python全角⇔半角文字列変換

strのtranslate関数の場合

ただし半角カナは非対応

ZEN = chr(0x3000) + "".join(chr(0xff01 + i) for i in range(94))
HAN = "".join(chr(0x20 + i) for i in range(95))
def to_hankaku(s, tdic=str.maketrans(ZEN, HAN)):
    return s.translate(tdic)
def to_zenkaku(s, tdic=str.maketrans(HAN, ZEN)):
    return s.translate(tdic)

結構速い

In [1]: %timeit to_hankaku("０１２３４５６７８９ABCＤＥＦ")
   ...: %timeit to_zenkaku("0123456789ＡＢＣDEF")
   ...:
1.09 µs ± 8.6 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
911 ns ± 2.09 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

C言語実装pythonライブラリにした場合

str.translateでやった場合より10倍速い。上記のmaketrans translateではカナエリアに非対応だが、
興味本位でカナ、半角カナもついでに対応表を作って実装してみたら遅くなるかと思ったが、
微塵も変化なかった。

#include <unordered_map>

std::unordered_map<char32_t, char32_t*> ZEN2HAN = {
    {U'\x30A1', U"\xff67\x00"},        //ァ
    {U'\x30A2', U"\xff71\x00"},        //ア
    {U'\x30A3', U"\xff68\x00"},        //ィ
    {U'\x30A4', U"\xff72\x00"},        //イ
    {U'\x30A5', U"\xff69\x00"},        //ゥ
    {U'\x30A6', U"\xff73\x00"},        //ウ
    {U'\x30A7', U"\xff6A\x00"},        //ェ
    {U'\x30A8', U"\xff74\x00"},        //エ
    {U'\x30A9', U"\xff6B\x00"},        //ォ
    {U'\x30AA', U"\xff75\x00"},        //オ
    {U'\x30AB', U"\xff76\x00"},        //カ
    {U'\x30AC', U"\xff76\xFF9E\x00"},  //ガ
    {U'\x30AD', U"\xff77\x00"},        //キ
    {U'\x30AE', U"\xff77\xFF9E\x00"},  //ギ
    {U'\x30AF', U"\xff78\x00"},        //ク
    {U'\x30B0', U"\xff78\xFF9E\x00"},  //グ
    {U'\x30B1', U"\xff79\x00"},        //ケ
    {U'\x30B2', U"\xff79\xFF9E\x00"},  //ゲ
    {U'\x30B3', U"\xff7A\x00"},        //コ
    {U'\x30B4', U"\xff7A\xFF9E\x00"},  //ゴ
    {U'\x30B5', U"\xff7B\x00"},        //サ
    {U'\x30B6', U"\xff7B\xFF9E\x00"},  //ザ
    {U'\x30B7', U"\xff7C\x00"},        //シ
    {U'\x30B8', U"\xff7C\xFF9E\x00"},  //ジ
    {U'\x30B9', U"\xff7D\x00"},        //ス
    {U'\x30BA', U"\xff7D\xFF9E\x00"},  //ズ
    {U'\x30BB', U"\xff7E\x00"},        //セ
    {U'\x30BC', U"\xff7E\xFF9E\x00"},  //ゼ
    {U'\x30BD', U"\xff7F\x00"},        //ソ
    {U'\x30BE', U"\xff7F\xFF9E\x00"},  //ゾ
    {U'\x30BF', U"\xff80\x00"},        //タ
    {U'\x30C0', U"\xff80\xFF9E\x00"},  //ダ
    {U'\x30C1', U"\xff81\x00"},        //チ
    {U'\x30C2', U"\xff81\xFF9E\x00"},  //ヂ
    {U'\x30C3', U"\xff6F\x00"},        //ッ
