[turnipdabeets]

💡 Motivation and Context

In screenshot replay, a ph-no-capture view masked only its own rect and the mask walk stopped there. A child of an unclipped container (clipChildren = false) can draw outside that rect, and that overflow was recorded unmasked, even overflowing text that default maskAllTextInputs would have masked without the tag. Reported in #908 review. It predates #908, but #908 makes screenshots the only mode, so it should land on main first.

The fix follows session-replay-privacy Behavior 4: the wrapper's rect is masked along with its descendants' rects.

• A ph-no-capture view, and every visible descendant, adds its rect, and the walk keeps descending.
• A ComposeView inside the subtree is masked whole instead of running the semantics pass. So a Compose postHogUnmask node that overflows a tagged View container stays masked. This was already the case inside the container's bounds, and it errs toward privacy.
• ph-no-mask on a View descendant still takes precedence, as before.
• Descendants are now checked like any other view, so a child that is animating or mid-layout makes the verified walk discard that frame. Untagged screens already do this.

💚 How did you test it?

• New Robolectric test no-capture container masks descendants that overflow its bounds. A tagged, unclipped wrapper has overflowing text, plain, Compose and ph-no-mask children. The test asserts that the wrapper and the first three are masked and the ph-no-mask child isn't. It also checks that the draw-time verification walk isn't misaligned, and keeps an untagged control where only the text is masked. It fails without the fix.
• ./gradlew :posthog-android:testReleaseUnitTest --tests '*replay*' passes; spotlessCheck is clean.
• Not tested on a device.

📝 Checklist

• All commits are signed (see Commit signing)
• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran pnpm changeset to generate a changeset file

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

• Written with Claude Code, then run through a pre-PR review loop: one review cycle, then triage. Triage kept only the missing changeset.
• Deferred: skipping descendant rects that fall inside the container rect. This is a perf-only change and hasn't been measured.
• The test root sets clipChildren = false because getGlobalVisibleRect clips a child to its parent based on the grandparent's flag.

🤖 Generated with Claude Code

[marandaneto]

The code looks good; no qualifying findings. Verdict: correct.

[ioannisj]

🤖 An AI agent wrote this on behalf of @ioannisj.

Nothing blocks merging here. The descendant walk lines up with the privacy spec and the new test covers the overflow, the verified re-walk and the untagged control. One thing worth a look: a ComposeView that carries the ph-no-capture tag itself still skips the no-capture branch, which predates this PR and could be a follow-up.