feat(push): capture a launch intent the SDK was installed too late to read

[turnipdabeets]

💡 Motivation and Context

The Android half of PostHog/posthog-flutter#558 — $push_notification_opened is never captured in a Flutter app.

The tray intent is read in onActivityCreated, and install() only calls registerActivityLifecycleCallbacks — it never seeds from an Activity that already exists. A host that configures the SDK from its own runtime installs too late to see that callback. Measured in the Flutter example app:

17:54:27.516  MainActivity.onCreate
17:54:27.939  onCreate END / onStart / onResume   ← all three complete
17:54:28.582  first PostHog event                 ← SDK installs ~640ms later

Every lifecycle callback for the launch Activity is missed. Native apps that call setup() from Application.onCreate are unaffected — this only bites hosts that initialise late, which today means Flutter and React Native.

Unlike iOS, no buffering is needed: the tap is durable state on the Intent, so the fix is to read it late rather than to hold it.

Changes

• PostHogAndroid.capturePushNotificationOpened(intent) — one new public entry point for late-installing hosts. Shares the extraction and google.message_id dedupe with the automatic path, so calling both cannot double-count.
• On that path only, recently opened ids are also persisted (pushOpenedMessageIds, a bounded history). A caller with no savedInstanceState cannot tell a process-kill restore — which hands the Activity back its original intent — from a real second tap. Read and write are both scoped to the new path, so a pure-native app's stored state is byte-identical to today.
• onActivityCreated keeps savedInstanceState as its restore gate and never touches the persisted id: it is the strictly better signal, because it separates a restore from a genuine re-tap.

💚 How did you test it?

• ./gradlew :posthog-android:testDebugUnitTest :posthog:test — BUILD SUCCESSFUL. spotlessCheck, :posthog:apiCheck, :posthog-android:apiCheck clean.
• Ten new tests. Three were verified to fail with their fix reverted — the ones guarding the silent regressions below: the ALL_INTERNAL_KEYS entry, the automatic path still captures a genuine re-tap in a new process, and with after setup must not redirect the manual entry to the secondary project.
• On device (Pixel 9 emulator, Flutter example app against a local build): cold start with a google.message_id extra → 1 capture (was 0); same id in a new process → 0; a new id → 1; no push extra → 0.

Testing

Beyond the unit tests, verified on a Pixel 9 emulator through two hosts:

posthog-android's own sample (setup() in Application.onCreate, plus the new onNewIntent forwarding this PR adds):

Scenario	Result
Cold launch from a tap	captured — also on unmodified origin/main, confirming no regression
Tap while running, forwarded from onNewIntent	captured
Same message id again	not captured
Fresh launch with a previously seen id	captured — a real second tap, not a restore
Launch with no push payload	not captured

A Flutter host (PostHog/posthog-flutter#557), where the SDK installs after the launch Activity has already resumed: cold and warm both captured, with the dedupe holding across process death.

Warm delivery needs android:launchMode="singleTop" — without it the system resumes the task instead of delivering the intent, which is why the sample now declares it.

📝 Checklist

• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran pnpm changeset to generate a changeset file

The changeset declares posthog as well as posthog-android — the core ALL_INTERNAL_KEYS entry is load-bearing (see below) and would not otherwise be released.

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Built with Claude Code (session), driven by @turnipdabeets. Root cause measured on an emulator before any code was written; a first attempt that hooked onActivityStarted/onActivityResumed was implemented, tested, and discarded once the ordering above was measured.

Six review rounds ran over this branch. Three findings are worth a reviewer's attention because each was a silent regression caught only by an executed test:

• The new preferences key was initially missing from ALL_INTERNAL_KEYS, so it rode on every event as a super property. getAll() feeds buildProperties(); every other internal key is listed.
• Persisting the dedupe id on the automatic path suppressed a genuine second tap of the same notification after a process death — demonstrated 2 → 1 captures. Hence the read/write scoping.
• An earlier "mark after delivery" guard was dead code (?: return on a Unit function), and the two synchronized blocks it required reopened the race the lock existed to close — two callers both delivered for one id.

