[vastin]

Issue #, if available: N/A

Description of changes:

ServiceEventsOtlpEmitter.mapToValue reversed the key order of every map it
converted to an OTLP Value. The last child pushed on the result stack was
popped into the map first, so a source map ordered
exception_type -> exception_message -> stack_trace -> call_path was emitted as
call_path -> stack_trace -> exception_message -> exception_type.

This makes the incident-snapshot body.exception_info serialize in reverse of
the schema order shared with the Python and JS SDKs. CloudWatch delivers the log
body verbatim, so the reversed order reaches consumers and breaks the ordered
e2e assertion on exception_info.

Fix: pop the children into a temp array in reverse index order, then insert
forward — preserving the source map's insertion order. This mirrors the existing
ListMarker branch. Arrays were already order-correct; only maps were affected.

Testing:

• Added testEmitIncidentSnapshotPreservesExceptionInfoFieldOrder — a
  regression guard that fails on the old reversing code.
• Full serviceevents emitter unit suite passes (9/9).
• Live dev-loop against CloudWatch Application Signals: the incident-snapshot
  e2e cell flips from FAIL to PASS with only this change (5/5 cells pass).

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[github-actions]

Claude finished @vastin's task in 1m 26s —— View job

---

Claude Code Review

• Read prior review comments to avoid duplicates (none found)
• Review the PR diff
• Analyze code changes for quality, correctness, tests
• Post findings

Summary

The fix is small, correct, and well-scoped. No blocking issues.

Correctness of the fix (ServiceEventsOtlpEmitter.java:473-489):
The producer side (line 524) pushes values in reverse index order, so the last-pushed child sits on top of resultStack. Popping into vals[m.keys.length - 1] first restores the intended index alignment, and forward-inserting into a LinkedHashMap preserves source insertion order. This mirrors the already-correct ListMarker branch — good consistency.

Test coverage (ServiceEventsOtlpEmitterTest.java:305-405):
The regression guard exercises both:

• Top-level exception_info map order (exception_type → exception_message → stack_trace → call_path)
• Nested call_path entry order (function_name → caller_function_name → duration_ns → error → is_async)

Nested-order verification is the important piece — it catches any future regression where the recursion/iteration is fixed at the top level but broken for children.

Minor observations (non-blocking):

• The two-loop pop-then-insert pattern could be written as a single addFirst-style loop, but the current form matches the ListMarker branch exactly, which is more valuable for readability than terseness. Leave as-is.
• The block comment at lines 476–479 is a good example of a WHY comment — the stack-order invariant is non-obvious and easy to break in future edits.
• Consider a small unit test targeting mapToValue directly (rather than only via emitIncidentSnapshot) to isolate ordering behavior from the emitter's other logic. Not required — the current e2e-shaped test is sufficient and mirrors real use.

Nothing else to flag. LGTM.
· Branch

[github-actions]

Claude finished @vastin's task in 57s —— View job

---

Claude Code Review

• Read prior review comments to avoid duplicates
• Review the PR diff
• Analyze code changes for quality, correctness, tests
• Post findings

Summary

No new findings on this synchronize. The substantive diff hasn't changed since the prior review at 21:18Z — the new commit is a merge from main (1c090ea). The fix at ServiceEventsOtlpEmitter.java:473-489 and the regression test at ServiceEventsOtlpEmitterTest.java:305-405 remain correct and well-scoped:

• Pop-into-temp-array-then-forward-insert preserves LinkedHashMap insertion order and mirrors the already-correct ListMarker branch.
• The regression test guards both the top-level exception_info map order and the nested call_path entry order.
• The block comment at lines 476–479 documents the non-obvious stack-order invariant.

PR is already approved by @mxiamxia. Nothing further to flag.
· Branch