aporter/jwzoomswitcher
Watch
1
0
Fork
You've already forked jwzoomswitcher
 0
Code Issues Pull requests Projects Releases 2 Packages Wiki Activity Actions

v0.2.1: focus hand-back, faster switch-back, smaller support file #6

Merged
aporter merged 4 commits from fix/v0.2.1-tester-findings into main 2026-11-04 11:38:08 +10:00
Conversation 0 Commits 4 Files changed 16 +413 -53
aporter commented 2026-11-04 11:36:29 +10:00
Owner

Fixes from a tester's 0.2.0 support file (Windows 11 26300, Zoom 7.2.5, JW Library Store app): 51 presses, all OK, but:

  1. Focus hand-back failed 9/9 after JW Library had to be re-activated. setForeground now tries SetForegroundWindow, then an empty injected mouse input (no movement or buttons; the PowerToys technique) and tries again, then the old Alt tap (now down+up in one SendInput). Injections run on koffi's worker thread, bounded to 250 ms; a stuck one is never stacked. A hand-back that lands late is watched for up to 300 ms inside the JW recovery deadline, and JW is re-checked afterwards. The log says which method worked.
  2. Switch-back took ~2 s: a fixed 1.5 s settle after JW was already showing. JW now has to stay showing for 300 ms (jwStableMs). VM: JW restore → showing 306–314 ms (was ~1.5 s).
  3. Diagnostics appeared twice in support files: the full JSON is no longer logged (33 KB line); the log gets the summary and the diagnostics-*.json name.
  4. Version 0.2.1 (package, lockfile, manifest, README, INSTALL).

Tests

  • tsc, 125/125 unit tests (10 new; each switch-back/focus test verified to fail without its fix), build, streamdeck validate, gitleaks, whitespace check.
  • Windows VM (real key presses on the emulated MK.2): toggle, triple press, JW minimised, JW minimised 20 s: 6/6 PASS, no WARN/ERROR in the plugin log. SendInput encoding verified on Windows (1 and 2 events inserted; wrong size refused). Results: testing/results/vm-v0.2.1-2026-11-04.md.
  • Not verifiable in the VM: the focus fallback itself (JW never takes the foreground there). The tester's next support file will show it.
  • workbench validate fails on workbench's own skill-link self-tests, unrelated to this repo.

Review

Review-gate: gpt-6.1-sol ship

  • Plan (Gate 1): fix-then-ship. Blockers resolved: JW re-checked after the focus wait; focus watch bounded by the recovery deadline.
  • Diff (Gate 2): fix-then-ship. Blocker resolved: SendInput moved off the plugin's thread (bounded worker, no stacking, regression tests). Re-check: ship; its two comment nits fixed.

🤖 Generated with Claude Code

Fixes from a tester's 0.2.0 support file (Windows 11 26300, Zoom 7.2.5, JW Library Store app): 51 presses, all OK, but: 1. **Focus hand-back failed 9/9** after JW Library had to be re-activated. `setForeground` now tries `SetForegroundWindow`, then an empty injected mouse input (no movement or buttons; the PowerToys technique) and tries again, then the old Alt tap (now down+up in one `SendInput`). Injections run on koffi's worker thread, bounded to 250 ms; a stuck one is never stacked. A hand-back that lands late is watched for up to 300 ms inside the JW recovery deadline, and JW is re-checked afterwards. The log says which method worked. 2. **Switch-back took ~2 s**: a fixed 1.5 s settle after JW was already showing. JW now has to stay showing for 300 ms (`jwStableMs`). VM: JW restore → showing 306–314 ms (was ~1.5 s). 3. **Diagnostics appeared twice in support files**: the full JSON is no longer logged (33 KB line); the log gets the summary and the `diagnostics-*.json` name. 4. Version **0.2.1** (package, lockfile, manifest, README, INSTALL). ### Tests - `tsc`, 125/125 unit tests (10 new; each switch-back/focus test verified to fail without its fix), build, `streamdeck validate`, gitleaks, whitespace check. - Windows VM (real key presses on the emulated MK.2): toggle, triple press, JW minimised, JW minimised 20 s: 6/6 PASS, no WARN/ERROR in the plugin log. `SendInput` encoding verified on Windows (1 and 2 events inserted; wrong size refused). Results: `testing/results/vm-v0.2.1-2026-11-04.md`. - Not verifiable in the VM: the focus fallback itself (JW never takes the foreground there). The tester's next support file will show it. - `workbench validate` fails on workbench's own skill-link self-tests, unrelated to this repo. ### Review Review-gate: gpt-6.1-sol ship - Plan (Gate 1): fix-then-ship. Blockers resolved: JW re-checked after the focus wait; focus watch bounded by the recovery deadline. - Diff (Gate 2): fix-then-ship. Blocker resolved: `SendInput` moved off the plugin's thread (bounded worker, no stacking, regression tests). Re-check: ship; its two comment nits fixed. 🤖 Generated with [Claude Code](https://claude.com/claude-code) 
From a tester's 0.2.0 support file (51 presses, all OK; Windows 11 26300):
- Focus hand-back after re-activating JW failed 9/9. setForeground now
  tries SetForegroundWindow, then an empty injected mouse input (no
  movement or buttons; the PowerToys technique) before trying again, then
  the old Alt tap. Returns which method worked; a hand-back that lands late
  is watched for up to 300 ms within the JW recovery deadline. The log says
  which method returned focus, or warns as before.
- JW is re-checked after the focus wait, so a JW that re-hides meanwhile is
  not reported as shown (record kept, next press retries).
- Switch-back took ~2 s: a fixed 1.5 s settle after JW was already
  showing. It must now stay showing for jwStableMs (300 ms, at most
  settleMs); waits for a change to show keep settleMs.
- The diagnostics snapshot is no longer logged as a 33 KB JSON line (it
  appeared twice in every support file); the log gets the summary and the
  diagnostics-*.json name.
- Version 0.2.1 (package, lockfile, manifest, README, INSTALL).

123 unit tests (8 new, each verified to fail without its fix).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Review (gpt-6.1-sol, fix-then-ship): SendInput can block inside the kernel
(PowerToys hit this with the same empty-input technique), and it ran on the
plugin's only thread. The empty mouse input and the Alt tap (which used
keybd_event, the same call) now go through koffi's .async on a worker
thread, waited for at most 250 ms; while an earlier injection hasn't
returned, further ones are skipped so stuck workers never pile up. The Alt
tap is now down+up in one SendInput call, so a late or partial injection
can't leave Alt held. setForeground is async and typed as
"direct" | "input" | "alt" | null.

125 unit tests (bounded worker: never-returning call, no stacking, recovery,
throwing start).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
VM test results for 0.2.1
All checks were successful
ci / secrets (pull_request) Successful in 14s
Details
ci / build (pull_request) Successful in 31s
Details
b5ca0e82d5 
Co-Authored-By: Claude Opus 5.5 <[email protected]>
aporter deleted branch fix/v0.2.1-tester-findings 2026-11-04 11:38:09 +10:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
aporter/jwzoomswitcher!6
No description provided.