fix(flags): dedupe exposures by latest assignment

[leoromanovsky]

Motivation

Android flags exposure deduplication currently treats every previously seen targetingKey + flagName + allocationKey + variationKey tuple as permanently sent. That suppresses valid exposure events when the same subject and flag cycles assignments, for example A -> B -> A, because the final A was seen earlier.

Related iOS PR: DataDog/dd-sdk-ios#2987 applies the same mobile exposure-cache contract.

Changes

• Cache exposures by targetingKey + flagName.
• Store the latest allocationKey + variationKey as the cache value.
• Emit an exposure when the subject/flag has no cached value or when the latest assignment differs from the cached value.
• Keep a bounded Android LruCache, but make it entry-count based with a 5,000-entry limit instead of using approximate object-size accounting.
• Add a comment documenting the expected high-water mark of two subjects and 2,500 flags each; normal flag keys are typically tens of characters, so count-based sizing is easier to reason about.
• Add a regression test proving A -> B -> A writes three exposure events.
• Update the existing mixed scenario to expect a cycle-back exposure.

Decisions

• Kept the change internal to the processor with no public API changes.
• Preserved an LRU bound for apps with many flags moving through a client process, while sizing it directly around the expected two-subject, 2,500-flag high-water mark.
• Avoided approximate byte accounting in sizeOf; AndroidX LruCache defaults to one unit per entry when sizeOf is not overridden.
• Matched the iOS PR's count-based mobile policy while using Android's platform cache primitive.
• Kept synchronization around cache check/update so duplicate suppression remains atomic.
• Validation: focused regression test failed before the implementation, then passed after the fix. JAVA_HOME=/opt/homebrew/Cellar/openjdk@17/17.0.19/libexec/openjdk.jdk/Contents/Home ANDROID_HOME=/opt/homebrew/share/android-commandlinetools ./gradlew :features:dd-sdk-android-flags:testDebugUnitTest passed.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.48%. Comparing base (c6e8fe9) to head (919fbba).

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #3526      +/-   ##
===========================================
- Coverage    72.52%   72.48%   -0.04%     
===========================================
  Files          978      978              
  Lines        36052    36037      -15     
  Branches      6038     6039       +1     
===========================================
- Hits         26146    26119      -27     
- Misses        8234     8235       +1     
- Partials      1672     1683      +11     

Files with missing lines	Coverage Δ
.../android/flags/internal/ExposureEventsProcessor.kt	100.00% <100.00%> (ø)

... and 42 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dd-octo-sts-6cbbf8]

🐑 PR Shepherd is maintaining this PR

I watch your PR and automatically fix CI failures, rebase your branch, handle flaky tests, and push it to the merge queue when it's ready.

More about what I do → Guide

To pause me on this PR, add the flow-skip label.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bed9e6ad17

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep cache updates ordered with exposure writes

When the same subject/flag is evaluated concurrently with different assignments, this cache update can happen before the previous exposure is actually written. For example, thread A caches A and pauses, thread B caches and writes B, then thread A writes A; the cache still says B, so the next B exposure is suppressed even though the latest emitted assignment was A. The new latest-assignment contract only holds if updating the cache and emitting the exposure remain ordered for a given cache key.

Useful? React with 👍 / 👎.

[dd-octo-sts]

👋 This PR is now being monitored by PR Shepherd.

It watches CI and the merge queue and helps the PR move toward merge — rebasing onto the base branch when it falls behind and re-queueing after a transient merge-queue failure.

Read more about PR shepherd hereNeed a hand? Reach the team in #ai-devx-flow. To opt out, add the flow-skip label.

[mikewoo-dd]

Please disregard comment from PR shepherd above. We're doing a dry run test and would not make any change to the PR.