Skip to content

performance-release-matrix.md

Latest commit

 

History

History
130 lines (106 loc) · 8.07 KB

performance-release-matrix.md

File metadata and controls

130 lines (106 loc) · 8.07 KB
Raw

Performance and interaction release matrix

Tracking issue: #20

Runtime baseline and branch base

Item Value
Runtime baseline commit be09b042d8386083cc49ecba88828a658e6d4257
Experiment branch base performance-baseline-v1
Release helper SHA-256 8ce79405bf59aa2afd516f1f6e0426887609993cd29d842bcba2fe03abff5a2b
Release helper bytes 432552
Architecture arm64
Swift Apple Swift 6.3.2, Swift 5.9 package mode

All isolated feature branches in this series start from the performance-baseline-v1 tag, which includes the measurement infrastructure. Worker comparisons use the runtime baseline values above because issue #20 did not change the release helper. Runtime candidates are not merged to main until their isolated reports are complete.

Candidate register

Issue Branch Performance report Isolated status Interaction status
#21 fix/reset-interrupted-gesture reset-interrupted-gesture.md accept gesture interaction accepted
#22 fix/use-cgevent-movement-deltas cgevent-movement-deltas.md reject not applicable
#23 fix/abort-stale-gesture-holds stale-gesture-hold.md accept gesture interaction accepted
#24 perf/reset-scroll-smoothing-state reset-scroll-smoothing-state.md accept clean-idle and single-device physical checks passed; active performance pending
#25 fix/preserve-scroll-delta-fields preserve-scroll-delta-fields.md accept clean-idle and single-device physical checks passed; active performance pending
#26 refactor/standardize-config-defaults standardize-config-defaults.md accept not applicable
#27 release/public-distribution-readiness public-distribution-readiness.md accept not applicable

Valid isolated states are pending, accept, redesign, and reject.

Required combinations

Group Candidates Reason Status
Gesture recovery #21 + #23 Both cancel or reset held-gesture state accept
Gesture movement #22 + #23 Both execute on active movement paths not applicable; #22 rejected
Scroll processing #24 + #25 Both change the transformed scroll callback clean-idle and single-device physical checks passed; active performance pending
Runtime set All isolated runtime candidates marked accept Detect cumulative or nonlinear cost clean-idle and single-device physical checks passed; active performance pending

Complete candidate

Branch Runtime commit Helper bytes __TEXT bytes Status
release/final-feature-candidate 727060f 433528 212992 static, packaging, clean-idle, and single-device physical checks complete; owner gates remain

The complete candidate combines #26 and #27 with the optimized runtime set without changing the runtime-only helper size or text layout. Use final-feature-candidate.md for the final delegated validation record.

Final validation gates

Gate Status Evidence or remaining scope
Integrated clean-idle measurement pass One controlled 150.086-second run at measured commit 727060f
CPU and wakeup stability pass Zero CPU-time, package-idle wakeup, interrupt-wakeup, context-switch, and syscall growth in that run
Thread and memory stability pass Two stable threads; RSS flat at 10,600,448 bytes and physical footprint flat at 3,375,584 bytes
Network, normal logging, and stop lifecycle pass Zero sockets, zero normal log files, and no worker remaining afterward
Physical MX Master behavior pass Maintainer reported gestures, one action per hold, stale cancellation, side buttons, ratchet/free-spin, direction reversal, and pause recovery all worked
MacBook trackpad behavior pass Maintainer reported vertical/horizontal scrolling and trackpad gestures remained normal with MX-only smoothing enabled
Mouse power cycle and disconnect/reconnect pass Maintainer reported gestures, scrolling, and side buttons worked after both recovery paths; transport was not separately recorded
Mac sleep and wake pass Maintainer reported worker and mouse behavior recovered and all mapped controls worked after wake
Controlled active-input performance pending Functional active use passed, but CPU and latency were not measured under a fixed combined workload
Publication-owner gates unchanged Developer ID, notarization, clean-machine Gatekeeper checks, repository visibility, license, and signed-out release links remain owner-controlled

The physical result was reported by the maintainer on 2026-07-27 from one MX Master 3S on macOS 26.5.1. It does not establish compatibility with every MX Master model, connection transport, Mac, or macOS release.

Integration procedure

  1. Create temporary perf/candidate-integration from performance-baseline-v1.
  2. Apply accepted candidates one at a time without merge commits.
  3. Rebuild and rerun strict idle after every addition.
  4. Run each feature-specific active workload.
  5. Compare the cumulative result with the baseline and individual deltas.
  6. Investigate any combined cost that exceeds expected measurement noise.
  7. Record the final accept, redesign, or reject decision here.
  8. Delete the temporary branch when the decision is complete.

Issue #20 infrastructure result

This branch changes documentation, source-archive exclusions, CI packaging checks, and development-only measurement tools. It does not change Swift worker or app sources.

Check Result
Release helper byte-identical to baseline pass: 432552 bytes, identical SHA-256
Strict idle measurement 150.085 s; 0.001569152 s CPU; 0.001046% of one core
Idle wakeups pass: 0 package-idle and 0 interrupt wakeup growth
Worker threads pass: 2 minimum and maximum after 10 s warm-up
RSS 7.547-8.812 MiB; exceeds the 8 MiB target transiently
Physical footprint 3.141-3.157 MiB with no sustained growth
Leak result pass: 0 leaks, 0 leaked bytes
No network socket pass
No normal log file pass
Lifecycle pass: 25 start/duplicate-start/status/stop cycles
Development probes absent from sharing ZIP pass
Tests absent from release helper and sharing ZIP pass
App archive size 438594-byte fresh baseline; 438631-byte candidate (+37 bytes, +0.0084%)
Local tests blocked: selected Command Line Tools do not provide XCTest
GitHub Actions tests pass: unit tests, release build, linkage check, and archive verification
Decision accept

Measurement context

  • MacBook Pro Mac16,8, Apple Silicon
  • macOS 26.5.1 (25F80)
  • Apple Swift 6.3.2
  • 10-second untimed warm-up, then six samples at 30-second intervals
  • normal worker mode with debug disabled
  • no matching MX Master was detected during this run
  • raw evidence remained under /tmp and was not committed

No app-bundle input changed on this branch. A fresh baseline build was 504 bytes larger when uncompressed despite producing a 37-byte smaller ZIP, so the archive difference is build/signing/compression variation rather than an added runtime artifact.

The current unchanged baseline does not satisfy a literal RSS ceiling of 8 MiB on every sample. Future candidate decisions therefore retain 8 MiB as the optimization target while requiring repeated, baseline-relative RSS and physical-footprint comparisons. No branch may claim zero impact by selecting only a favorable sample.