Skip to content

Sync the latest develop changes into develop-v2 - #1918

Open
PratimMallick wants to merge 39 commits into
develop-v2from
chore/sync-develop-into-develop-v2
Open

PratimMallick wants to merge 39 commits into
develop-v2from
chore/sync-develop-into-develop-v2

Conversation

@PratimMallick

@PratimMallick PratimMallick commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor

Goal

Closes AND-1999

Bring the fixes that landed on develop since the last sync onto develop-v2, and move v2 onto repackaged WebRTC 156.0.7.

Implementation

  • Merge develop into develop-v2. v2 keeps Kotlin 2.3.21, kotlin.time.Instant models, the io.getstream.android.video.api.generated package, and JSON query parameters.
  • WebRTC is io.getstream:stream-video-webrtc-android-repackaged:156.0.7. Types stay in io.getstream.webrtc, and the native library stays libstream_jingle_peerconnection_so. Noise cancellation stays on the non-libwebrtc artifact.
  • 156.0.7 ships its own consumer ProGuard rules for io.getstream.webrtc and org.jni_zero, so the SDK no longer keeps those packages.
  • Generated models use @JsonClass(generateAdapter = true) with Moshi KSP codegen. Core no longer depends on moshi-kotlin or KotlinJsonAdapterFactory, and the consumer rules no longer keep kotlin-reflect or the generated model packages.

Testing

  • Core, Compose, XML, and the demo app compiled against repackaged WebRTC 156.0.5. GeneratedModelCodegenTest passed (7 tests). Pre-push spotlessCheck and apiCheck passed on that commit.
  • The 156.0.7 classes.jar is byte-identical to 156.0.5, and the AAR still ships libstream_jingle_peerconnection_so. This pin changes the version and drops the keeps the AAR now provides.
Open in Web Open in Cursor 

aleksandar-apostolov and others added 30 commits October 9, 2026 08:10
Syncs signal.proto with GetStream/protocol#2116, which removed the deprecated
SendStatsRequest fields. Android only populated publisher_stats and
subscriber_stats; sendCallStats() no longer takes a CallStatsReport and the
collectStats() calls that only existed to fill it are gone.

Co-authored-by: Pratim Mallick <34009202+PratimMallick@users.noreply.github.com>
Bumps [actions/download-artifact](https://github.com/actions/download-artifact) from 8.0.1 to 8.0.2.
- [Release notes](https://github.com/actions/download-artifact/releases)
- [Commits](actions/download-artifact@v8.0.1...v8.0.2)

---
updated-dependencies:
- dependency-name: actions/download-artifact
  dependency-version: 8.0.2
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
A flapping network can keep a call escalating to REJOIN, and each REJOIN or MIGRATE issues a
new backend join. The loop's own attempt counter resets on every reconnect, so nothing bounded
how many joins a call could issue.

RejoinRateLimiter allows 10 REJOIN/MIGRATE attempts per rolling 120 seconds; FAST reconnects
do not count. Once the budget is spent the call leaves with rejoin_attempt_limit_exceeded.
A fresh join resets the budget. Attempts spaced further apart than the window are unaffected.
* temp

* Here’s a cleaner commit message:

Bump AudioSwitch and fix dropped Telecom actions

1. Bump AudioSwitch to 1.1.0, which includes Twilio 1.2.5 with BLE listener support.
2. Fix dropped Telecom actions by buffering the channel. Previously, the Bluetooth request could take up to 2 seconds, causing the call accept action to be dropped when both events arrived simultaneously.

