Extract key and flag handling into pure, testable functions - #441

Last of the three test-focused PRs (#439, #440). Two behaviours were only reachable by writing bytes to stdin or by running the CLI, so neither had tests. Both are now ordinary functions.

keyToAction — src/key-actions.ts

Maps a keypress to a GameAction, or undefined for keys the game ignores. Escape stays in ui.tsx, since quitting the app isn't a game action.

GameAction had to move to types.ts first. Otherwise the new module and ui.tsx would import each other. That also removes the existing reducer.ts → ui.tsx cycle (finding 2.2), since the reducer now takes GameAction from types.ts — src/ui.js has no importers left besides cli.tsx.

The extraction is not purely mechanical. useInput ran its checks as four independent ifs, so ctrl+Q dispatched input-letter 'Q' and give-up, and any ctrl chord typed its letter into the row. Returning a single action forces a resolution — ctrl chords are checked first and are never letters. The stray letter was invisible before because the loss screen hides the current row, so no user-visible behaviour changes.

parseFlags — src/cli-options.ts

Turns meow's flags into App's props, or into a message explaining why the game can't start. cli.tsx keeps the process.exit and console.log; only the decision moved, which takes cli.tsx from 73 lines to 51.

A regression from #439, fixed here

While testing the flag path I found that #439 broke --num-boards 0. It replaced opts?.numBoards || 1 with numBoards ?? 1, and ?? does not catch 0 — so the game reached newGame with zero boards and crashed on the first render, where it used to fall back to one board. meow also hands back NaN for a non-numeric value, which broke the same way.

Confirmed against a clean worktree at 6be12e7, before any of these PRs:

Normalising with || inside parseFlags restores the old behaviour and puts it somewhere a test can reach. The negative case is untouched and recorded as it.fails for finding 1.3.

Verified by mutation

Checks

yarn build, yarn lint, yarn prettier --check ., the --test midgame --quit smoke run and the --test bogus error path all pass. 36 → 59 passing plus 4 expected failures.

Where the review stands after these three

Done: reducer coverage, invariants, GameAction cycle, expectEqual argument order, and the two extractions.

Still open, and the natural next step: finding 1.1 — the multi-board guess limit. Three it.fails tests are already sitting on it and will flip to passing the moment guessesUsed moves onto GameState. After that, 1.3 (validate the flags, one it.fails waiting) and 1.2 (floor the blank-row count).

Generated by Claude Code