Fix #43: add ComplianceEvalRecord round-trip test

tests/test_compliance_roundtrip.py

@@ -0,0 +1,180 @@
+"""Round-trip test: a ComplianceEvalRecord produced against the canonical schema
+serialises to JSON and deserialises back to an equivalent object without loss."""
+
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from evomerge.schemas.compliance import (
+    ComplianceEvalRecord,
+    ComplianceError,
+    ConstraintCategory,
+    ConstraintIR,
+    ConstraintLevel,
+    ConstraintViolation,
+    EvidenceSpan,
+    RepairPolicy,
+    RepairStrategy,
+    RepairTraceEntry,
+    RunMode,
+    TokenCost,
+    ViolationStage,
+)
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _minimal_record() -> ComplianceEvalRecord:
+    """Minimal valid ComplianceEvalRecord (all required fields set)."""
+    return ComplianceEvalRecord(
+        task_id="task-001",
+        task_spec_hash="abc123",
+        model="gpt-4o",
+        mode=RunMode.direct,
+        final_pass=True,
+        artifact="The answer is 42.",
+    )
+
+
+def _full_record() -> ComplianceEvalRecord:
+    """ComplianceEvalRecord with all optional fields populated."""
+    violation = ConstraintViolation(
+        constraint_id="c-format-1",
+        level=ConstraintLevel.hard,
+        category=ConstraintCategory.format,
+        hint="Output must start with 'Answer:'",
+        evidence_span=EvidenceSpan(char_range=(0, 6), line_range=(1, 1)),
+        detected_at=ViolationStage.post_decode,
+    )
+    repair = RepairTraceEntry(
+        round=1,
+        violation_ids=["c-format-1"],
+        strategy=RepairStrategy.patch,
+        ok=True,
+        rolled_back=False,
+        remaining_violation_ids=[],
+        token_cost={"prompt": 120, "generation": 40},
+        latency_ms=210.5,
+    )
+    return ComplianceEvalRecord(
+        task_id="task-002",
+        task_spec_hash="def456",
+        model="claude-3-5-sonnet",
+        mode=RunMode.full_pcl,
+        violations=[violation],
+        repair_trace=[repair],
+        repair_rounds=1,
+        final_pass=True,
+        token_cost=TokenCost(prompt=800, generation=200, repair=160),
+        latency_ms=1342.7,
+        artifact="Answer: The answer is 42.",
+        error=None,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Round-trip tests
+# ---------------------------------------------------------------------------
+
+def test_minimal_record_json_round_trip() -> None:
+    record = _minimal_record()
+    raw = record.model_dump_json()
+    restored = ComplianceEvalRecord.model_validate_json(raw)
+
+    assert restored.task_id == record.task_id
+    assert restored.task_spec_hash == record.task_spec_hash
+    assert restored.model == record.model
+    assert restored.mode == RunMode.direct
+    assert restored.final_pass is True
+    assert restored.artifact == record.artifact
+    assert restored.schema_version == "compliance-eval-record/v1"
+    assert restored.violations == []
+    assert restored.repair_trace == []
+    assert restored.error is None
+
+
+def test_full_record_json_round_trip_preserves_nested_fields() -> None:
+    record = _full_record()
+    raw = record.model_dump_json()
+    restored = ComplianceEvalRecord.model_validate_json(raw)
+
+    assert len(restored.violations) == 1
+    v = restored.violations[0]
+    assert v.constraint_id == "c-format-1"
+    assert v.level == ConstraintLevel.hard
+    assert v.category == ConstraintCategory.format
+    assert v.evidence_span is not None
+    assert v.evidence_span.char_range == (0, 6)
+    assert v.detected_at == ViolationStage.post_decode
+
+    assert len(restored.repair_trace) == 1
+    r = restored.repair_trace[0]
+    assert r.round == 1
+    assert r.ok is True
+    assert r.strategy == RepairStrategy.patch
+
+    assert restored.token_cost.prompt == 800
+    assert restored.repair_rounds == 1
+    assert abs(restored.latency_ms - 1342.7) < 0.01
+
+
+def test_json_output_is_valid_json_and_contains_schema_version() -> None:
+    raw = _minimal_record().model_dump_json()
+    parsed = json.loads(raw)  # must not raise
+    assert parsed["schema_version"] == "compliance-eval-record/v1"
+    assert parsed["task_id"] == "task-001"
+
+
+def test_record_with_error_field_round_trips() -> None:
+    record = ComplianceEvalRecord(
+        task_id="task-err",
+        task_spec_hash="err-hash",
+        model="gpt-4o-mini",
+        mode=RunMode.prompt_retry,
+        final_pass=False,
+        artifact="",
+        error=ComplianceError(
+            kind="model_error",
+            message="timeout after 30s",
+            stage="generate",
+        ),
+    )
+    restored = ComplianceEvalRecord.model_validate_json(record.model_dump_json())
+    assert restored.error is not None
+    assert restored.error.kind == "model_error"
+    assert restored.error.stage == "generate"
+    assert restored.final_pass is False
+
+
+def test_round_trip_preserves_enum_string_values() -> None:
+    record = ComplianceEvalRecord(
+        task_id="t",
+        task_spec_hash="h",
+        model="m",
+        mode=RunMode.full_pcl,
+        final_pass=False,
+        artifact="x",
+    )
+    raw_dict = json.loads(record.model_dump_json())
+    assert raw_dict["mode"] == "full_pcl"
+    restored = ComplianceEvalRecord.model_validate(raw_dict)
+    assert restored.mode == RunMode.full_pcl
+
+
+def test_multiple_records_survive_jsonl_round_trip(tmp_path) -> None:
+    """Verify JSONL write→read cycle using evomerge.io helpers."""
+    from evomerge.io import load_jsonl, write_jsonl
+
+    records = [_minimal_record(), _full_record()]
+    path = tmp_path / "compliance.jsonl"
+    write_jsonl(records, path)
+
+    loaded = load_jsonl(path, ComplianceEvalRecord)
+    assert len(loaded) == 2
+    assert loaded[0].task_id == "task-001"
+    assert loaded[1].task_id == "task-002"
+    assert len(loaded[1].violations) == 1