    {U'\x30C4', U"\xff82\x00"},        //ツ
    {U'\x30C5', U"\xff82\xFF9E\x00"},  //ヅ
    {U'\x30C6', U"\xff83\x00"},        //テ
    {U'\x30C7', U"\xff83\xFF9E\x00"},  //デ
    {U'\x30C8', U"\xff84\x00"},        //ト
    {U'\x30C9', U"\xff84\xFF9E\x00"},  //ド
    {U'\x30CA', U"\xff85\x00"},        //ナ
    {U'\x30CB', U"\xff86\x00"},        //ニ
    {U'\x30CC', U"\xff87\x00"},        //ヌ
    {U'\x30CD', U"\xff88\x00"},        //ネ
    {U'\x30CE', U"\xff89\x00"},        //ノ
    {U'\x30CF', U"\xff8A\x00"},        //ハ
    {U'\x30D0', U"\xff8A\xFF9E\x00"},  //バ
    {U'\x30D1', U"\xff8A\xFF9F\x00"},  //パ
    {U'\x30D2', U"\xff8B\x00"},        //ヒ
    {U'\x30D3', U"\xff8B\xFF9E\x00"},  //ビ
    {U'\x30D4', U"\xff8B\xFF9F\x00"},  //ピ
    {U'\x30D5', U"\xff8C\x00"},        //フ
    {U'\x30D6', U"\xff8C\xFF9E\x00"},  //ブ
    {U'\x30D7', U"\xff8C\xFF9F\x00"},  //プ
    {U'\x30D8', U"\xff8D\x00"},        //ヘ
    {U'\x30D9', U"\xff8D\xFF9E\x00"},  //ベ
    {U'\x30DA', U"\xff8D\xFF9F\x00"},  //ペ
    {U'\x30DB', U"\xff8E\x00"},        //ホ
    {U'\x30DC', U"\xff8E\xFF9E\x00"},  //ボ
    {U'\x30DD', U"\xff8E\xFF9F\x00"},  //ポ
    {U'\x30DE', U"\xff8F\x00"},        //マ
    {U'\x30DF', U"\xFF90\x00"},        //ミ
    {U'\x30E0', U"\xFF91\x00"},        //ム
    {U'\x30E1', U"\xFF92\x00"},        //メ
    {U'\x30E2', U"\xFF93\x00"},        //モ
    {U'\x30E3', U"\xff6C\x00"},        //ャ
    {U'\x30E4', U"\xFF94\x00"},        //ヤ
    {U'\x30E5', U"\xff6D\x00"},        //ュ
    {U'\x30E6', U"\xFF95\x00"},        //ユ
    {U'\x30E7', U"\xff6E\x00"},        //ョ
    {U'\x30E8', U"\xFF96\x00"},        //ヨ
    {U'\x30E9', U"\xFF97\x00"},        //ラ
    {U'\x30EA', U"\xFF98\x00"},        //リ
    {U'\x30EB', U"\xFF99\x00"},        //ル
    {U'\x30EC', U"\xFF9A\x00"},        //レ
    {U'\x30ED', U"\xFF9B\x00"},        //ロ
    {U'\x30EF', U"\xFF9C\x00"},        //ワ
    {U'\x30F2', U"\xff66\x00"},        //ヲ
    {U'\x30F3', U"\xFF9D\x00"},        //ン
    {U'\x30F4', U"\xff73\xFF9E\x00"},  //ヴ
    {U'\x30FB', U"\xff65\x00"},        //・
    {U'\x30FC', U"\xff70\x00"},        //ー
    {U'\x3001', U"\xff64\x00"},        //、
    {U'\x300C', U"\xff62\x00"},        //「
    {U'\x300D', U"\xff63\x00"},        //」
    {U'\x309B', U"\xFF9E\x00"},        //゛
    {U'\x309C', U"\xFF9F\x00"},        //゜
};

