Skip to content

test

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

test

README.md

Solver regression suite

The solver was rewritten from a single 3,754-line file into small modules (js/cube.js, js/predicates.js, js/tables.js, js/solver.js, js/ui.js) without changing the algorithm — the same permutation brute-force over the F2L / beginner's method, keeping the shortest solution across 6 orientations × 24×24 insertion orderings.

These tools load the solver into a Node vm sandbox (mocked DOM, frozen clock so the 800 ms invalid-config timeout never fires and the full deterministic search runs) and check that it produces the exact same solution string for a given scramble — the contract that proves the algorithm is unchanged.

Files

  • lib.mjs — shared core: sandbox loader, scramble runner, seeded generator. Exports SOLVER, the module list loaded in dependency order (same order as index.html).
  • harness.mjs — quick CLI: golden baseline + regression check.
  • compare.mjs — parallel head-to-head comparator (new vs. a reference build).
  • golden.json — recorded solutions for 14 named scrambles (the regression baseline).
  • Rubik.original.js — the pre-refactor solver, kept as the equivalence reference.
  • transpile.mjs — generated js/tables.js (the WHITE_CROSS / SECOND_LAYER lookup tables) from the original goWC / goL2 ladders.
  • gen-cycles.mjs — extracted the Cube.MOVE_CYCLES data (the data-driven turns) from the verified swap-based moves.
  • gen-svg.mjs — generated images/orientation.svg (the isometric "how to hold" cube).

Usage

# Quick regression: does the split solver still match the golden solutions?
node test/harness.mjs --check

# (Re)generate the golden baseline, verifying every solution actually solves its cube:
node test/harness.mjs --golden

# High-coverage proof: run named + N random scrambles through both builds and
# assert byte-identical solutions (sharded across CPU cores).
node test/compare.mjs "js/cube.js,js/predicates.js,js/tables.js,js/solver.js,js/ui.js" test/Rubik.original.js 300

# Regenerate the generated assets:
node test/transpile.mjs > js/tables.js
node test/gen-svg.mjs   > images/orientation.svg

The refactor was validated this way against 1,600+ random scrambles plus the named set, all byte-identical to the original.