[aw][test audit] `_detect_resume`: missing regression test for `ASSISTANT_TURN_START`-only post-shutdown not marking session as [Content truncated due to length]

[microsasa]

Root Cause Analysis

_detect_resume (src/copilot_usage/parser.py) iterates events after the last shutdown and handles four types differently:

Event type	Effect
ASSISTANT_MESSAGE	session_resumed = True + accumulate output tokens
USER_MESSAGE	session_resumed = True + count post-shutdown user messages
SESSION_RESUME	session_resumed = True + capture timestamp
ASSISTANT_TURN_START	post_shutdown_turn_starts += 1 only — does not set session_resumed = True
All others	Ignored

The design intention is clear: a turn start alone (with no assistant message produced) does not constitute a resumed session — the session's shutdown is still its final state. However, this invariant is nowhere explicitly tested.

TestDetectResumeNonIndicatorEvents (line 6079 of test_parser.py) documents non-indicator events by testing TOOL_EXECUTION_COMPLETE + SESSION_ERROR, but ASSISTANT_TURN_START is structurally different: it is a partially handled event (it has a counter) that intentionally does not trigger session_resumed. Any developer reading the elif chain without the test context could reasonably consider adding session_resumed = True there — and no test would catch the regression.

Existing coverage (for context)

• test_counts_user_messages_and_turn_starts — combines USER_MESSAGE and ASSISTANT_TURN_START after shutdown; session_resumed becomes True via the user messages, not turn starts.
• TestDetectResumeRangeIndex.test_correctness_with_large_event_list — mixes all event types including turn starts; session_resumed=True comes from user/assistant messages.

There is no test where ASSISTANT_TURN_START events are the only post-shutdown events and the expectation is session_resumed=False.

Tests to add

In TestDetectResumeNonIndicatorEvents (or a new sibling class)

test_turn_start_alone_does_not_set_session_resumed

• Build a session: session.start → user.message → session.shutdown → assistant.turn_start × N
• Assert result.session_resumed is False
• Assert result.post_shutdown_turn_starts == N
• Assert result.post_shutdown_user_messages == 0
• Assert result.post_shutdown_output_tokens == 0
• Assert result.last_resume_time is None

test_full_build_session_summary_turn_start_only_is_completed

• Same event sequence, run through build_session_summary
• Assert summary.is_active is False
• Assert summary.end_time is not None (shutdown time preserved)
• Assert summary.active_model_calls == N (turn starts still counted)
• Assert summary.active_user_messages == 0

Regression scenarios to cover

• A refactor adds session_resumed = True to the ASSISTANT_TURN_START branch → the first test above would catch it.
• A refactor promotes ASSISTANT_TURN_START into the "session_resumed indicators" set → the second test above would catch the incorrect is_active=True.

Generated by Test Suite Analysis · ● 20M · ◷

[microsasa]

Closing as resolved — the function this issue targeted (_detect_resume) was deleted in PR #990, which folded resume detection into _first_pass as rolling post-shutdown accumulators.

The specific scenario described here (ASSISTANT_TURN_START-only after shutdown should NOT set session_resumed=True) is covered by TestOrphanedTurnStartDoesNotInflateActiveCalls (test_parser.py line 8913), added as part of issue #908. In the current code, ASSISTANT_TURN_START only increments _ps_turn_starts and never sets _ps_resumed = True (parser.py lines 853-856).