* private val

* add ut

* remove ut
The release workflow now puts the Stream repository card in the release
body itself, so a repo-local release-notes-header job goes in the same change.
* fix: wait for delayed Telecom audio endpoints

* fix race-condition in starting jobs for requesting endpoint change and add UT

---------

Co-authored-by: Aleksandar Apostolov <apostolov.alexandar@gmail.com>
* improvement: update network call error logs

* improvement: remove sensitive info from logs

* improvement: log only whitelisted query param

* improvement: update tests

* revert changes

* update kdoc and test cases

* add try-catch around response read

* Throw the original Exception (as a new instance) and keep all its diagnostic data as it is
…erator (#1889)

The API key hardcoded in the ringing tutorial was revoked, and the
bundled user tokens were signed with its secret. Fetch the API key and
token from the demo token generator at login instead, so the tutorial
no longer hardcodes keys.
* fix(core): announce publisher mute from the track enabled bit

SetPublisher was marking video muted whenever the capturer was not LIVE yet, so web, iOS, and the recorder kept audio only after 1.31.

Co-authored-by: Cursor <cursoragent@cursor.com>

* test(core): restore device LIVE checks and pin LIVE on the disabled-track case

AndroidDeviceTest asserts that capture actually started, which the fix does not change. The disabled-track announce test now stubs LIVE so a return to muted = !LIVE fails it.

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>
* chore(ci): move the build-conventions pins to v1.1.2

* chore(ci): pin the build-conventions workflows by commit SHA

* chore(ci): move the build-conventions pins to v1.1.3

---------

Co-authored-by: Pratim Mallick <34009202+PratimMallick@users.noreply.github.com>
…1900)

Branch protection required each shard by name, so changing the shard
count meant editing protection in lockstep. The gate job needs every
shard and is the one check to require.

Co-authored-by: Pratim Mallick <34009202+PratimMallick@users.noreply.github.com>
* ci(e2e): build the apks from the Actions cache instead of S3

The S3 build cache cost the e2e build ~2 min a run in download overhead.
The build now uses Gradle's local cache, restored from the Actions cache
that scheduled and manual runs save. Pull requests only restore it.

* ci(e2e): keep one gradle cache snapshot per branch

Each scheduled or manual run saves a ~1 GB snapshot under its run id.
Delete the branch's older snapshots after saving so they don't fill the
repo's Actions cache.

* ci(e2e): keep older gradle caches when the save failed

actions/cache/save only warns on failure, so the cleanup deleted the
older snapshots and left the branch without a cache. Delete them only
when the new snapshot is listed.
Spreading the keys into remember made its key count depend on the caller. When a caller passed fewer keys than before, remember returned an old key as the factory and the cast failed. Pass the keys as one list instead. A factory that read Compose state through a delegate kept its first value, because no remember key changed. Build the factory inside derivedStateOf, so Compose state read in the factory builds a new one when it changes.
* Hide system ANR dialogs on the E2E emulators

* Save the emulator logcat from boot until the first E2E test

* Close a system ANR dialog left from the emulator boot

* Save the boot logcat before the E2E setup steps
)

* Use the release bot App instead of the bot token in the workflows

* Take the release bot App's client ID instead of its app ID

* Point at the latest release workflow changes

* Point the conventions workflows at the v1.2.0 release

* Drop the GITHUB_TOKEN write permissions the App token replaced
)

* build(deps): try stream-video-webrtc-android 156.0.2-SNAPSHOT

Resolves the snapshot from the Stream snapshot repo, limited to the WebRTC module.

* build(deps): bump WebRTC to 156.0.3-SNAPSHOT

156.0.2 packed Kotlin 2.5 into the AAR and failed to compile against Kotlin 1.9.

* build(deps): bump WebRTC to 156.0.4-SNAPSHOT

156.0.4 publishes org.webrtc without embedding Kotlin, so the SDK can compile and run e2e against the current snapshot.

* Update to stable version 156.0.5

* Updated to latest stable release of m156 and remove the proguard rules as 156.0.7 ships with those rules
* perf(core): narrow the R8 consumer keep rules

Broad Serializable, kotlin-reflect, and stream.video.sfu keeps stopped
R8 from shrinking host apps. Scope the first two, drop the SFU keep, and
trace SFU events with Wire toString() so the payload stays readable.

* perf(core): parse models with Moshi codegen instead of kotlin-reflect

Generated adapters replace KotlinJsonAdapterFactory, so the blanket generated-model keep and the kotlin-reflect runtime can leave the SDK.

