"""Clean captions and apply controlled noisy perturbations (Section 2.2.4 style)."""

from __future__ import annotations

import random
import re
from typing import Optional

# Small synonym table for fashion-ish tokens (extend as needed)
SYNONYMS: dict[str, list[str]] = {
    "slim": ["slim", "fitted", "narrow"],
    "long": ["long", "extended"],
    "short": ["short", "brief"],
    "dress": ["dress", "gown"],
    "sleeve": ["sleeve", "sleeves"],
}


def clean_caption(text: str) -> str:
    t = text.replace("\n", " ").replace("\r", " ")
    t = re.sub(r"\s+", " ", t).strip()
    t = re.sub(r"\s*,\s*", ", ", t)
    parts = [p.strip() for p in t.split(",") if p.strip()]
    return ", ".join(parts)


def _typo_noise(rng: random.Random, s: str, p: float) -> str:
    if len(s) < 2:
        return s
    out = list(s)
    for i in range(len(out)):
        if rng.random() < p and out[i].isalpha():
            op = rng.choice(["swap", "drop", "ins"])
            if op == "swap" and i < len(out) - 1:
                out[i], out[i + 1] = out[i + 1], out[i]
            elif op == "drop":
                out[i] = ""
            elif op == "ins":
                out[i] = out[i] + rng.choice("aeiou")
    return "".join(out)


def _token_dropout(rng: random.Random, caption: str, p: float) -> str:
    parts = [p.strip() for p in caption.split(",") if p.strip()]
    if len(parts) <= 1:
        return caption
    kept = [x for x in parts if rng.random() > p]
    if not kept:
        kept = [parts[0]]
    return ", ".join(kept)


def _truncate_tail(rng: random.Random, caption: str, p: float) -> str:
    parts = [p.strip() for p in caption.split(",") if p.strip()]
    if len(parts) <= 2 or rng.random() > p:
        return caption
    k = rng.randint(max(1, len(parts) // 2), len(parts) - 1)
    return ", ".join(parts[:k])


def _synonym_replace(rng: random.Random, caption: str) -> str:
    out = []
    for tok in re.split(r"(\W+)", caption):
        low = tok.lower()
        if low in SYNONYMS:
            out.append(rng.choice(SYNONYMS[low]))
        else:
            out.append(tok)
    return "".join(out)


def noisy_caption(
    clean: str,
    *,
    seed: int,
    typo_prob: float = 0.03,
    token_dropout_prob: float = 0.12,
    truncate_tail_prob: float = 0.08,
) -> str:
    rng = random.Random(seed + hash(clean) % (2**31))
    t = clean
    t = _truncate_tail(rng, t, truncate_tail_prob)
    t = _token_dropout(rng, t, token_dropout_prob)
    t = _synonym_replace(rng, t)
    t = _typo_noise(rng, t, typo_prob)
    t = clean_caption(t)
    return t if t else clean