Dynamic bitset used to track visited nodes in labels efficiently.
`constants.h`	Global numeric constants and status codes shared across the library.