Reduce Juice idle power usage

[EClinick]

What changed

• Make live-power sampling demand-aware: 5 seconds for menu-bar-only background work and 2 seconds while the popover or Stats window is visible.
• Stop background samples from invalidating hidden SwiftUI charts, while keeping the AppKit menu-bar watt label current.
• Cache the helper protocol handshake per connection and cancel settled request timeouts.
• Tear down closed Stats window hosting work and skip battery observers only on batteryless Mac minis.
• Add concurrency, cadence, lifecycle, and laptop battery regression coverage.

Why

The installed app could sustain roughly 38-41% CPU because hidden app-scoped SwiftUI views were re-rendering every two seconds. Helper requests also repeated a connection-scoped handshake. In signed isolated runtime sampling, the patched app was usually at 0% CPU with brief 0.5-0.7% activity, about a 99.4% sampled reduction. A short-window coalition comparison also moved directionally from about 1.07 W to 0.0006 W; that watt figure is diagnostic rather than a calibrated benchmark.

Laptop battery behavior is intentionally preserved: laptops still perform the initial battery read, keep the 60-second polling cadence, and react to Low Power Mode notifications. Only Mac mini mode avoids battery-specific background resources that can produce no readings.

Verification

• DEVELOPER_DIR=/Applications/Xcode.app/Contents/Developer CLANG_MODULE_CACHE_PATH=/tmp/juice-clang-module-cache SWIFTPM_MODULECACHE_OVERRIDE=/tmp/juice-swiftpm-module-cache swift test --no-parallel — 275 tests in 35 suites passed
• swift test --filter BatteryViewModelTests — 2 focused battery tests passed
• git diff --check
• Signed isolated dev-build runtime comparison described above

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
website	Ignored	Preview	Jul 29, 2026 9:09pm

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 58287f489f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Retry stale checkState transport failures

When the connection is invalidated or interrupted while this handshake is in flight and its error handler wins, the caught error may belong to a context that is no longer current. This branch nevertheless publishes .unavailable and returns immediately, so the replacement connection is never checked and a helper crash/update race can falsely report that the helper is unavailable. Mirror the stale-success path by checking isCurrent(context) in the catch and continuing when it is false.

Useful? React with 👍 / 👎.

[EClinick]

Benchmark repeat: launch-idle compute usage

Repeated on the M4 Mac mini (Mac16,10) against exact PR head 8549e42 and installed Juice 0.2.9.

Method

• Restarted production Juice and Juice Dev within the same second.
• Allowed both processes to settle for 45 seconds.
• Sampled both variants simultaneously for 180.803 seconds.
• Confirmed through CGWindowListCopyWindowInfo that neither popover was instantiated or on-screen before or after the run. This result is therefore specifically the true launch-idle workload.
• App metrics come from cumulative proc_pid_rusage(RUSAGE_INFO_V6) deltas sampled once per second.
• App and root-owned helper energy come from their separate resource-coalition CPU/GPU/ANE counters. The total below is app coalition plus its matching helper coalition.

Metric	Installed 0.2.9	PR head	Reduction
Average app CPU	0.732617%	0.055652%	92.4%
Maximum 1-second CPU	2.696750%	1.961529%	27.3%
Instructions/second	28.423M	1.091M	96.2%
CPU cycles/second	17.458M	1.305M	92.5%
Package-idle wakeups/second	0.774	0.100	87.1%
Interrupt wakeups/second	3.667	0.653	82.2%
App coalition compute	0.002317774 W	0.000132155 W	94.3%
Helper coalition compute	0.000869322 W	0.000160737 W	81.5%
Total app + helper compute	0.003187096 W	0.000292892 W	90.8%

Interpretation

This repeat independently confirms that the PR materially reduces background compute. Under true launch-idle conditions, attributable app-plus-helper compute energy was about 10.9x lower. The result is also consistent with the earlier hidden-instantiated-popover run, where total coalition compute fell from 0.03467 W to 0.00945 W (72.7% lower) and app CPU fell from 10.08% to 4.03% (60.0% lower).

The watt values are kernel coalition CPU/GPU/ANE energy attributable to Juice, not whole-system wall-outlet power. They are useful for comparing these two builds under the same workload but should not be presented as total Mac power draw.