cyclone/backend/tests/test_db_migrate.py

"""Tests for the PRAGMA user_version migration runner."""

from __future__ import annotations

from pathlib import Path

import pytest
import sqlalchemy as sa
import sqlalchemy.exc

from cyclone import db_migrate


def _fresh_engine(tmp_path: Path) -> sa.Engine:
    return sa.create_engine(f"sqlite:///{tmp_path}/m.db", future=True)


def _user_version(engine: sa.Engine) -> int:
    with engine.connect() as c:
        return c.exec_driver_sql("PRAGMA user_version").scalar() or 0


def test_run_on_empty_db_bumps_to_latest(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setattr(db_migrate, "MIGRATIONS_DIR", tmp_path)
    (tmp_path / "0001_initial.sql").write_text("-- version: 1\nCREATE TABLE t (id INTEGER);\n")

    engine = _fresh_engine(tmp_path)
    db_migrate.run(engine)

    assert _user_version(engine) == 1
    with engine.connect() as c:
        rows = c.exec_driver_sql("SELECT name FROM sqlite_master WHERE type='table' AND name='t'").all()
    assert len(rows) == 1


def test_run_is_idempotent(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setattr(db_migrate, "MIGRATIONS_DIR", tmp_path)
    (tmp_path / "0001_initial.sql").write_text("-- version: 1\nCREATE TABLE t (id INTEGER);\n")

    engine = _fresh_engine(tmp_path)
    db_migrate.run(engine)
    db_migrate.run(engine)  # second call should be a no-op

    assert _user_version(engine) == 1


def test_run_skips_already_applied_migrations(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setattr(db_migrate, "MIGRATIONS_DIR", tmp_path)
    (tmp_path / "0001_initial.sql").write_text("-- version: 1\nCREATE TABLE t1 (id INTEGER);\n")
    (tmp_path / "0002_add_col.sql").write_text("-- version: 2\nALTER TABLE t1 ADD COLUMN name TEXT;\n")

    engine = _fresh_engine(tmp_path)
    db_migrate.run(engine)
    assert _user_version(engine) == 2

    # Modify 0002 (a no-op for SQLite ALTER, but we want to verify the
    # runner doesn't re-run already-applied migrations).
    db_migrate.run(engine)
    assert _user_version(engine) == 2


def test_run_raises_on_missing_version_header(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setattr(db_migrate, "MIGRATIONS_DIR", tmp_path)
    (tmp_path / "0001_bad.sql").write_text("CREATE TABLE t (id INTEGER);\n")  # no header

    engine = _fresh_engine(tmp_path)
    with pytest.raises(RuntimeError, match="missing '-- version"):
        db_migrate.run(engine)


def test_run_rolls_back_version_on_failed_statement(
    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
) -> None:
    """If a DDL statement fails, the version bump must also roll back."""
    monkeypatch.setattr(db_migrate, "MIGRATIONS_DIR", tmp_path)
    (tmp_path / "0001_initial.sql").write_text(
        "-- version: 1\nCREATE TABLE t (id INTEGER);\n"
    )

    engine = _fresh_engine(tmp_path)
    db_migrate.run(engine)  # applies 0001, user_version=1
    assert _user_version(engine) == 1

    # Now drop in a second migration whose DDL conflicts with the first.
    # The runner should apply 0001 and 0002 in one call; 0002 fails, the
    # transaction rolls back, and user_version must stay at 1 (not bump to 2).
    (tmp_path / "0002_bad.sql").write_text(
        "-- version: 2\nCREATE TABLE t (id INTEGER);\n"  # table already exists!
    )

    with pytest.raises(sqlalchemy.exc.SQLAlchemyError):
        db_migrate.run(engine)

    assert _user_version(engine) == 1  # bumped atomically, rolled back on failure


def test_run_ignores_non_sql_files(
    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
) -> None:
    """README.md and similar files in MIGRATIONS_DIR must not be executed."""
    monkeypatch.setattr(db_migrate, "MIGRATIONS_DIR", tmp_path)
    (tmp_path / "0001_initial.sql").write_text(
        "-- version: 1\nCREATE TABLE t (id INTEGER);\n"
    )
    (tmp_path / "README.md").write_text("Don't run me!\nCREATE TABLE should_not_exist (x INT);\n")

    engine = _fresh_engine(tmp_path)
    db_migrate.run(engine)

    with engine.connect() as c:
        rows = c.exec_driver_sql(
            "SELECT name FROM sqlite_master WHERE type='table' AND name='should_not_exist'"
        ).all()
    assert len(rows) == 0
    assert _user_version(engine) == 1


