Skip to content

liter8 · concurrency

The bug that only happens on the fourteenth Tuesday.

Concurrency is the one area where testing genuinely does not work. A suite that passes a thousand times has taken a thousand samples out of an enormous space, and every engineer knows it — which is why most people’s model of concurrency is a collection of superstitions.

This does something else. You write concurrent code against an injected API, and the engine does not run it once: it explores the space of interleavings, finds the one that breaks your assertion, and hands it back as a schedule you can step through instruction by instruction — forwards and backwards.

Every run reproduces

A run is a list of scheduler decisions and nothing else, so it replays exactly — and the failing one replays on purpose. A concurrency bug stops being a story and becomes an artefact with a link.

The checker is the grader

Nothing here is marked against a rubric. The engine explores, and what it found is the answer — including in predict mode, where naming an output the program cannot produce is exactly as wrong as missing one it can.

It says what it checked

Exhaustive within a bound is a real claim and a limited one. Every result is shown with the space it actually walked, and the word “correct” appears nowhere.

Lessons
21
Programs
21
Named patterns
17

No account, no backend, no analytics. Everything you do stays in this browser, which is why the whole thing can be free.

Everything you do here stays in this browser.