Skip to content

Latest commit

 

History

History
94 lines (79 loc) · 5.42 KB

104-windows-pointer-gestures.md

File metadata and controls

94 lines (79 loc) · 5.42 KB

Windows native pointer gestures

Status: source and focused x64 VM qualification complete; cleanup complete. Signed installed, ARM64 live and physical qualification remain open. Owning topics: Windows desktop, Windows resident control.

Objective and completion conditions

Complete Windows desktop parity for native move, drag and scroll through the existing common CLI and retained owner contract. Independently observe pointer motion, drag endpoints/releases and both signed wheel axes. Prove access/scope and operator protections, Pause/Stop interruption and unlocked-origin cleanup. Commit the result, remove owned staging and release the target claim.

Boundaries

Normal input remains target-native. No host pointer, VM window manipulation, arbitrary held-button API or automatic action replay is added. The desktop UAC profile admits pointer operations only on unlocked Default; secure consent remains typed. Windows wheel units are explicit and differ from Mac pixel scroll units. Signed packages, live ARM64, multiple displays, physical devices, elevated-app pointer effects and secure-desktop transitions need separate qualification.

Ordered work and validation

  1. Add typed requests, native capabilities, grant scopes and CLI translation.
  2. Implement physical-pixel motion, bounded drags, signed wheel events and authority-independent release cleanup restricted to the original desktop.
  3. Protect operator endpoints, crossing paths, current wheel routing and pointer capture. Verify refusal geometry, secure policy and client fields.
  4. Use an independent native window-message fixture and the real operator UI to test effects, scopes, Pause/Stop and absence of a held button afterward.
  5. Run formatting verification, desktop/unlock contracts, x64/ARM64 publishes and the embedded desktop build. Remove owned test state, verify baseline readiness and stored credentials, shut down the VM and release its claim.

Result

Current: formatting verification, desktop/unlock contracts, client pointer translation and locked-use client checks pass. The one unavailable secret-pipe fixture remains skipped on this controller. Self-contained x64/ARM64 runtime and native fixture publishes and the embedded desktop build pass.

The staged desktop runtime SHA-256 is 74240fc881b7b8f55d3fab1363bd3c001ed1cf9f0681559a09f10684660987ee. The real desktop operator and independent native window-message fixture pass 36 checks: access-off and observation-only refusals, advertised operations, operator endpoint and crossing-path protection, requested movement, signed horizontal/vertical wheel messages, left/right release at the requested drag endpoint and independently measured hold timing, malformed requests, mid-drag Pause/Stop, native button release, retained consent on Pause, fresh Resume and revoked consent on Stop. Unlocked-origin cleanup leaves the console unlocked. API delivery continues to report application effect as unverifiable; fixture effects are separate.

Inspection of the first passing report found that a nominal 500 ms drag took about five seconds because every delayed sample was still sent. The scheduler now skips elapsed samples after authority checks, binds display geometry and adds independent event timing. The final timing checks require each nominal 500 ms hold to finish in under 2500 ms on this VM; independently measured holds were 514.8 ms and 598.4 ms. This is a focused timing bound, not a hard real-time API guarantee.

An initial run reached the input request before the app's next polling tick hid the protected control announcement. The resident correctly refused the operator overlap. Waiting for the UI tick corrected the harness. A later run passed effects and both interruptions, then hit a harness assertion expecting an explicit null grant; the serializer omits null values. Correcting that assertion produced the complete passing run without weakening product guards.

One bounded scheduler recovery restored guest routes after the known boot stall. Canonical secret-safe cold login and read-only doctor then confirmed full readiness. That infrastructure recovery is separate from feature evidence; no host pointer or VM window was manipulated. The common CLI audit does not reconstruct raw provider steps, and its attribution is self-asserted.

Signed installed, live ARM64, physical/multiple-display, elevated pointer application effects, owner-disconnect/expiry and desktop-transition cleanup remain separate qualification gates. Source cancellation/cleanup fences are present; the focused live interruption evidence is specifically Pause/Stop. Raw observations stay in ignored private evidence, outside public Git history.

Implementation is committed in 9646c7d. Owned fixture and actor processes, the exact scheduled task and guest staging were removed using native PowerShell with containment, task-identity and ancestor/child link checks. An initial combined controller cleanup command was blocked by command policy; separate script preparation and checked execution completed the same cleanup. No protected helper or controller grant was installed by this slice, and no capture artifact was created. Baseline doctor and canonical stored credential verification passed after cleanup. Shutdown reached independently confirmed power-off in 38.0 seconds using the declared 30-second scheduler assistance; the exact target claim was released. This is assisted shutdown evidence.