* test(core): cover the SFU event trace branches

SonarCloud requires 80% coverage on new code, and the five traceEvent arms were untested.

* refactor(core): let Moshi load the generated model adapters

Moshi 1.15.1 already resolves @JsonClass adapters, so the extra factory only called an internal API. A test now fails if a generated model loses that annotation.
github-actions Bot and others added 9 commits November 17, 2026 15:49
…1905)

* fix(core): stop noise-cancellation policy acting on a stale settings/capabilities pair

combine collects each upstream on its own coroutine, so it could pair new
settings with the capabilities from before them and withdraw noise
cancellation the server had just granted. Decide on the latest state instead.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(core): read noise-cancellation policy inputs as one snapshot

A response writes own capabilities and settings one after the other, so a
policy decision could still see one updated and the other not, e.g. latch
auto-on against the old settings. Responses that write both now hold a lock
the policy reads take too.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(core): publish noise-cancellation inputs after both fields are written

The lock is reentrant, so a collector resumed on the writer thread still
saw capabilities updated and settings not. Publish one snapshot after
both writes and decide from that.

---------

Co-authored-by: Cursor <cursoragent@cursor.com>
* refactor(core): add EmptyRtcSession, an RtcSession without an SFU connection

RtcSession becomes open and gains an internal initializeRtc flag. When the
flag is false, the session never creates the SFU connection or any peer
connection and leaves the shared peer connection factory untouched.
EmptyRtcSession uses it and turns every public member into a no-op.

cleanup() no longer assumes the SFU connection module exists.

Prepares for joining a call without WebRTC media (VID-1543). No behaviour
change for existing sessions.

Co-authored-by: Cursor <cursoragent@cursor.com>

* refactor(core): log ignored calls on EmptyRtcSession

Co-authored-by: Cursor <cursoragent@cursor.com>

* refactor(core): reject EmptyRtcSession as the active session

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>
Bring the fixes landed on develop since the last sync onto develop-v2. Keep v2's Kotlin 2.3, repackaged WebRTC 145.19.0, Instant models, and Moshi reflection.
The repackaged 145 artifact kept a blanket model keep and kotlin-reflect in the consumer rules. 156.0.7 publishes org.webrtc with its own ProGuard rules, and @JsonClass adapters replace reflection so R8 can shrink the model package.
v2 must stay on stream-video-webrtc-android-repackaged so WebRTC types remain in io.getstream.webrtc. 156.0.7 is only published as the org.webrtc artifact; 156.0.5 is the newest repackaged release.
156.0.7 is published as stream-video-webrtc-android-repackaged and keeps the io.getstream.webrtc package. Its classes match 156.0.5, and the AAR now ships consumer ProGuard rules, so the SDK drops the blanket webrtc and jni_zero keeps.

Co-authored-by: Pratim Mallick <PratimMallick@users.noreply.github.com>
@PratimMallick
PratimMallick requested a review from a team as a code owner November 20, 2026 16:18
@PratimMallick PratimMallick added the pr:internal Internal or infra-only changes label Nov 20, 2026 — with Cursor
@github-actions

github-actions Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor

PR checklist ✅

All required conditions are satisfied:

  • Title length is OK (or ignored by label).
  • At least one pr: label exists.
  • Sections ### Goal, ### Implementation, and ### Testing are filled, or the PR is bot-authored.
  • An issue is linked (Linear ticket or GitHub issue), or the PR is bot-authored.

🎉 Great job! This PR is ready for review.

@github-actions

Copy link
Copy Markdown
Contributor

SDK Size Comparison 📏

SDK Before After Difference Status
stream-video-android-core 12.38 MB 13.25 MB 0.88 MB 🔴
stream-video-android-ui-xml 5.65 MB 5.72 MB 0.07 MB 🟢
stream-video-android-ui-compose 6.19 MB 3.16 MB -3.03 MB 🚀

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

pr:internal Internal or infra-only changes

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants