Fix pipelines and tests

[bmehta001]

• Modernize all GitHub Actions workflows to use latest action versions and fix pre-existing bugs across iOS, Android, and Windows that cause CI failures, crashes, data races, and flaky tests.

• Several CI pipelines were failing or logging extensive warnings due to:

  • Deprecated GitHub Actions (checkout v1/v2, setup-java v3, setup-msbuild v1.1) running on EOL Node.js 16
  • Removed macos-13 runner images (deprecated Dec 2024)
  • Hardcoded paths that break on newer runner images
  • Pre-existing bugs in iOS HTTP client, threading, and test infrastructure

  Changes

1. Modernize GitHub Actions (11 workflow files)

• Updated actions/checkout from v1/v2/v3 to v4
• Updated actions/setup-java from v3 to v5
• Updated android-actions/setup-android from v2 to v3
• Updated microsoft/setup-msbuild from v1.1 to v2
• Updated actions/upload-artifact from v2 to v4

2. Pipeline improvements

• Add concurrency groups to cancel superseded runs
• Replace deprecated macos-13 with macos-14 and update simulator matrix
• Remove hardcoded C:\Users\runneradmin\ → use $Env:USERPROFILE
• Remove hardcoded cmdline-tools\7.0\bin (setup-android v3 adds sdkmanager to PATH)
• Update vs-version from [16,) to [17,) for VS 2022

3. Fix iOS CI: Simulator UUID Targeting (build-tests-ios.sh)

xcodebuild on macos-14 lists duplicate simulator entries, causing it to pick the "Any iOS Device" placeholder instead of a concrete simulator (actions/runner-images#8621). Fix by resolving the simulator UUID via xcrun simctl and using -destination "id=".

4. Fix iOS HTTP Client Crashes and Leaks (5 files)

• Use-after-destroy crash or deadlocks: Replace shared NSURLSession with per-instance ephemeral sessions with safe teardown via invalidateAndCancel.
  • Unlikely, aside from tests, for the SDK to be torn down and then be created again, but the ephemeral session over the "dispatch once" session can also potentially help to recover from zombie sessions, since we will re-create a new session.
• Session invalidation: Invalidated in the destructor. Originally, HttpRequestApple::Cancel() called [session getTasksWithCompletionHandler:] to cancel every task on the shared static NSURLSession. The completion handler callback fires asynchronously on an internal dispatch queue, so when one SDK session tears down and a new session starts sending requests, the stale callback arrives and cancels the new session's in-flight tasks. This caused receivedRequests.size() == 0 in the second block of the restart test. The fix removes the blanket cancel and keeps only [m_dataTask cancel], which cancels the specific task.
• Response body leak: Add delete before setting httpResponse to nullptr in HttpResponseDecoder. The Abort outcome path set ctx->httpResponse = nullptr without delete, leaking the SimpleHttpResponse allocated by the HTTP client. Added the missing delete like RetryNetwork already had

FlushAndTeardown() remains required before shutdown, as documented.

5. Fix Data Race and Memory Leak (2 files)

• TransmissionPolicyManager.cpp: Move m_scheduledUploadTime write inside LOCKGUARD to decrease chance of concurrent read/write and made m_runningLatency atomic for same reason (to avoid adding additional uses of LOCKGUARD, especially ones that could cause deadlocks if we make them too broad)
• Added m_httpCallbacksMtx around .empty() check in HttpClientManager: cancelAllRequests() which reads m_httpCallbacks.empty(), since onHttpResponse() removes entries under that mutex on another thread.
• WorkerThread.cpp: Clean up remaining task queues on shutdown instead of just logging warnings.

6. Fix Static-Destruction-Order Crash (Logger.cpp)

Remove LOG_TRACE from ~Logger() to prevent crash when logging subsystem is destroyed before Logger instances during static destruction. Order of namespace destruction can change depending on system, but the original implementation was causing issues for the simulator (where debugLogMutex may be destroyed before the Logger instance). If this is an issue, I can revert it, as I am not sure if it would be an actual issue in a real iOS deployment scenario.

7. Fix Android Build Issues (3 files)

• HttpClientFactory.cpp: Use standard __ANDROID__ macro instead of non-standard ANDROID
• HttpClient_Android.cpp: Mark unused log tag with attribute((unused))
• config-default.h: Add #ifndef guards to prevent duplicate macro definitions

8. Fix Flaky Tests (5 files)

• Replace localhost with 127.0.0.1 in functional tests to avoid IPv6 resolution failures
• Add synchronization loop for storage-full callback in APITest to fix assertion race
• Increase timing tolerances in PalTests, BasicFuncTests, and MultipleLogManagerTests for slower CI environments

Testing

All changes have been validated through:

• Local Windows builds (VS 2022), WSL, Android Simulator on WSL, macOS, iOS Simulator on macOS