util.py 文件源码

def check_spelling(spelling_lang, txt):
    """
    Check the spelling in the text, and compute a score. The score is the
    number of words correctly (or almost correctly) spelled, minus the number
    of mispelled words. Words "almost" correct remains neutral (-> are not
    included in the score)

    Returns:
        A tuple : (fixed text, score)
    """
    if os.name == "nt":
        assert(not "check_spelling() not available on Windows")
        return
    with _ENCHANT_LOCK:
        # Maximum distance from the first suggestion from python-enchant

        words_dict = enchant.request_dict(spelling_lang)
        try:
            tknzr = enchant.tokenize.get_tokenizer(spelling_lang)
        except enchant.tokenize.TokenizerNotFoundError:
            # Fall back to default tokenization if no match for 'lang'
            tknzr = enchant.tokenize.get_tokenizer()

        score = 0
        offset = 0
        for (word, word_pos) in tknzr(txt):
            if len(word) < _MIN_WORD_LEN:
                continue
            if words_dict.check(word):
                # immediately correct words are a really good hint for
                # orientation
                score += 100
                continue
            suggestions = words_dict.suggest(word)
            if (len(suggestions) <= 0):
                # this word is useless. It may even indicates a bad orientation
                score -= 10
                continue
            main_suggestion = suggestions[0]
            lv_dist = Levenshtein.distance(word, main_suggestion)
            if (lv_dist > _MAX_LEVENSHTEIN_DISTANCE):
                # hm, this word looks like it's in a bad shape
                continue

            logger.debug("Spell checking: Replacing: %s -> %s"
                         % (word, main_suggestion))

            # let's replace the word by its suggestion

            pre_txt = txt[:word_pos + offset]
            post_txt = txt[word_pos + len(word) + offset:]
            txt = pre_txt + main_suggestion + post_txt
            offset += (len(main_suggestion) - len(word))

            # fixed words may be a good hint for orientation
            score += 5

        return (txt, score)