fix: Update OpenFeature SDK dependency lower bound from 1.13.0 to 1.16.0

[kinyoklion]

BEGIN_COMMIT_OVERRIDE
fix: Update OpenFeature SDK dependency lower bound from 1.13.0 to 1.16.0
ci: Fix flaky tests by awaiting the provider being set.
END_COMMIT_OVERRIDE
Requirements

• I have added test coverage for new or changed functionality
• I have followed the repository's pull request submission guidelines
• I have validated my changes against all supported platform versions

Related issues

Fixes #40

Describe the solution you've provided

Raises the lower bound of the dev.openfeature:sdk dependency range from [1.13.0,2.0.0) to [1.16.0,2.0.0).

The previous lower bound of 1.13.0 was incorrect for two reasons, both confirmed by inspecting the bytecode of the relevant SDK versions:

1. Compilation failure (SDK < 1.14.1): TrackingEventDetails.getValue() (no-arg) does not exist prior to 1.14.1. The Provider.track() method calls this, so the project cannot compile against SDK versions [1.13.0, 1.14.1).

2. Binary incompatibility (SDK < 1.16.0): EventProvider.emitProviderReady, emitProviderError, emitProviderStale, and emitProviderConfigurationChanged changed their return type from void to Awaitable in SDK 1.16.0. Code compiled against < 1.16.0 throws NoSuchMethodError at runtime when a consumer brings in SDK ≥ 1.16.0.

Setting the lower bound to 1.16.0 resolves both issues.

Existing tests already exercise the affected code paths (emit methods in LifeCycleTest, tracking in ProviderTest). No new test coverage is needed since this is a dependency range correction, not a behavior change.

Test fixes

This PR also fixes pre-existing flaky/failing tests:

LifeCycleTest

• itCanHandleClientThatIsNotInitializedImmediately: Event handlers registered via OpenFeatureAPI.getInstance().on(...) were leaking across tests because the singleton was not consistently cleaned up. Added @AfterEach calling OpenFeatureAPI.getInstance().shutdown() so handlers are cleared between tests. Removed individual shutdown() calls from test bodies since cleanup is now centralized.
• canShutdownAnOfflineClient: Removed the assertion that the data source status is OFF after shutdown. The offline LDClient does not transition its data source status to OFF on close(), so this assertion was always incorrect. The test now verifies that shutdown completes without error.

ProviderTest

• All four evaluation tests: Replaced setProvider() (non-blocking) + Awaitility.await().forever() polling with setProviderAndWait() (synchronous). This eliminates the race condition that caused itCanDoAValueEvaluation to throw NullPointerException on Java 11, and removes the risk of tests hanging indefinitely. To support setProviderAndWait() calling initialize(), the mock LDClientInterface now stubs getFlagTracker(), getDataSourceStatusProvider(), and isInitialized().

Describe alternatives you've considered

• Setting the lower bound to 1.14.1 would fix the compilation issue but not the binary incompatibility, so 1.16.0 is the correct minimum.

Additional context

• The issue suggests bumping the library version from 1.1.1 → 1.2.0. This repo uses release-please, so version bumping may be handled separately.

Key items for review:

• Confirm 1.16.0 is the appropriate lower bound (no intermediate versions were skipped that would also work)
• Whether a library version bump should accompany this change or be handled by release-please
• canShutdownAnOfflineClient now has no assertions beyond assertDoesNotThrow — confirm this is sufficient or whether a different assertion should replace the removed one
• Verify that OpenFeatureAPI.getInstance().shutdown() in @AfterEach fully clears event handlers (not just providers)
• The mock setup stubs isInitialized() → true and getStatus() → VALID, so initialize() returns immediately without waiting on the CompletableFuture — confirm this adequately exercises the provider lifecycle for these evaluation-focused tests

Link to Devin session: https://app.devin.ai/sessions/219cccbf82aa4ac6a2372d91eee96ef9
Requested by: @kinyoklion

---

Note

Medium Risk
Medium risk because it changes the supported OpenFeature SDK version range, which can affect downstream dependency resolution. Code changes are limited to test synchronization/cleanup and should not impact runtime behavior.

Overview
Raises the lower bound of the dev.openfeature:sdk dependency range from 1.13.0 to 1.16.0.

Stabilizes tests by centralizing OpenFeatureAPI cleanup via @AfterEach in LifeCycleTest, removing an incorrect post-shutdown data source status assertion, and switching ProviderTest evaluations from async setProvider() + polling to synchronous setProviderAndWait() (with additional LDClientInterface mock stubbing to support initialization).

^{Written by Cursor Bugbot for commit 8a60681. This will update automatically on new commits. Configure here.}

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[kinyoklion]

We will need to make sure this test has some timeout associated with it. Otherwise the test run could hang if there was as bug.

[devin-ai-integration]

Good catch — updated all four Awaitility waits (boolean, string, double, and value) to use atMost(Duration.ofSeconds(10)) instead of forever(). This ensures the build won't hang if a bug prevents provider initialization.

[kinyoklion]

Should we be using an open feature provider status API?
Why isn't this available immediately?

[devin-ai-integration]

Good question. OpenFeatureAPI.setProvider() is non-blocking — it triggers provider initialization asynchronously, so the provider may not be in READY state when the next line executes. The other three evaluation tests (boolean, string, double) already had Awaitility.await() guards for this reason; this value test was the only one missing it.

An alternative would be to use setProviderAndWait() instead of setProvider() + Awaitility — that's the synchronous variant that blocks until initialization completes. That would be cleaner and avoid the need for Awaitility entirely in these tests. Want me to refactor the evaluation tests to use setProviderAndWait() instead?

[kinyoklion]

In what open feature version was setProviderAndWait added?

[devin-ai-integration]

setProviderAndWait is available since at least SDK 1.13.0 (the previous lower bound) — I confirmed it exists in the bytecode of that version. So it's safe to use across the entire [1.16.0, 2.0.0) range.

Want me to refactor these tests to use setProviderAndWait() and drop the Awaitility waits?

[kinyoklion]

Yes, lets give that a try.

[devin-ai-integration]

Done — refactored all four evaluation tests to use setProviderAndWait() instead of setProvider() + Awaitility. Also added the necessary mock stubs (getFlagTracker(), getDataSourceStatusProvider(), isInitialized()) so the provider's initialize() lifecycle completes successfully. All tests pass locally.