A warm-start tap arrives through Activity.onNewIntent, which ActivityLifecycleCallbacks does not expose — so a native host forwards it with one line, which the sample now demonstrates. PostHog/posthog-flutter#557 does it automatically for Flutter apps.

Consumer: PostHog/posthog-flutter#557 (which closes PostHog/posthog-flutter#558) raises its floor to [3.62.0,4.0.0) and is held until this releases.

🤖 Generated with Claude Code

https://claude.ai/code/session_012txiHBCZRkShMdE7V25Jrd

[github-actions]

posthog-android Compliance Report

Date: 2026-09-02 16:09:07 UTC
Duration: 118462ms

✅ All Tests Passed!

46/46 tests passed

---

Capture Tests

✅ 29/29 tests passed

View Details

Test	Status	Duration
Format Validation.Event Has Required Fields	✅	378ms
Format Validation.Event Has Uuid	✅	33ms
Format Validation.Event Has Lib Properties	✅	29ms
Format Validation.Distinct Id Is String	✅	27ms
Format Validation.Token Is Present	✅	29ms
Format Validation.Custom Properties Preserved	✅	28ms
Format Validation.Event Has Timestamp	✅	23ms
Retry Behavior.Retries On 503	✅	7025ms
Retry Behavior.Does Not Retry On 400	✅	4024ms
Retry Behavior.Does Not Retry On 401	✅	4029ms
Retry Behavior.Respects Retry After Header	✅	7028ms
Retry Behavior.Implements Backoff	✅	17035ms
Retry Behavior.Retries On 500	✅	7019ms
Retry Behavior.Retries On 502	✅	7020ms
Retry Behavior.Retries On 504	✅	7021ms
Retry Behavior.Max Retries Respected	✅	17035ms
Deduplication.Generates Unique Uuids	✅	43ms
Deduplication.Preserves Uuid On Retry	✅	7018ms
Deduplication.Preserves Uuid And Timestamp On Retry	✅	12025ms
Deduplication.Preserves Uuid And Timestamp On Batch Retry	✅	7018ms
Deduplication.No Duplicate Events In Batch	✅	37ms
Deduplication.Different Events Have Different Uuids	✅	23ms
Compression.Sends Gzip When Enabled	✅	23ms
Batch Format.Uses Proper Batch Structure	✅	18ms
Batch Format.Flush With No Events Sends Nothing	✅	12ms
Batch Format.Multiple Events Batched Together	✅	32ms
Error Handling.Does Not Retry On 403	✅	4021ms
Error Handling.Does Not Retry On 413	✅	4022ms
Error Handling.Retries On 408	✅	5025ms

Feature_Flags Tests

✅ 17/17 tests passed

View Details

Test	Status	Duration
Request Payload.Request With Person Properties Device Id	✅	37ms
Request Payload.Flags Request Uses V2 Query Param	✅	25ms
Request Payload.Flags Request Hits Flags Path Not Decide	✅	24ms
Request Payload.Flags Request Omits Authorization Header	✅	31ms
Request Payload.Token In Flags Body Matches Init	✅	26ms
Request Payload.Groups Round Trip	✅	39ms
Request Payload.Groups Default To Empty Object	✅	25ms
Request Payload.Disable Geoip False Propagates As Geoip Disable False	✅	24ms
Request Payload.Disable Geoip Omitted Defaults To False	✅	23ms
Request Payload.Flag Keys To Evaluate Contains Only Requested Key	✅	27ms
Request Lifecycle.No Flags Request On Init Alone	✅	14ms
Request Lifecycle.No Flags Request On Normal Capture	✅	21ms
Request Lifecycle.Two Flag Calls Produce Two Remote Requests	✅	49ms
Request Lifecycle.Mock Response Value Is Returned To Caller	✅	22ms
Retry Behavior.Retries Flags On 502	✅	323ms
Retry Behavior.Retries Flags On 504	✅	326ms
Side Effect Events.Get Feature Flag Captures Feature Flag Called Event	✅	21ms