gh-84508: tool to generate cjk traditional chinese mappings (gh-93272)

Author: sorcio

Modules/cjkcodecs/mappings_hk.h

@@ -1,3 +1,4 @@
+// AUTO-GENERATED FILE FROM genmap_tchinese.py: DO NOT EDIT
 static const ucs2_t __big5hkscs_decmap[6219] = {
 17392,19506,17923,17830,17784,29287,19831,17843,31921,19682,31941,15253,18230,
 18244,19527,19520,17087,13847,29522,28299,28882,19543,41809,18255,17882,19589,

Modules/cjkcodecs/mappings_tw.h

@@ -1,3 +1,4 @@
+// AUTO-GENERATED FILE FROM genmap_tchinese.py: DO NOT EDIT
 static const ucs2_t __big5_decmap[16702] = {
 12288,65292,12289,12290,65294,8226,65307,65306,65311,65281,65072,8230,8229,
 65104,65380,65106,183,65108,65109,65110,65111,65372,8211,65073,8212,65075,
@@ -2631,3 +2632,4 @@ static const struct unim_index cp950ext_encmap[256] = {
 0,0},{0,0,0},{0,0,0},{0,0,0},{0,0,0},{0,0,0},{__cp950ext_encmap+342,81,104},{
 __cp950ext_encmap+366,15,229},
 };
+

