"""Export key numbers from results/tables as LaTeX macros (paper/tables/numbers.tex) so that prose never hard-codes values."""
from __future__ import annotations
import re, pandas as pd, numpy as np, json
from pathlib import Path
from ..util import ROOT, read_json

TAB = ROOT / "results/tables"; OUT = ROOT / "paper/tables/numbers.tex"

def _key(s: str) -> str:
    s = re.sub(r"[^A-Za-z0-9]", " ", s).title().replace(" ", "")
    return re.sub(r"\d", lambda m: "ZOTTFFSSEN"[int(m.group(0))], s)   # LaTeX macro names cannot contain digits

def main():
    lines = ["% auto-generated by wp30.analysis.numbers — do not edit"]
    def add(name, val, fmt="{:.2f}"):
        if val is None or (isinstance(val, float) and np.isnan(val)):
            lines.append(f"\\newcommand{{\\{name}}}{{--}}")
        else:
            lines.append(f"\\newcommand{{\\{name}}}{{{fmt.format(val) if not isinstance(val, str) else val}}}")
    prim = TAB / "E1_core_primary.csv"
    if prim.exists():
        t = pd.read_csv(prim)
        add("nCoreTrials", int(pd.read_csv(TAB / "E1_core_failures.csv").iloc[:, 1:].to_numpy().sum()) if (TAB / "E1_core_failures.csv").exists() else None, "{:,}")
        for _, r in t.iterrows():
            k = _key(f"{r['family']} {r['pair']} {r['condition']}")
            add("Q" + k, r["quality"] * 100, "{:.0f}"); add("Qlo" + k, r["q_lo"] * 100, "{:.0f}"); add("Qhi" + k, r["q_hi"] * 100, "{:.0f}")
            add("B" + k, r["payload_bytes_median"], "{:,.0f}"); add("Tcold" + k, r["cold_ms_p50"], "{:,.0f}"); add("Twarm" + k, r["warm_ms_p50"], "{:,.0f}")
        tests = pd.read_csv(TAB / "E1_core_primary_tests.csv") if (TAB / "E1_core_primary_tests.csv").exists() else None
        if tests is not None:
            for _, r in tests.iterrows():
                k = _key(r["name"].split(" ")[0])
                add("Diff" + k, r["diff"] * 100, "{:+.1f}"); add("Lo" + k, r["lo"] * 100, "{:+.1f}"); add("Hi" + k, r["hi"] * 100, "{:+.1f}"); add("P" + k, r["p_holm"], "{:.3f}"); add("N" + k, int(r["n"]), "{:d}")
    e6 = TAB / "E6_network.csv"
    if e6.exists():
        t6 = pd.read_csv(e6); t6 = t6[t6["label"] != "loopback"]
        for _, r in t6.iterrows():
            tag = _key(f"{r['method']} {'' if pd.isna(r['bits']) else int(r['bits'])} {r['codec'] if isinstance(r['codec'], str) else ''} {int(r['target_tokens'])}")
            add("Esix" + tag + "Rt", r["roundtrip_ms_p50"], "{:,.0f}"); add("Esix" + tag + "Bytes", r["bytes"] / 1e6 if not pd.isna(r["bytes"]) else None, "{:.0f}")
            add("Esix" + tag + "Send", r["send_ms_p50"], "{:,.0f}"); add("Esix" + tag + "Prefill", r["remote_prefill_p50"] if not pd.isna(r.get("remote_prefill_p50", float("nan"))) else r.get("local_prefill_p50"), "{:,.0f}")
            add("Esix" + tag + "Load", r["remote_load_ms_p50"], "{:,.0f}"); add("Esix" + tag + "Gbps", r["eff_gbps"], "{:.2f}")
    if (ROOT / "protocol/frozen.json").exists():
        add("protocolHash", read_json(ROOT / "protocol/frozen.json")["protocol_hash"][:16], "{}")
    OUT.parent.mkdir(parents=True, exist_ok=True); OUT.write_text("\n".join(lines) + "\n")
    print(f"{len(lines)-1} macros -> {OUT}")

if __name__ == "__main__":
    main()