# =============================================================================
# Author: Simon-Pierre Boucher
# Contact: contact@spboucher.ai
# =============================================================================
"""Compare regenerated result CSVs (_verify/results) against the originals
(results/). Numeric columns must match within rtol=1e-9 (and we report
whether they are byte-identical); non-numeric columns must match exactly."""

import sys
from pathlib import Path

import numpy as np
import pandas as pd

ROOT = Path(__file__).resolve().parents[1]
ORIG = ROOT / "results"
NEW = ROOT / "_verify" / "results"

overall_ok = True
rows = []
for new_file in sorted(NEW.glob("*.csv")):
    name = new_file.name
    orig_file = ORIG / name
    if not orig_file.exists():
        rows.append((name, "NEW (no original to compare)"))
        continue

    byte_identical = new_file.read_bytes() == orig_file.read_bytes()
    if byte_identical:
        rows.append((name, "IDENTICAL (byte-for-byte)"))
        continue

    a = pd.read_csv(orig_file)
    b = pd.read_csv(new_file)
    if a.shape != b.shape:
        rows.append((name, f"MISMATCH shape {a.shape} vs {b.shape}"))
        overall_ok = False
        continue
    if list(a.columns) != list(b.columns):
        rows.append((name, f"MISMATCH columns"))
        overall_ok = False
        continue

    bad_cols = []
    max_rel = 0.0
    for col in a.columns:
        if pd.api.types.is_numeric_dtype(a[col]) and pd.api.types.is_numeric_dtype(b[col]):
            av, bv = a[col].values.astype(float), b[col].values.astype(float)
            both_nan = np.isnan(av) & np.isnan(bv)
            close = np.isclose(av, bv, rtol=1e-9, atol=1e-15, equal_nan=True)
            if not (close | both_nan).all():
                bad_cols.append(col)
            with np.errstate(all="ignore"):
                rel = np.abs(av - bv) / np.maximum(np.abs(av), 1e-300)
                rel = rel[~(both_nan | np.isnan(rel))]
                if len(rel):
                    max_rel = max(max_rel, np.nanmax(rel))
        else:
            if not a[col].fillna("§na§").astype(str).equals(
                    b[col].fillna("§na§").astype(str)):
                bad_cols.append(col)

    if bad_cols:
        rows.append((name, f"MISMATCH in columns {bad_cols} (max rel diff {max_rel:.2e})"))
        overall_ok = False
    else:
        rows.append((name, f"EQUAL numerically (max rel diff {max_rel:.2e})"))

width = max(len(r[0]) for r in rows) + 2
for name, status in rows:
    print(f"{name:<{width}} {status}")

print("\n" + ("ALL COMPARED FILES MATCH" if overall_ok else "DISCREPANCIES FOUND"))
sys.exit(0 if overall_ok else 1)