"""Modern country-year panel: V-Dem v16 (institutions, outcomes) + Maddison 2023 (population, GDP per capita).

Variables fixed in research/preregistration.md §3 and the codebook readings recorded in research/measurement.md:
  K items: v2clrspct, v2stcritrecadm, v2cltrnslw, v2stfisccap, v2svstterr, v2elembcap
  D: Maddison `pop` (thousands) ; e_area (km2, 1886+)
  Outcomes (events in year t): Y4 civil war onset = e_civil_war 0->1 ; Y5 coup = e_coups >= 1 ;
           Y6 irregular regime end = last year of a regime (v2regidnr changes) whose v2regendtype is in {0,1,2,3,5,6,7,8}
  Covariates: Maddison gdppc, v2x_polyarchy, e_regionpol_6C
Crosswalk: ISO3 (`country_text_id` == `countrycode`); unmatched countries are listed in
data/metadata/crosswalk_vdem_maddison.csv and kept without Maddison values (no substitution from e_pop).
"""
from __future__ import annotations
from pathlib import Path
import numpy as np
import pandas as pd
import pyreadr

ROOT = Path(__file__).resolve().parents[3]
VDEM = ROOT / "data" / "raw" / "vdem_v16_rdata" / "vdem_v16.RData"
MADD = ROOT / "data" / "raw" / "maddison_2023_xlsx" / "mpd2023_web.xlsx"
META = ROOT / "data" / "metadata"

K_ITEMS = ["v2clrspct", "v2stcritrecadm", "v2cltrnslw", "v2stfisccap", "v2svstterr", "v2elembcap"]
IRREGULAR_END = {0, 1, 2, 3, 5, 6, 7, 8}
KEEP = ["country_name", "country_text_id", "country_id", "year", "e_regionpol_6C", "e_regionpol", "v2x_polyarchy",
        "e_civil_war", "e_miinterc", "e_coups", "e_pt_coup", "v2regendtype", "v2regint", "v2regdur", "e_pop", "e_area",
        "e_gdppc", "v2x_regime", "v2x_rule", "v2regidnr"] + K_ITEMS


def build() -> dict[str, pd.DataFrame]:
    vd = pyreadr.read_r(VDEM)["vdem"]
    cols = [c for c in KEEP if c in vd.columns]
    v = vd[cols].copy()
    v["year"] = v.year.astype(int)
    md = pd.read_excel(MADD, sheet_name="Full data")
    md = md.rename(columns={"countrycode": "country_text_id", "pop": "maddison_pop", "gdppc": "maddison_gdppc"})
    md["year"] = md.year.astype(int)
    m = v.merge(md[["country_text_id", "year", "maddison_pop", "maddison_gdppc"]], on=["country_text_id", "year"], how="left")
    # crosswalk audit
    vset, mset = set(v.country_text_id.unique()), set(md.country_text_id.unique())
    cw = pd.DataFrame({"code": sorted(vset | mset)})
    cw["in_vdem"] = cw.code.isin(vset); cw["in_maddison"] = cw.code.isin(mset)
    names = v.drop_duplicates("country_text_id").set_index("country_text_id").country_name
    cw["vdem_name"] = cw.code.map(names)
    META.mkdir(parents=True, exist_ok=True)
    cw.to_csv(META / "crosswalk_vdem_maddison.csv", index=False)
    m = m.sort_values(["country_text_id", "year"]).reset_index(drop=True)
    g = m.groupby("country_text_id")
    # outcomes in year t
    cw_prev = g.e_civil_war.shift(1)
    m["y4_civilwar_onset"] = np.where(m.e_civil_war.isna() | cw_prev.isna(), np.nan,
                                      ((m.e_civil_war == 1) & (cw_prev == 0)).astype(float))
    m["y5_coup"] = np.where(m.e_coups.isna(), np.nan, (m.e_coups >= 1).astype(float))
    # v2regendtype is carried on every year of a regime (how that regime eventually ended; 13 = still exists).
    # The event is placed in the regime's LAST year: v2regidnr changes between t and t+1 (within country).
    next_id = g.v2regidnr.shift(-1)
    last_year = m.v2regidnr.notna() & next_id.notna() & (m.v2regidnr != next_id)
    m["regime_end_year"] = last_year.astype(float)
    m["y6_irregular_regime_end"] = np.where(m.v2regendtype.isna() | next_id.isna(), np.nan,
                                           (last_year & m.v2regendtype.isin(IRREGULAR_END)).astype(float))
    m["log_pop"] = np.log(m.maddison_pop.where(m.maddison_pop > 0))
    m["log_area"] = np.log(m.e_area.where(m.e_area > 0))
    m["log_gdppc"] = np.log(m.maddison_gdppc.where(m.maddison_gdppc > 0))
    return {"panel": m, "crosswalk": cw}