diff --git a/test-coverage.html b/test-coverage.html deleted file mode 100644 index 247dd145660..00000000000 --- a/test-coverage.html +++ /dev/null @@ -1,75 +0,0 @@ - - - - - -Orca E2E & Integration Coverage - - - -
-
Orca test inventory · snapshot 2026-08-12
-

E2E & integration coverage

-

A repository-based view of executable test inventory, CI platform execution, and recorded reliability-gate evidence. It separates tests that exist from tests that actually run on a platform.

-
Bottom line: the largest exposure is cross-platform live evidence. The standard Electron E2E suite is broad but Linux-only in CI. Windows has valuable targeted native, ConPTY, crash-survival, update, and computer-E2E lanes, but not broad desktop E2E parity.
-
-
Playwright Electron specs
246
Headless by default; headful for tagged specs
-
Vitest-included files
4,764
4,643 src · 101 config · 1 tool · 19 E2E unit
-
Computer E2E files
5
macOS 2 · Linux 1 · Windows 2
-
Reliability gates
77
64 partial protection · 13 no protection
-
-

Platform evidence

-

Counts overlap because a gate may target several platforms. “Evidence” means the gate metadata records coverage; it does not imply complete product coverage.

-
-
macOS62 / 71 gates
-
Linux4 / 69 gates
-
Windows3 / 73 gates
-
iOS1 / 2 gates
-
Android0 / 2 gates
-
Mobile0 / 4 gates
-
-

What runs where

- - - - - - - -
LayerCurrent executionLimitation
Standard Electron E2E246 tests/e2e/*.spec.ts; Playwright headless; 10 Ubuntu CI shardsLinux-biased
macOS and Windows are not part of the default E2E workflow.
Headful Electron E2ETagged @headful specs through a separate Playwright projectTargeted
Used for pointer capture and focus-sensitive paths.
Computer/accessibility E2EDedicated macOS, Linux, and Windows jobsAll 3 desktop OSes
Specialized coverage, not full desktop-app parity.
Vitest contracts/integration4,764 included files; real Git, PTY, sockets, HTTP, SQLite, and relay fixtures appear in the suiteMostly deterministic
No formal integration-test taxonomy or source coverage percentage.
Windows native/recoveryWindows PR smoke, ConPTY tests, crash survival, update, and terminal restart workflowsTargeted
Strongest Windows confidence is concentrated in terminal and recovery behavior.
SSH/WSL/paired runtimeDirect SSH, Docker relay, paired-runtime, and WSL fixturesScenario coverage
Live evidence is frequently macOS-hosted; Linux/Windows clients and WSL remain gaps.
-

Highest-value missing coverage

-

Windows behavior

    -
  • WSL restore and shell-resolution parity for PowerShell 5/7, cmd, Git Bash, WSL, and missing shells.
  • -
  • ConPTY keyboard reset after agents/TUIs and modified-Enter routing.
  • -
  • Windows CJK/wide-glyph repaint and pixel-refresh correctness.
  • -
  • Windows input throughput and terminal backpressure budgets.
  • -
  • Startup color-query leakage, restore convergence, and persisted-session recovery.
  • -

Cross-platform live journeys

    -
  • Paired-runtime and remote browser/terminal journeys with Linux and Windows clients.
  • -
  • Real sleep/wake, network partitions, multi-client churn, and reconnect behavior.
  • -
  • WSL-specific live process and PTY restore evidence.
  • -
  • Mixed-version client/host journeys in both skew directions.
  • -
  • CI soak history and per-OS flake tracking.
  • -
-

Registered gaps with no executable protection

-

These are the clearest candidates for adding tests, based on config/reliability-gates.jsonc.

- - - - - - - - - - -
AreaMissing invariantTarget
Startup upgradeRecover old production persisted sessionsmacOS · Linux · Windows
Terminal providerPreserve WSL launch identity, liveness, and restore boundariesWindows
Terminal inputReset Windows ConPTY keyboard protocol after agents/TUIs exitWindows
Terminal renderingRepaint Windows CJK/wide glyphs without stale cellsWindows
Terminal shellResolve Windows local/daemon shells and startup commands consistentlyWindows
Terminal performanceKeep input latency and throughput within explicit budgetsmacOS · Linux · Windows
Terminal capabilityPrevent startup color queries leaking into streamsmacOS · Linux · Windows
ObservabilityDetect renderer/mirror restore divergencemacOS · Linux · Windows
Terminal renderingProve forced repaint does not alter revealed-pane pixelsmacOS · Linux · Windows
-

The gate file contains 13 no-protection entries; nine are explicitly marked “Registered gap only.” The remaining four are tracked as pending-stack or contract gaps.

-

Recommended order

-
  1. Add a Windows Playwright smoke lane covering golden flows, terminal restore/input, file-browser drives, titlebar, settings, and onboarding.
  2. Wire ConPTY keyboard reset, shell parity, and CJK repaint into executable reliability gates.
  3. Promote Linux and Windows paired/SSH journeys from contracts to live CI evidence.
  4. Add explicit unit/integration/E2E/performance classification so coverage reports do not rely on filename heuristics.
  5. Track pass history and flake rate per OS.
-

Sources

- - -