Tools/unicode/genmap_tchinese.py

@@ -0,0 +1,239 @@
+#
+# genmap_tchinese.py: Traditional Chinese Codecs Map Generator
+#
+# Original Author:  Hye-Shik Chang <perky@FreeBSD.org>
+#
+import os
+
+from genmap_support import *
+
+
+# ranges for (lead byte, follower byte)
+BIG5_C1 = (0xa1, 0xfe)
+BIG5_C2 = (0x40, 0xfe)
+BIG5HKSCS_C1 = (0x87, 0xfe)
+BIG5HKSCS_C2 = (0x40, 0xfe)
+
+MAPPINGS_BIG5 = 'https://unicode.org/Public/MAPPINGS/OBSOLETE/EASTASIA/OTHER/BIG5.TXT'
+MAPPINGS_CP950 = 'https://www.unicode.org/Public/MAPPINGS/VENDORS/MICSFT/WINDOWS/CP950.TXT'
+
+HKSCS_VERSION = '2004'
+# The files for HKSCS mappings are available under a restrictive license.
+# Users of the script need to download the files from the HKSARG CCLI website:
+MAPPINGS_HKSCS = f'https://www.ccli.gov.hk/en/archive/terms_hkscs-{HKSCS_VERSION}-big5-iso.html'
+
+
+def bh2s(code):
+    return ((code >> 8) - 0x87) * (0xfe - 0x40 + 1) + ((code & 0xff) - 0x40)
+
+
+def split_bytes(code):
+    """Split 0xABCD into 0xAB, 0xCD"""
+    return code >> 8, code & 0xff
+
+
+def parse_hkscs_map(fo):
+    fo.seek(0, 0)
+    table = []
+    for line in fo:
+        line = line.split('#', 1)[0].strip()
+        # We expect 4 columns in supported HKSCS files:
+        # [1999]: unsupported
+        # [2001]: unsupported
+        # [2004]: Big-5; iso10646-1:1993; iso10646-1:2000; iso10646:2003+amd1
+        # [2008]: Big-5; iso10646-1:1993; iso10646-1:2000; iso10646:2003+amd6
+        # [2016]: not supported here--uses a json file instead
+        #
+        # In both supported cases, we only need the first and last column:
+        #  * Big-5 is a hex string (always 4 digits)
+        #  * iso10646:2003 is either a hex string (4 or 5 digits) or a sequence
+        #    of hex strings like: `<code_point1,code_point2>`
+        try:
+            hkscs_col, _, _, uni_col = line.split()
+            hkscs = int(hkscs_col, 16)
+            seq = tuple(int(cp, 16) for cp in uni_col.strip('<>').split(','))
+        except ValueError:
+            continue
+        table.append((hkscs, seq))
+    return table
+
+
+def make_hkscs_map(table):
+    decode_map = {}
+    encode_map_bmp, encode_map_notbmp = {}, {}
+    is_bmp_map = {}
+    sequences = []
+    beginnings = {}
+    single_cp_table = []
+    # Determine multi-codepoint sequences, and sequence beginnings that encode
+    # multiple multibyte (i.e. Big-5) codes.
+    for mbcode, cp_seq in table:
+        cp, *_ = cp_seq
+        if len(cp_seq) == 1:
+            single_cp_table.append((mbcode, cp))
+        else:
+            sequences.append((mbcode, cp_seq))
+        beginnings.setdefault(cp, []).append(mbcode)
+    # Decode table only cares about single code points (no sequences) currently
+    for mbcode, cp in single_cp_table:
+        b1, b2 = split_bytes(mbcode)
+        decode_map.setdefault(b1, {})
+        decode_map[b1][b2] = cp & 0xffff
+    # Encode table needs to mark code points beginning a sequence as tuples.
+    for cp, mbcodes in beginnings.items():
+        plane = cp >> 16
+        if plane == 0:
+            encode_map = encode_map_bmp
+        elif plane == 2:
+            encode_map = encode_map_notbmp
+            is_bmp_map[bh2s(mbcodes[0])] = 1
+        else:
+            assert False, 'only plane 0 (BMP) and plane 2 (SIP) allowed'
+        if len(mbcodes) == 1:
+            encode_value = mbcodes[0]
+        else:
+            encode_value = tuple(mbcodes)
+        uni_b1, uni_b2 = split_bytes(cp & 0xffff)
+        encode_map.setdefault(uni_b1, {})
+        encode_map[uni_b1][uni_b2] = encode_value
+
+    return decode_map, encode_map_bmp, encode_map_notbmp, is_bmp_map
+
+
+def load_big5_map():
+    mapfile = open_mapping_file('python-mappings/BIG5.txt', MAPPINGS_BIG5)
+    with mapfile:
+        big5decmap = loadmap(mapfile)
+    # big5 mapping fix: use the cp950 mapping for these characters as the file
+    # provided by unicode.org doesn't define a mapping. See notes in BIG5.txt.
+    # Since U+5341, U+5345, U+FF0F, U+FF3C already have a big5 mapping, no
+    # roundtrip compatibility is guaranteed for those.
+    for m in """\
+    0xA15A      0x2574
+    0xA1C3      0xFFE3
+    0xA1C5      0x02CD
+    0xA1FE      0xFF0F
+    0xA240      0xFF3C
+    0xA2CC      0x5341
+    0xA2CE      0x5345""".splitlines():
+        bcode, ucode = list(map(eval, m.split()))
+        big5decmap[bcode >> 8][bcode & 0xff] = ucode
+    # encoding map
+    big5encmap = {}
+    for c1, m in list(big5decmap.items()):
