test(adapters): hermetic coverage pack for mock, go, rust, and python adapters

[debugmcpdev]

Part of the coverage-expansion sprint (follows the source-map-derive rewrite). Covers the four adapter files on the deficit leaderboard, entirely through the existing hermetic harnesses — no real go/dlv/cargo/python anywhere.

mock-debug-adapter (was 65% — the worst ratio on the board)

The gap was simply the IDebugAdapter surface never being called. Added: full-shape getCapabilities() assertions (the single largest uncovered block) for both a configured and a defaults adapter; buildAdapterCommand incl. the npx .cjs bundled fallback (fs routed through a controllable override — the ESM namespace isn't spy-able); handleDapEvent stopped/continued/terminated/exited state tracking; dispose(); invalid-transition throw; error-scenario initialize; connection delay; and the trivial config surface (transform/default configs, resolvers, install metadata, feature requirements).

go-debug-adapter

Scripted-spawn coverage of the validateEnvironment arms the happy-path tests never hit: Go-too-old (1.17), version-undetermined warning, Delve-without-DAP incl. the sanitized stderr hint; resolveExecutablePath caching; checkDelveVersion banner parse + cache; buildAdapterCommand's DEBUG logging flags; one CI=true test sweeping the six diagnostic console.error lines; trivia getters.

rust-debug-adapter (toolchain suite)

The toolchain harness (full-replacement vi.mock of every utils module) already existed but never called initialize(). Added: initialize success/warnings-logged/invalid-throws-ENVIRONMENT_INVALID/ERROR-state; dispose(); RUST_NOT_FOUND + CARGO_NOT_FOUND warnings; VALIDATION_ERROR catch; win32 dlltool discovery; neither-cargo-nor-rustc throw; the MCP_CONTAINER=true relaxed prebuilt mode; darwin search paths via the existing platform ctor param; resolveAutoSuggestGnu '0'/'false'/'no'; and the cargo.example/cargo.test/default-binary branches of transformLaunchConfig.

python (adapter + utils)

• Adapter: validateEnvironment through the real private helpers (no monkey-patching): version-unknown warning, virtualenv detection incl. the spawn-error arm, checkPythonVersion cache-miss, checkDebugpyInstalled cache reuse; CI=true initialize diagnostics; platform-stubbed getDefaultExecutableName/getExecutableSearchPaths for all three arms; attach stopOnEntry propagation; dependency/feature metadata.
• python-utils (WhichCommandFinder internals): the real Windows fixes that had zero tests — the ComSpec/COMSPEC fallback and the PATH ← Path copy; Store-alias-only candidate lists throwing CommandNotFoundError; the interpreter-validator spawn-error arm; env-var interpreter falling through to auto-detection on CommandNotFoundError; and one test each for the DEBUG_PYTHON_DISCOVERY=true success-side and failure-side diagnostic blocks (incl. the direct-spawn probe) — those two are coverage-value-only by design, one test per block, no refactor.

Not covered on purpose: mock buildAdapterCommand's fileURLToPath-throw fallback, python-utils' dead if (!candidate) continue guard (path.join can't yield falsy).

Full npm test green, npm run lint clean. Leaderboard deltas recorded after merge.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!