import json
import tempfile
import unittest
from pathlib import Path

from prepare_distill_dataset import DatasetError, prepare


ROOT = Path(__file__).parent
FIXTURE = ROOT / "teacher_candidates.jsonl"


class DistillationDatasetTests(unittest.TestCase):
    def test_prepare_emits_separate_sft_files_and_manifest(self):
        with tempfile.TemporaryDirectory() as temp:
            manifest = prepare(FIXTURE, Path(temp))
            self.assertEqual(manifest["counts"], {"eval": 2, "train": 3})
            self.assertTrue(manifest["known_reasoning_fields_and_think_markers_rejected"])
            self.assertFalse(manifest["semantic_content_audited"])
            train_rows = [json.loads(line) for line in (Path(temp) / "train.jsonl").read_text().splitlines()]
            self.assertEqual(train_rows[0]["messages"][1]["role"], "assistant")
            self.assertNotIn("quality_pass", train_rows[0])
            self.assertNotIn("teacher_response", train_rows[0])

    def test_task_key_contamination_is_rejected(self):
        rows = [json.loads(line) for line in FIXTURE.read_text().splitlines()]
        rows[-1]["task_key"] = rows[0]["task_key"]
        with tempfile.TemporaryDirectory() as temp:
            source = Path(temp) / "bad.jsonl"
            source.write_text("\n".join(json.dumps(row) for row in rows) + "\n")
            with self.assertRaisesRegex(DatasetError, "train/eval task_key contamination"):
                prepare(source, Path(temp) / "out")

    def test_prompt_contamination_is_rejected(self):
        rows = [json.loads(line) for line in FIXTURE.read_text().splitlines()]
        rows[-1]["prompt"] = rows[0]["prompt"]
        with tempfile.TemporaryDirectory() as temp:
            source = Path(temp) / "bad.jsonl"
            source.write_text("\n".join(json.dumps(row) for row in rows) + "\n")
            with self.assertRaisesRegex(DatasetError, "normalized prompt contamination"):
                prepare(source, Path(temp) / "out")

    def test_private_reasoning_is_rejected(self):
        rows = [json.loads(line) for line in FIXTURE.read_text().splitlines()]
        rows[0]["private_reasoning"] = "do not export"
        with tempfile.TemporaryDirectory() as temp:
            source = Path(temp) / "bad.jsonl"
            source.write_text("\n".join(json.dumps(row) for row in rows) + "\n")
            with self.assertRaisesRegex(DatasetError, "private reasoning"):
                prepare(source, Path(temp) / "out")

    def test_no_partial_output_when_split_missing(self):
        rows = [json.loads(line) for line in FIXTURE.read_text().splitlines()]
        with tempfile.TemporaryDirectory() as temp:
            source, out = Path(temp) / "one.jsonl", Path(temp) / "out"
            source.write_text("\n".join(json.dumps(r) for r in rows if r["split"] == "train"))
            with self.assertRaisesRegex(DatasetError, "non-empty"):
                prepare(source, out)
            self.assertFalse(out.exists())

    def test_existing_output_is_preserved(self):
        with tempfile.TemporaryDirectory() as temp:
            marker = Path(temp) / "keep.txt"
            marker.write_text("original")
            with self.assertRaisesRegex(DatasetError, "refusing overwrite"):
                prepare(FIXTURE, Path(temp))
            self.assertEqual(marker.read_text(), "original")

    def test_nested_reasoning_and_think_are_rejected(self):
        for field, value in [("metadata", {"reasoning_content": "private"}),
                             ("teacher_response", "<think>private</think>final")]:
            with self.subTest(field=field), tempfile.TemporaryDirectory() as temp:
                rows = [json.loads(line) for line in FIXTURE.read_text().splitlines()]
                rows[0][field] = value
                source = Path(temp) / "bad.jsonl"
                source.write_text("\n".join(json.dumps(r) for r in rows))
                with self.assertRaisesRegex(DatasetError, "private reasoning"):
                    prepare(source, Path(temp) / "out")


if __name__ == "__main__":
    unittest.main()
