…or consistency (#5555)

…5568)

Exposes a `BuildConfig.USE_SAGP` property from the recently introduced -PuseSagp flag ([#5538](#5538)).

Lets us update the SQLite screen in the Android sample app so that it swizzles between auto-instrumenting vs manually wrapping `SQLiteDriver`, depending on the whether SAGP was applied to the build.

…n standalone app start transaction (#5552)

* feat(android): Report app start reason on standalone app start transaction

Read ApplicationStartInfo.getReason() (API 35+) and attach it as
app.start.reason trace data on the standalone app.start transaction
in both the foreground and headless paths.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* feat(android): Register StandaloneAppStart SDK integration marker

Advertise that standalone app start tracing is active by adding a
StandaloneAppStart marker to the SDK metadata integrations when the
feature is enabled. Internal SDK metadata only.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* docs(changelog): Note app.start.reason is searchable in Trace Explorer

Address review feedback to mention that customers can search and group
by the app.vitals.start.reason attribute.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>

* fix(replay): Release MediaMuxer when no frames are encoded

The MediaMuxer is created when the video encoder is constructed, but its
release() was reachable only on the happy path. Two cases leaked it:

- createVideoOf returned early when frameCount was 0 without releasing the
  encoder.
- SimpleMp4FrameMuxer.release() called muxer.stop() before muxer.release().
  stop() throws if the muxer was never started (no frame ever muxed), so
  release() was skipped.

This surfaced as a CloseGuard "resource was acquired but never released"
warning. Guard stop() behind the started flag so release() is always
reached, and release the encoder on the no-frames return path.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* changelog

---------

Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>

…5586)

* docs(replay): Add THIRD_PARTY_NOTICES entry for SimpleMp4FrameMuxer

SimpleMp4FrameMuxer is adapted from the flutter_screen_recorder library
and carries a complete attribution header, but the corresponding entry
in THIRD_PARTY_NOTICES.md was never added. Warden's check-code-attribution
flags the missing entry as independently required regardless of header
completeness.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* docs(replay): Cover all adapted flutter_screen_recorder and Curtains files

SimpleFrameMuxer and SimpleVideoEncoder are adapted from the same
flutter_screen_recorder library as SimpleMp4FrameMuxer; fold all three
into one notice entry. Also extend the existing Square Curtains scope to
list io.sentry.android.replay.Windows, which is adapted from Curtains but
was not mentioned.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* fix(replay): Correct adapted-from URL in SimpleVideoEncoder header

The attribution header pointed at the upstream SimpleFrameMuxer.kt
instead of SimpleVideoEncoder.kt.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>

… tasks (#5522)

* build(samples): Remove outputs.upToDateWhen { false } from systemTest tasks

The systemTest tasks in the sample modules forced Gradle to always treat
their outputs as out of date, disabling up-to-date checks and build cache
reuse. Removing this lets Gradle rely on its normal input/output tracking
for the Test tasks.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* build(samples): Track systemTest app archive via convention plugin

The system tests launch the packaged sample (war/shadowJar/bootJar) from
build/libs as a separate process, so the archive is a real input to the
systemTest task even though it is not on the test classpath. Without it,
removing outputs.upToDateWhen { false } would let Gradle mark systemTest
up-to-date while a separate jar build refreshed the artifact, skipping
verification against the rebuilt sample.

Move that wiring into a single io.sentry.systemtest convention plugin in
build-logic instead of repeating it in every sample build file. The
plugin auto-detects the packaging task (war, else shadowJar, else
bootJar), mirroring the selection in test/system-test-runner.py, and
declares its archive as an input and dependency. Each sample just
applies the plugin.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* build(samples): Track OpenTelemetry agent jar as systemTest input

The agent-based OpenTelemetry samples are launched by the runner with
-javaagent:<sentry-opentelemetry-agent>, started outside the test JVM.
That jar is not on the test classpath nor one of the app archives, so
without tracking it systemTest could stay up-to-date and be skipped
while the runner launches a newer agent.

Add a usesOpenTelemetryAgent opt-in to the io.sentry.systemtest plugin;
the three agent samples enable it and the agent jar is then tracked as a
content input. The runner already builds and launches the agent before
invoking the task, so it is tracked by path without a cross-project task
dependency, which keeps it configuration-on-demand and configuration
cache compatible.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>

…ed from a wrong process (#5597)

* fix(android): Fix crash when getHistoricalProcessStartReasons is called from a wrong process

* test(android): Add test and changelog for getHistoricalProcessStartReasons crash fix

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* SecurityException -> RuntimeException

---------

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>

The `when sentry-unmask modifier is set unmasks the node` test intermittently
failed because Robolectric can report zero bounds for some nodes when running
the full test class, making them invisible (shouldMask = isVisible && ...).

Restructure the test to:
- Explicitly find the "Make Request" node and assert it IS visible and unmasked
- Assert other visible nodes remain masked, with a guard against empty iteration
- Tolerate intermittent zero-bounds on non-identifiable nodes (Robolectric artifact)

Validated with the repro from getsentry/repro#51: 20/20 passes (vs ~10% flake
rate before the fix).

Fixes #5585

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>

Fork PRs don't have access to the SENTRY_AUTH_TOKEN secret, so the
sentry-cli snapshot upload would fail anyway. Guard the step to run
only on pushes and same-repo PRs.

Co-authored-by: Claude Opus 4.8 <[email protected]>

Fork PRs don't have access to SENTRY_AUTH_TOKEN, so the upload step
would attempt to run without credentials. Guard it the same way the
replay snapshot upload is guarded so fork PRs cleanly skip it.

)

* fix: use System.nanoTime() for cron check-in duration measurement

System.currentTimeMillis() is a wall-clock value and is subject to
NTP adjustments and DST transitions. For long-running cron jobs this
can produce incorrect or even negative durations in the check-in payload.

Switch the start/end capture in CheckInUtils.withCheckIn() and the three
SentryCheckInAdvice implementations (sentry-spring, sentry-spring-jakarta,
sentry-spring-7) to System.nanoTime(), which is guaranteed monotonic.
Use DateUtils.nanosToSeconds() (already present) to convert the delta.

Fixes #5579

* changelog

---------

Co-authored-by: Roman Zavarnitsyn <[email protected]>

Co-authored-by: GitHub <[email protected]>