"""英文文本归一化 (TN): 数字 / 货币 / 百分比 / 日期 / 年份 / 电话号码等 NSW -> 英文单词。

[导出包 vendored 版本] 与 bridge/text_norm.py 逻辑一致, FST 路径相对模型根目录
(env DUBBING_HOME 或 DUBBING_WETEXT 覆盖, 缺省取本文件上三级/wetext_en_tn)。

规则版本: wetext-0.0.4-en-tn-v1。实现: wetext 0.0.4 en TN (kaldifst FST, 本地文件,
不触网); wetext 不可用时回退 inflect 手写规则 (仅兜底, 打 warning)。
"""
import logging
import os
import re

RULE_VERSION = 'wetext-0.0.4-en-tn-v1'

_MODEL_DIR = os.environ.get('DUBBING_HOME') or os.path.abspath(
    os.path.join(os.path.dirname(__file__), '..', '..'))

# en TN FST (导出包 wetext_en_tn/; 可用 DUBBING_WETEXT 覆盖)
WETEXT_EN_TN = os.environ.get('DUBBING_WETEXT') or os.path.join(_MODEL_DIR, 'wetext_en_tn')

_NORMALIZER = None
_BACKEND = None

# ST LM 译文偶带的元标记: <|...|> 特殊 token、[...]/(...) 时间戳
_RE_SPECIAL = re.compile(r'<\|[^|]*\|>')
_RE_TS_BRACKET = re.compile(r'[\(\[]\s*\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?\s*[\)\]]')
_RE_WS = re.compile(r'\s+')


def _get_backend():
    """惰性加载 wetext en TN (本地 FST); 失败回退 inflect 规则。返回 'wetext'/'inflect'。"""
    global _NORMALIZER, _BACKEND
    if _BACKEND is None:
        try:
            from wetext import Normalizer
            tagger = os.path.join(WETEXT_EN_TN, 'tagger.fst')
            verbalizer = os.path.join(WETEXT_EN_TN, 'verbalizer.fst')
            if not (os.path.exists(tagger) and os.path.exists(verbalizer)):
                raise FileNotFoundError(WETEXT_EN_TN)
            _NORMALIZER = Normalizer(tagger_path=tagger, verbalizer_path=verbalizer, lang='en')
            _BACKEND = 'wetext'
        except Exception as e:
            logging.warning(f'text_norm: wetext unavailable ({type(e).__name__}: {e}), '
                            f'fallback to inflect rules')
            _BACKEND = 'inflect'
    return _BACKEND


_MONTHS = ['january', 'february', 'march', 'april', 'may', 'june', 'july',
           'august', 'september', 'october', 'november', 'december']


def _inflect_normalize(text):
    """兜底规则 (无 wetext 时): 货币 / 百分比 / ISO 日期 / 序数 / 小数 / 整数。子集, 仅兜底。"""
    import inflect
    p = inflect.engine()

    def int_words(s):
        return p.number_to_words(int(s)).replace('-', ' ').replace(',', '')

    def num_words(s):
        if '.' in s:
            i, f = s.split('.', 1)
            head = int_words(i) if i and int(i) else 'zero'
            return head + ' point ' + ' '.join(int_words(d) for d in f)
        return int_words(s)

    cur = {'$': 'dollars', '€': 'euros', '£': 'pounds'}
    text = re.sub(r'([$€£])\s*(\d+(?:\.\d+)?)',
                  lambda m: num_words(m.group(2)) + ' ' + cur[m.group(1)], text)
    text = re.sub(r'(\d+(?:\.\d+)?)\s*%',
                  lambda m: num_words(m.group(1)) + ' percent', text)
    text = re.sub(r'\b(\d{4})-(\d{2})-(\d{2})\b',
                  lambda m: f"the {p.ordinal(int_words(m.group(3)))} of "
                            f"{_MONTHS[int(m.group(2)) - 1]} {int_words(m.group(1))}", text)
    text = re.sub(r'\b(\d+)(st|nd|rd|th)\b', lambda m: p.ordinal(int_words(m.group(1))), text)
    text = re.sub(r'\b\d+\.\d+\b', lambda m: num_words(m.group(0)), text)
    text = re.sub(r'\b\d+\b', lambda m: num_words(m.group(0)), text)
    return text


def strip_meta(text):
    """剥掉 ST LM 输出里可能夹带的特殊 token / 时间戳标记。"""
    text = _RE_SPECIAL.sub(' ', text)
    text = _RE_TS_BRACKET.sub(' ', text)
    return _RE_WS.sub(' ', text).strip()


def normalize_en(text):
    """英文译文归一化入口。幂等 (normalize_en(normalize_en(x)) == normalize_en(x))。"""
    text = strip_meta(text)
    if not text:
        return text
    if _get_backend() == 'wetext':
        out = _NORMALIZER.normalize(text)
    else:
        out = _inflect_normalize(text)
    return _RE_WS.sub(' ', out).strip()