Skip to content

CALIBER_LEARNINGS.md

Latest commit

 

History

History
26 lines (19 loc) · 1.41 KB

CALIBER_LEARNINGS.md

File metadata and controls

26 lines (19 loc) · 1.41 KB
Raw

candy-testing CALIBER_LEARNINGS

Accumulated patterns and anti-patterns for this library.

Self-testing note

candy-testing is itself a test framework, so its tests are meta — they assert that it correctly tests other things. Self-test by simulating a trivial counter Program. Don't introduce real I/O in candy-testing's own tests — use fixtures.

Patterns

  • Counter fixture — A simple CounterModel that increments on KeyMsg('+') and decrements on KeyMsg('-') is sufficient for self-testing ProgramSimulator's message dispatch, model update, and view capture.

2026-05-28 — ProgramSimulator drives tea programs via scripted input

Pattern: Use ProgramSimulator::for($program)->send(...)->run() to drive TEA programs deterministically without real I/O. Chain multiple ->send() calls for complex input sequences. Override withFakeCmdRunner() to intercept side-effecting commands. Anti-pattern: Don't introduce real stdin/stdout in self-tests — use memory streams and fixtures instead. Source: step-04 ai/candy-testing-new

2026-05-28 — assertGoldenAnsi auto-creates golden files when UPDATE_GOLDENS=1

Pattern: First run with UPDATE_GOLDENS=1 to scaffold a .golden fixture; subsequent runs assert byte-exact match. Golden files live in tests/fixtures/. Anti-pattern: Don't commit golden files that capture non-deterministic output (timestamps, entropy). Source: step-04 ai/candy-testing-new