+        for c2, code in list(m.items()):
+            big5encmap.setdefault(code >> 8, {})
+            if code & 0xff not in big5encmap[code >> 8]:
+                big5encmap[code >> 8][code & 0xff] = c1 << 8 | c2
+    # fix unicode->big5 priority for the above-mentioned duplicate characters
+    big5encmap[0xFF][0x0F] = 0xA241
+    big5encmap[0xFF][0x3C] = 0xA242
+    big5encmap[0x53][0x41] = 0xA451
+    big5encmap[0x53][0x45] = 0xA4CA
+
+    return big5decmap, big5encmap
+
+
+def load_cp950_map():
+    mapfile = open_mapping_file('python-mappings/CP950.TXT', MAPPINGS_CP950)
+    with mapfile:
+        cp950decmap = loadmap(mapfile)
+    cp950encmap = {}
+    for c1, m in list(cp950decmap.items()):
+        for c2, code in list(m.items()):
+            cp950encmap.setdefault(code >> 8, {})
+            if code & 0xff not in cp950encmap[code >> 8]:
+                cp950encmap[code >> 8][code & 0xff] = c1 << 8 | c2
+    # fix unicode->big5 duplicated mapping priority
+    cp950encmap[0x53][0x41] = 0xA451
+    cp950encmap[0x53][0x45] = 0xA4CA
+    return cp950decmap, cp950encmap
+
+
+def main_tw():
+    big5decmap, big5encmap = load_big5_map()
+    cp950decmap, cp950encmap = load_cp950_map()
+
+    # CP950 extends Big5, and the codec can use the Big5 lookup tables
+    # for most entries. So the CP950 tables should only include entries
+    # that are not in Big5:
+    for c1, m in list(cp950encmap.items()):
+        for c2, code in list(m.items()):
+            if (c1 in big5encmap and c2 in big5encmap[c1]
+                    and big5encmap[c1][c2] == code):
+                del cp950encmap[c1][c2]
+    for c1, m in list(cp950decmap.items()):
+        for c2, code in list(m.items()):
+            if (c1 in big5decmap and c2 in big5decmap[c1]
+                    and big5decmap[c1][c2] == code):
+                del cp950decmap[c1][c2]
+
+    with open('mappings_tw.h', 'w') as fp:
+        print_autogen(fp, os.path.basename(__file__))
+        write_big5_maps(fp, 'BIG5', 'big5', big5decmap, big5encmap)
+        write_big5_maps(fp, 'CP950', 'cp950ext', cp950decmap, cp950encmap)
+
+
+def write_big5_maps(fp, display_name, table_name, decode_map, encode_map):
+    print(f'Generating {display_name} decode map...')
+    writer = DecodeMapWriter(fp, table_name, decode_map)
+    writer.update_decode_map(BIG5_C1, BIG5_C2)
+    writer.generate()
+    print(f'Generating {display_name} encode map...')
+    writer = EncodeMapWriter(fp, table_name, encode_map)
+    writer.generate()
+
+
+class HintsWriter:
+    filler_class = BufferedFiller
+
+    def __init__(self, fp, prefix, isbmpmap):
+        self.fp = fp
+        self.prefix = prefix
+        self.isbmpmap = isbmpmap
+        self.filler = self.filler_class()
+
+    def fillhints(self, hintfrom, hintto):
+        name = f'{self.prefix}_phint_{hintfrom}'
+        self.fp.write(f'static const unsigned char {name}[] = {{\n')
+        for msbcode in range(hintfrom, hintto+1, 8):
+            v = 0
+            for c in range(msbcode, msbcode+8):
+                v |= self.isbmpmap.get(c, 0) << (c - msbcode)
+            self.filler.write('%d,' % v)
+        self.filler.printout(self.fp)
+        self.fp.write('};\n\n')
+
+
+def main_hkscs():
+    filename = f'python-mappings/hkscs-{HKSCS_VERSION}-big5-iso.txt'
+    with open_mapping_file(filename, MAPPINGS_HKSCS) as f:
+        table = parse_hkscs_map(f)
+    hkscsdecmap, hkscsencmap_bmp, hkscsencmap_nonbmp, isbmpmap = (
+        make_hkscs_map(table)
+    )
+    with open('mappings_hk.h', 'w') as fp:
+        print('Generating BIG5HKSCS decode map...')
+        print_autogen(fp, os.path.basename(__file__))
+        writer = DecodeMapWriter(fp, 'big5hkscs', hkscsdecmap)
+        writer.update_decode_map(BIG5HKSCS_C1, BIG5HKSCS_C2)
+        writer.generate()
+
+        print('Generating BIG5HKSCS decode map Unicode plane hints...')
+        writer = HintsWriter(fp, 'big5hkscs', isbmpmap)
+        writer.fillhints(bh2s(0x8740), bh2s(0xa0fe))
+        writer.fillhints(bh2s(0xc6a1), bh2s(0xc8fe))
+        writer.fillhints(bh2s(0xf9d6), bh2s(0xfefe))
+
+        print('Generating BIG5HKSCS encode map (BMP)...')
+        writer = EncodeMapWriter(fp, 'big5hkscs_bmp', hkscsencmap_bmp)
+        writer.generate()
+
+        print('Generating BIG5HKSCS encode map (non-BMP)...')
+        writer = EncodeMapWriter(fp, 'big5hkscs_nonbmp', hkscsencmap_nonbmp)
+        writer.generate()
+
+
+if __name__ == '__main__':
+    main_tw()
+    main_hkscs()