const char32_t han2zen[62] = {U'\x30C4', U'\x30C6', U'\x30C8', U'\x30CA', U'\x30CB', U'\x30CC', U'\x30CD', U'\x30CE',
                              U'\x30CF', U'\x30D2', U'\x30D5', U'\x30D8', U'\x30DB', U'\x30DE', U'\x30DF', U'\x30E0',
                              U'\x30E1', U'\x30E2', U'\x30E4', U'\x30E6', U'\x30E8', U'\x30E9', U'\x30EA', U'\x30EB',
                              U'\x30EC', U'\x30ED', U'\x30EF', U'\x30F3', U'\x309B', U'\x309C', U'\x300C', U'\x300D',
                              U'\x3001', U'\x30FB', U'\x30F2', U'\x30A1', U'\x30A3', U'\x30A5', U'\x30A7', U'\x30A9',
                              U'\x30E3', U'\x30E5', U'\x30E7', U'\x30C3', U'\x30FC', U'\x30A2', U'\x30A4', U'\x30A6',
                              U'\x30A8', U'\x30AA', U'\x30AB', U'\x30AD', U'\x30AF', U'\x30B1', U'\x30B3', U'\x30B5',
                              U'\x30B7', U'\x30B9', U'\x30BB', U'\x30BD', U'\x30BF', U'\x30C1'};


static PyObject* to_hankaku(PyObject* str) {
    std::size_t len = (std::size_t)PyObject_Length(str);
    if(len == (std::size_t)-1 || PyUnicode_READY(str) == -1)
        return NULL;

    unsigned int kind = PyUnicode_KIND(str);
    if(kind == 1)
        return str;

    void* data = PyUnicode_DATA(str);
    PyObject* res = PyUnicode_New((Py_ssize_t)len, 0x10ffff);
    void* ret = PyUnicode_DATA(res);
    std::size_t i = 0, j = 0;
    char32_t writer;
    for(; i < len; ++i, ++j) {
        auto s = PyUnicode_READ(kind, data, i);
        writer = NULL;

        if(s > 0xff00 && s < 0xff5f)
            writer = (0x20 + (s % 0xff));
        else if(s == 0x3000)
            writer = 0x20;
        else if(s > 0x3098 && s < 0x30FD) {
            auto r = ZEN2HAN[s];
            if(r && r[1] != 0x00) { //半濁音、濁音付け
                PyUnicode_WRITE(4, ret, j++, r[0]);
                writer = r[1];
            } else if(r[0] != 0x00) {
                writer = r[0];
            }
        }

        PyUnicode_WRITE(4, ret, j, writer ? writer : s);
    }
    return res;
}

static PyObject* to_zenkaku(PyObject* str) {
    std::size_t len = (std::size_t)PyObject_Length(str);
    if(len == (std::size_t)-1 || PyUnicode_READY(str) == -1)
        return NULL;

    unsigned int kind = PyUnicode_KIND(str);
    void* data = PyUnicode_DATA(str);
    PyObject* res = PyUnicode_New((Py_ssize_t)len, 0x10ffff);
    void* ret = PyUnicode_DATA(res);
    std::size_t i = 0, j = 0;
    char32_t writer;

    for(; i < len; ++i, ++j) {
        auto s = PyUnicode_READ(kind, data, i);
        writer = NULL;

        if(s > 0x20 && s < 0x7f)
            writer = s + 0xfee0;
        else if(s == 0x20)
            writer = 0x3000;
        else if(s > 0xff62 && s < 0xff9f) {
            writer = han2zen[(int)s % 62];

            if(s == 0xff73 || (s > 0xff75 && s < 0xff82) || (s > 0xff89 && s < 0xff8F)) {
                auto next = PyUnicode_READ(kind, data, i + 1);
                if(next == 0xFF9E || next == 0x309B)  // 濁音付け
                    ++writer, ++i;
                else if(next == 0xFF9F || next == 0x309C)  // 半濁音付け
                    ++++writer, ++i;
            }
        }

        PyUnicode_WRITE(4, ret, j, writer ? writer : s);
    }
    return res;
}

圧倒的に速いので自作pythonライブラリに追加しよう。

In [3]: %timeit to_hankaku("０１２３４５６７８９ABCＤＥＦ")
   ...: %timeit to_zenkaku("0123456789ＡＢＣDEF")
   ...:
87 ns ± 0.472 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
86 ns ± 0.322 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)