def test_migration_latest_idempotent_on_fresh_db(tmp_path: Path) -> None:
    """All migrations up to the current head run cleanly on a fresh DB,
    and a second run is a no-op (no version bump). SP22 bumped the
    expected head from 14 to 15 with the new UNIQUE-drop migration.
    SP27 Task 11 bumped it to 16 with the claims.matched_remittance_id
    index (drift-check perf).
    SP27 Task 17 bumped it to 17 with the patient_control_number
    backfill UPDATE (the migration runner now applies DML too).
    SP28 bumped it to 18 with the claim_acks join table.
    SP32 bumped it to 19 with rendering_provider_npi +
    service_provider_npi on claims and remittances.
    SP37 bumped it to 20 with batch transaction_set_control_number.
    SP39 bumped it to 21 with the resubmissions audit table.
    SP41 bumped it to 22 with submission_dedup.
    """
    engine = _fresh_engine(tmp_path)
    db_migrate.run(engine)
    v_after_first = _user_version(engine)
    assert v_after_first == 23, f"expected head=23, got {v_after_first}"

    db_migrate.run(engine)
    assert _user_version(engine) == 23, "second run should not bump version"


def test_drop_claims_unique_constraint_migration(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
    """SP22: migration 0015 recreates the `claims` table without the inline
    `UNIQUE(batch_id, patient_control_number)` constraint, so two claims in
    one batch can share a patient_control_number (real 837P multi-claim
    subscriber loops do this).

    Discovery (2026-06-23): the inline UNIQUE does not exist in the current
    production DB or in main's fresh-DB schema, so this migration is a
    defensive no-op against the current state. The test still proves
    migration correctness: (a) all migrations up to v15 run cleanly,
    (b) two rows with the same (batch_id, patient_control_number) can
    be inserted after the migration (proving no UNIQUE was re-introduced
    by the table recreation).
    """
    # Real migrations dir so the test exercises the actual 0015 file.
    monkeypatch.setattr(
        db_migrate,
        "MIGRATIONS_DIR",
        Path(__file__).parent.parent / "src" / "cyclone" / "migrations",
    )
    engine = _fresh_engine(tmp_path)

    db_migrate.run(engine)
    assert _user_version(engine) == 23, f"expected head=23, got {_user_version(engine)}"

    # Two claims in one batch with the same patient_control_number
    # must be insertable. If 0015's table recreation re-introduced a
    # UNIQUE(batch_id, patient_control_number), this would raise
    # IntegrityError. (The test also implicitly asserts the FK from
    # claims to batches still works after the recreation.)
    with engine.begin() as conn:
        conn.exec_driver_sql(
            "INSERT INTO batches (id, kind, input_filename, parsed_at) "
            "VALUES ('B1', '837p', 'test.txt', '2026-01-01 00:00:00')"
        )
        conn.exec_driver_sql(
            "INSERT INTO claims (id, batch_id, patient_control_number, charge_amount) "
            "VALUES ('CLM-1', 'B1', 'SAME-PCN', 100)"
        )
        conn.exec_driver_sql(
            "INSERT INTO claims (id, batch_id, patient_control_number, charge_amount) "
            "VALUES ('CLM-2', 'B1', 'SAME-PCN', 200)"
        )
        rows = conn.exec_driver_sql(
            "SELECT id, charge_amount FROM claims "
            "WHERE patient_control_number='SAME-PCN' ORDER BY id"
        ).all()
        assert [r[0] for r in rows] == ["CLM-1", "CLM-2"]
        assert [float(r[1]) for r in rows] == [100.0, 200.0]



# ---------------------------------------------------------------------------
# SP39: migration 0021 creates the resubmissions table
# ---------------------------------------------------------------------------


def test_migration_0021_creates_resubmissions_table(tmp_path: Path) -> None:
    """SP39: 0021_resubmissions.sql adds the resubmissions audit table
    with the documented columns + unique constraint on
    (claim_id, interchange_control_number)."""
    # Point the runner at the REAL migrations dir so we exercise 0021.
    real_dir = Path(db_migrate.__file__).parent / "migrations"
    import cyclone.db_migrate as real_migrate_mod
    real_dir_str = str(real_dir)

    engine = _fresh_engine(tmp_path)
    # monkeypatch the module-level MIGRATIONS_DIR
    import importlib
    monkey_save = real_migrate_mod.MIGRATIONS_DIR
    real_migrate_mod.MIGRATIONS_DIR = Path(real_dir_str)
    try:
        db_migrate.run(engine)
    finally:
        real_migrate_mod.MIGRATIONS_DIR = monkey_save

    # Verify the table exists with the expected columns + unique index.
    with engine.connect() as conn:
        version = conn.exec_driver_sql("PRAGMA user_version").scalar()
        assert version >= 21
        cols = conn.exec_driver_sql(
            "PRAGMA table_info(resubmissions)"
        ).all()
        col_names = {row[1] for row in cols}
        assert col_names == {
            "id", "claim_id", "batch_id", "resubmitted_at",
            "source_corrected_path",
            "interchange_control_number", "group_control_number",
        }
        idx_rows = conn.exec_driver_sql(
            "SELECT name, sql FROM sqlite_master "
            "WHERE type='index' AND tbl_name='resubmissions'"
        ).all()
        idx_names = {row[0] for row in idx_rows}
        assert "ix_resubmissions_claim_id" in idx_names
        assert "ix_resubmissions_batch_id" in idx_names
        assert "ux_resubmissions_claim_icn" in idx_names