"""Sample N parallel "Program and Code" conversations and write each language as .funes.jsonl.

usage: python3 convert.py N SEED
writes corpus/{zh,en}/<conversation_id>.funes.jsonl and corpus/ids.txt
"""
import json, random, sys
from pathlib import Path

n, seed = int(sys.argv[1]), int(sys.argv[2])
src = Path("sharegpt_jsonl")


def load(name):
    out = {}
    for line in open(src / name, encoding="utf-8"):
        r = json.loads(line)
        if r.get("category") == "Program and Code" and r.get("conversation"):
            out[r["conversation_id"]] = r["conversation"]
    return out


zh = load("computer_zh_26k(fixed).jsonl")
en = load("computer_en_26k(fixed).jsonl")
both = sorted(
    cid for cid in zh.keys() & en.keys()
    if ":" not in cid and len(zh[cid]) == len(en[cid])
)
print(f"{len(zh)} zh, {len(en)} en, {len(both)} parallel with matching turn counts")
ids = random.Random(seed).sample(both, n)

for lang, convs in (("zh", zh), ("en", en)):
    d = Path("corpus") / lang
    d.mkdir(parents=True, exist_ok=True)
    for i, cid in enumerate(ids):
        ts = f"2026-01-{1 + i % 28:02d}T{(i // 28) % 24:02d}:00:00Z"
        lines, seq = [], 0
        for pair in convs[cid]:
            for role, key in (("user", "human"), ("assistant", "assistant")):
                text = pair.get(key) or ""
                lines.append(json.dumps({
                    "session_id": cid,
                    "turn_uuid": f"t{seq}",
                    "seq": seq,
                    "ts": ts,
                    "role": role,
                    "harness": "sharegpt",
                    "blocks": [{"block_type": "text", "text": text}] if text else [],
                }, ensure_ascii=False))
                seq += 1
        (d / f"{cid}.funes.jsonl").write_text("\n".join(lines) + "\n", encoding="utf-8")

Path("corpus/ids.txt").write_text("\n".join(ids) + "\n")
print(f"wrote {n} conversations per language")