"""scheduled_jobs UUID-seed guard.

Hard CI invariant preventing a FOURTH recurrence of the no-dash
``scheduled_jobs.id`` defect that migrations 059 / 162 / 248 each had to
repair: a seed migration that writes the primary key as
``REPLACE(UUID(), '-', '')`` produces a 32-char no-dash id in the
CHAR(36) column. The ``GUID`` TypeDecorator then normalises it to the
dashed form on read, so the worker schedules APScheduler with the dashed
string but the run-wrapper's ``session.get(ScheduledJob, uuid.UUID(...))``
queries the dashed value, never matches the stored no-dash row, and the
cron silently never runs (``run_count`` stays 0 forever).

The fix is always the same: seed ``scheduled_jobs.id`` with a plain
dashed ``UUID()``. This guard rejects any NEW migration that reintroduces
the ``REPLACE(UUID(), '-', '')`` pattern anywhere near a
``scheduled_jobs`` INSERT, turning a thrice-recurring production incident
class into a hard stop at PR time.

Runs without a database connection — pure text scan of the version files.
Exits 0 on success, 1 with a human-readable diagnostic on failure.
"""

from __future__ import annotations

import re
import sys
from pathlib import Path

# The exact offending pattern (whitespace-tolerant). We only care about
# the no-dash form: ``REPLACE(UUID(), '-', '')``. The dashed no-op
# ``REPLACE(UUID(), '-', '-')`` and plain ``UUID()`` are fine.
_BAD_REPLACE = re.compile(r"REPLACE\(\s*UUID\(\)\s*,\s*'-'\s*,\s*''\s*\)")

# Migrations that pre-date this guard and legitimately carry the pattern
# for NON-scheduled_jobs tables (auth_roles / email-group seeds, etc.),
# or are the historical fix migrations whose DOCSTRINGS quote the pattern.
# These are grandfathered: the guard's job is to stop NEW scheduled_jobs
# seeds, not to rewrite history. The eight scheduled_jobs offenders
# (041/064/114/115/134/135/155) were corrected in the 248 PR and are NOT
# on this list, so a regression in any of them re-trips the guard.
_GRANDFATHERED = {
    "001_initial_mysql.py",
    "040_drop_cloud_admin_role.py",
    "043_hr_finance_roles_and_email_groups.py",
    "092_normalize_auth_role_ids.py",
    "059_fix_scheduled_jobs_id_dashes.py",   # docstring quote only
    "162_fix_scheduled_jobs_id_dashes_2.py",  # docstring quote only
    "248_fix_scheduled_jobs_id_dashes_3.py",  # docstring quote only
}


def _versions_dir() -> Path:
    return Path(__file__).resolve().parent.parent / "alembic" / "versions"


def main() -> int:
    versions = _versions_dir()
    if not versions.is_dir():
        print(f"[scheduled-jobs-guard] versions dir not found: {versions}")
        return 1

    offenders: list[str] = []
    for path in sorted(versions.glob("*.py")):
        if path.name in _GRANDFATHERED:
            continue
        text = path.read_text(encoding="utf-8")
        if not _BAD_REPLACE.search(text):
            continue
        # Only flag when the pattern co-occurs with a scheduled_jobs INSERT
        # in the same file — the defect is specific to that table's id.
        if "scheduled_jobs" not in text:
            continue
        # Drive the gating decision off the SAME whole-text matcher as the
        # pre-filter (line numbers via finditer). A per-line re-scan would
        # miss a match split across physical lines (``\s`` spans newlines),
        # silently passing the guard — so iterate the full text instead.
        for m in _BAD_REPLACE.finditer(text):
            line_no = text.count("\n", 0, m.start()) + 1
            offenders.append(f"  {path.name}:{line_no}: {m.group(0)}")

    if offenders:
        print(
            "[scheduled-jobs-guard] FAIL — REPLACE(UUID(), '-', '') found in a "
            "migration that touches scheduled_jobs. Seed scheduled_jobs.id with "
            "a plain dashed UUID() instead; the no-dash 32-char form strands the "
            "cron (see migration 248 docstring).\n" + "\n".join(offenders)
        )
        return 1

    print("[scheduled-jobs-guard] OK — no no-dash UUID seeds in scheduled_jobs.")
    return 0


if __name__ == "__main__":
    sys.exit(main())