fix(network-activity-plugin): align Nitro timestamps with wall time

[lorenc-tomasz]

Description

Fix incorrect start times and excessively stretched timelines for Nitro HTTP and WebSocket traffic in Network Activity.

Calculate the timestamp origin using Date.now() - performance.now() so Nitro events align with built-in network events.

Related Issue

No related issue.

Context

On affected React Native iOS runtimes, performance.timeOrigin and performance.now() can use different clock bases. The adapter previously trusted timeOrigin whenever it was available, causing Nitro timestamps to shift relative to wall-clock time.

The fix derives the offset from the clock Nitro actually uses. Event timestamps remain in epoch milliseconds, and elapsed durations are preserved.

Includes regression coverage for shifted, aligned, and missing timeOrigin values across HTTP requests, cancellations, and WebSocket events, plus a patch changeset.

Testing

• pnpm checks:affected — passed.
• pnpm test:affected — 268 tests passed.
• pnpm release:plan — passed.
• Repeated 25 adapter, formatting, and timeline tests in each of six time zones: UTC, Europe/Warsaw, America/New_York, Asia/Kolkata, Asia/Kathmandu, and Pacific/Kiritimati.
• Verified clock formatting across daylight-saving transitions and date boundaries; elapsed timeline durations remained correct.
• Confirmed the regression tests fail before the fix and pass afterward.

Manual:

• Verified on Android and iOS in our React Native application using the equivalent patch-package fix.
• Verified Nitro GET and POST requests in the Rozenite playground on an iPhone 16 Pro Max simulator running iOS 18.6, 26.1 and 27.0. Network Activity displayed the correct local start times and request durations.

[V3RON]

Thanks, the diagnosis holds. In React Native, NativePerformance::now() on Apple uses mach_absolute_time(), which pauses during sleep, while timeOrigin() subtracts libc++'s steady_clock (CLOCK_MONOTONIC_RAW, which counts sleep). So timeOrigin + now() lags wall time by the device's total sleep since boot. Android uses CLOCK_MONOTONIC for both, so it doesn't have this particular mismatch.

I pushed one follow-up commit: the offset is now sampled on every conversion instead of once at module load. The monotonic clock pauses during sleep on both iOS and Android, so a cached offset drifts after the device locks mid-session. The commit also adds a regression test for that case and updates the stale timeOrigin comment in recorder.ts.

[lorenc-tomasz]

Thanks :)