perf(core): Skip redundant setBreadcrumbs call on the addBreadcrumb path

[runningcode]

Resolves JAVA-606.

Description

Scope.addBreadcrumb notified every scope observer twice per breadcrumb: observer.addBreadcrumb(crumb) followed by observer.setBreadcrumbs(all). For PersistingScopeObserver the second call is a guaranteed no-op right after an add — it only does work when the collection is empty (the clear case) — yet it still acquires the SynchronizedQueue lock via isEmpty() and re-iterates the observer list on every breadcrumb.

This notifies observers once per add. clearBreadcrumbs still calls setBreadcrumbs.

⚠️ Reviewer note — public API contract

IScopeObserver is a public interface. In principle a third-party observer could implement only setBreadcrumbs and rely on it being called on every add, rather than the incremental addBreadcrumb. Removing the per-add setBreadcrumbs call changes the observer notification contract on the add path.

This PR treats that as an intentional contract clarification (observers must implement addBreadcrumb for incremental updates). Flagging for a maintainer decision — if we'd rather preserve the old contract, close this in favor of documentation only. Kept as draft for that reason.

Source

Found via on-device Perfetto method-trace analysis of the Android SDK breadcrumb path. Part of Reduce SDK init time [Android].

🤖 Generated with Claude Code

Benchmark

Measured on-device with androidx Microbenchmark 1.4.1 (benchmark-junit4) on a Samsung Galaxy A55 (SM-A556B), Android API 36 (release build, non-minified).

The benchmark exercises the caller-side cost of a scope mutation with scope persistence enabled (enableScopePersistence = true, cacheDirPath set, no beforeBreadcrumb). The async serialize + disk write is dispatched to a no-op executor so it never runs on the measured thread — mirroring production, where it runs off-thread. Each branch is compared against its base commit (aab952b82e); the per-branch delta isolates this change. allocationCount is deterministic; timeNs has ~±10 ns run-to-run noise on a non-rooted device (unlocked clocks), so it is reported as a median and reproduced across 2 runs.

How it was done

A local, uncommitted androidx microbenchmark module (sentry-android-integration-tests/sentry-uitest-android-microbenchmark) depending on :sentry, with a BenchmarkRule looping the scope mutation. Run with:

./gradlew :sentry-android-integration-tests:sentry-uitest-android-microbenchmark:connectedReleaseAndroidTest \
  -Pandroid.testInstrumentationRunnerArguments.androidx.benchmark.suppressErrors=UNLOCKED

Result — scope.addBreadcrumb()

Metric	main	This PR	Delta
allocationCount	7	7	none (expected)
timeNs (median)	~242 ns	229 ns	~-6%

As expected, dropping the redundant per-observer setBreadcrumbs call is a small time/lock win with no allocation change. The setTag control path was unchanged.

[linear-code]

JAVA-606

[sentry]

📲 Install Builds

Android

🔗 App Name	App ID	Version	Configuration
SDK Size	io.sentry.tests.size	8.47.0 (1)	release

⚙️ sentry-android Build Distribution Settings

[github-actions]

Performance metrics 🚀

	Plain	With Sentry	Diff
Startup time	315.27 ms	358.67 ms	43.40 ms
Size	0 B	0 B	0 B

Baseline results on branch: main

Startup times

Revision	Plain	With Sentry	Diff
abfcc92	304.04 ms	370.33 ms	66.29 ms
4e3e79d	365.83 ms	477.62 ms	111.79 ms
d8912da	329.94 ms	389.68 ms	59.74 ms
d15471f	294.13 ms	399.49 ms	105.36 ms
abf451a	332.82 ms	403.67 ms	70.85 ms
604a261	380.65 ms	451.27 ms	70.62 ms
806307f	357.85 ms	424.64 ms	66.79 ms
22f4345	307.87 ms	354.51 ms	46.64 ms
d217708	375.27 ms	415.68 ms	40.41 ms
c3ee041	310.64 ms	361.90 ms	51.26 ms

App size

Revision	Plain	With Sentry	Diff
abfcc92	1.58 MiB	2.13 MiB	557.31 KiB
4e3e79d	0 B	0 B	0 B
d8912da	0 B	0 B	0 B
d15471f	1.58 MiB	2.13 MiB	559.54 KiB
abf451a	1.58 MiB	2.20 MiB	635.29 KiB
604a261	1.58 MiB	2.10 MiB	533.42 KiB
806307f	1.58 MiB	2.10 MiB	533.42 KiB
22f4345	1.58 MiB	2.29 MiB	719.83 KiB
d217708	1.58 MiB	2.10 MiB	532.97 KiB
c3ee041	0 B	0 B	0 B