About the port

@johnhenry/math began as a faithful port of the original Mallory ActionScript 3 library, but it fixed bugs rather than carrying them over, and it took advantage of modern JavaScript/TypeScript: ES classes, generics, iterators, Array subclassing, tagged unions, and .ts-native execution. Every module was translated test-first — a test was written against the original’s documented/intended behavior before its TypeScript equivalent existed, which is how the bugs below were caught rather than silently reproduced.

A sampling of the ~40 bugs found and fixed

• RealMath.subtract returned a * b. (Yes, really.)
• Order statistics were broken: sort used the default lexicographic sort, so minimum/maximum/median were wrong for numbers.
• ComplexMath.divide compared alpha.ivalue (lowercase v, always undefined), so all eight directed-infinity results were dead code.
• ComplexMath.normalDistribution reciprocated the whole expression, putting the exponential in the denominator (wrong sign) — the normal PDF was inverted.
• solveN wasn’t Newton’s method — it assumed a derivative of 1 and diverged; now a real numeric Newton–Raphson.
• integrateN sampled at 2·x + interval; differentiateN was a forward difference despite claiming to be symmetric.
• invertMatrix had no pivoting (divide-by-zero on a zero pivot); now uses partial pivoting (and a zero-pivot row swap in Structure, for finite fields).
• crossProduct read index 3 for the z-component and used truthiness tests that dropped zero components; powerMatrix never actually multiplied.
• Vector.setElement infinitely recursed on a falsy slot.
• ComplexNumber string parsing crashed on non-matches and couldn’t round-trip negative imaginary parts; it also swallowed every *, so "4*2" parsed as 42.
• StringEvaluator was uniformly right-associative (10-2-3 → 11); now left-associative for + - * / % and right-associative for ^.
• Polynomial.multiply referenced a non-existent dimension.value; Polynomial.antiderivative dropped its highest-degree term.
• IntegerMath.modulus recursed one step at a time (stack overflow); primeFactors infinite-looped on 0 and negatives.
• Structure’s matrix section was non-functional (static/instance confusion, undefined references) and is reconstructed as working generic linear algebra.
• IntUtils spelled “forty” and “ninety” as “fourty” and “ninty”.

Where the original had no verifiable intended behavior

One area — the Flash 3D ribbon geometry in Graph3DUtils — had no recoverable specification: no test, no comment, no reference implementation to check against, just a block of matrix math whose intended output couldn’t be independently verified. Rather than “fix” it by guesswork, the math is ported verbatim, with an explicit note in the source marking it as such. The Flash rendering utilities (GraphUtils/Graph3DUtils) more broadly now return renderer-agnostic geometry — 2D paths, 3D meshes, as plain data — instead of Flash display objects, since there’s no Flash runtime to hand those to anymore.

That distinction matters for how to read the rest of this history: everywhere else, a fix means the original’s documented or inferable intent was identifiably violated by its own code, and the test written against that intent is what proves the fix is correct — not a preference for “better” math.

This is a snapshot of the port as it happened, not a description of @johnhenry/math’s current API — the library has grown substantially since (CellGraph, Rotor4, exact-fraction and interval arithmetic, and more; see the main page for what’s actually in it today).