[aw][code health] Duplicated ModelMetrics merging logic in parser.py and report.py

[microsasa]

Root Cause

ModelMetrics dict merging is implemented twice with different approaches:

parser.py — build_session_summary() (lines 279–299): Constructs new ModelMetrics, RequestMetrics, and TokenUsage objects on every merge. Safe, but verbose.

report.py — _aggregate_model_metrics() (lines 544–564): Uses model_copy() on first insertion, then mutates the copy's fields in-place for subsequent merges. This is fragile — if the model_copy() is ever removed, it silently corrupts session data passed from the caller.

Because this logic is duplicated, any future bug fix or field addition (e.g., a new token type on TokenUsage) must be applied in both places. They have already diverged in style.

Proposed Fix

Extract a shared helper, e.g. in models.py or a new copilot_usage/_utils.py:

def merge_model_metrics(
    base: dict[str, ModelMetrics],
    additional: dict[str, ModelMetrics],
) -> dict[str, ModelMetrics]:
    """Return a new dict merging *additional* into *base* without mutation."""
    result = {name: ModelMetrics(...) for name, mm in base.items()}  # deep copy
    for name, mm in additional.items():
        if name in result:
            existing = result[name]
            result[name] = ModelMetrics(
                requests=RequestMetrics(
                    count=existing.requests.count + mm.requests.count,
                    cost=existing.requests.cost + mm.requests.cost,
                ),
                usage=TokenUsage(
                    inputTokens=existing.usage.inputTokens + mm.usage.inputTokens,
                    outputTokens=existing.usage.outputTokens + mm.usage.outputTokens,
                    cacheReadTokens=existing.usage.cacheReadTokens + mm.usage.cacheReadTokens,
                    cacheWriteTokens=existing.usage.cacheWriteTokens + mm.usage.cacheWriteTokens,
                ),
            )
        else:
            result[name] = mm
    return result

Replace both implementations with calls to this helper.

Testing Requirements

• Unit test merge_model_metrics directly: empty dicts, single entry, overlapping keys, disjoint keys, all token fields accumulate correctly.
• Regression: existing test_parser.py and test_report.py coverage must continue to pass unchanged.
• Verify the extracted function is the single source of truth by removing the inline logic from both callers.

Generated by Code Health Analysis · ◷