private void FillLetterNFKD (int i, bool checkUpper, bool greekRemap)
{
if (map [i].Defined)
return;
int up = (int) ti.ToUpper ((char) i);
if (checkUpper && map [up].Category == 0xF) {
if (i == up)
return;
FillLetterNFKD (up, checkUpper, greekRemap);
map [i] = new CharMapEntry (0xF,
map [up].Level1,
map [up].Level2);
} else {
int idx = decompIndex [i];
if (idx == 0)
return;
int primary = decompValues [decompIndex [i]];
FillLetterNFKD (primary, checkUpper, greekRemap);
int lv2 = map [primary].Level2;
byte off = 0;
for (int l = 1; l < decompLength [i]; l++) {
int tmp = decompValues [idx + l];
if (map [tmp].Category != 1)
return;
if (greekRemap && map [tmp].Level2 == 0xC)
off += 3;
else
off += map [tmp].Level2;
}
if (off > 0) {
if (lv2 == 0)
lv2 += 2;
lv2 += off;
}
// ... but override if the value already exists.
if (diacritical [i] != 0)
lv2 = diacritical [i];
map [i] = new CharMapEntry (
map [primary].Category,
map [primary].Level1,
(byte) lv2);
}
}
private bool AddCharMap (char c, byte category, byte increment, byte alt)
{
if (IsIgnorable ((int) c) || map [(int) c].Defined)
return false; // do nothing
map [(int) c] = new CharMapEntry (category,
category == 1 ? alt : fillIndex [category],
category == 1 ? fillIndex [category] : alt);
fillIndex [category] += increment;
return true;
}
void GenerateCore ()
{
UnicodeCategory uc;
#region Specially ignored // 01
// This will raise "Defined" flag up.
// FIXME: Check If it is really fine. Actually for
// Japanese voice marks this code does remapping.
foreach (char c in specialIgnore)
map [(int) c] = new CharMapEntry (0, 0, 0);
#endregion
#region Extenders (FF FF)
fillIndex [0xFF] = 0xFF;
char [] specialBiggest = new char [] {
'\u3005', '\u3031', '\u3032', '\u309D',
'\u309E', '\u30FC', '\u30FD', '\u30FE',
'\uFE7C', '\uFE7D', '\uFF70'};
foreach (char c in specialBiggest)
AddCharMap (c, 0xFF, 0);
#endregion
#region Variable weights
// Controls : 06 03 - 06 3D
fillIndex [0x6] = 3;
for (int i = 0; i < 65536; i++) {
if (IsIgnorable (i))
continue;
char c = (char) i;
uc = Char.GetUnicodeCategory (c);
// NEL is whitespace but not ignored here.
if (uc == UnicodeCategory.Control &&
!Char.IsWhiteSpace (c) || c == '\u0085')
AddCharMap (c, 6, 1);
}
// Apostrophe 06 80
fillIndex [0x6] = 0x80;
AddCharMap ('\'', 6, 0);
AddCharMap ('\uFF07', 6, 1);
AddCharMap ('\uFE63', 6, 1);
// SPECIAL CASE: fill FE32 here in prior to be added
// at 2013. Windows does not always respect NFKD.
map [0xFE32] = new CharMapEntry (6, 0x90, 0);
// Hyphen/Dash : 06 81 - 06 90
for (int i = 0; i < char.MaxValue; i++) {
if (!IsIgnorable (i) &&
Char.GetUnicodeCategory ((char) i) ==
UnicodeCategory.DashPunctuation) {
AddCharMapGroup2 ((char) i, 6, 1, 0);
if (i == 0x2011) {
// SPECIAL: add 2027 and 2043
// Maybe they are regarded the
// same hyphens in "central"
// position.
AddCharMap ('\u2027', 6, 1);
AddCharMap ('\u2043', 6, 1);
}
}
}
// They are regarded as primarily equivalent to '-'
map [0x208B] = new CharMapEntry (6, 0x82, 0);
map [0x207B] = new CharMapEntry (6, 0x82, 0);
map [0xFF0D] = new CharMapEntry (6, 0x82, 0);
// Arabic variable weight chars 06 A0 -
fillIndex [6] = 0xA0;
// vowels
for (int i = 0x64B; i <= 0x650; i++)
AddArabicCharMap ((char) i, 6, 1, 0);
// sukun
AddCharMapGroup ('\u0652', 6, 1, 0);
// shadda
AddCharMapGroup ('\u0651', 6, 1, 0);
#endregion
#region Nonspacing marks // 01
// FIXME: 01 03 - 01 B6 ... annoyance :(
// Combining diacritical marks: 01 DC -
fillIndex [0x1] = 0x41;
for (int i = 0x030E; i <= 0x0326; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
for (int i = 0x0329; i <= 0x0334; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
fillIndex [0x1]++;
for (int i = 0x0339; i <= 0x0341; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
fillIndex [0x1] = 0x74;
for (int i = 0x0346; i <= 0x0348; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
for (int i = 0x02BE; i <= 0x02BF; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
for (int i = 0x02C1; i <= 0x02C5; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
for (int i = 0x02CE; i <= 0x02CF; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
fillIndex [0x1]++;
for (int i = 0x02D1; i <= 0x02D3; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
AddCharMap ('\u02DE', 0x1, 1);
for (int i = 0x02E4; i <= 0x02E9; i++)
if (!IsIgnorable (i))
AddCharMap ((char) i, 0x1, 1);
// FIXME: needs more love here (it should eliminate
// all the hacky code above).
for (int i = 0x0300; i < 0x0370; i++)
if (!IsIgnorable (i) && diacritical [i] != 0
&& !map [i].Defined)
map [i] = new CharMapEntry (
0x1, 0x1, diacritical [i]);
// Cyrillic and Armenian nonspacing mark
fillIndex [0x1] = 0x94;
for (int i = 0x400; i < 0x580; i++)
if (!IsIgnorable (i) &&
Char.GetUnicodeCategory ((char) i) ==
UnicodeCategory.NonSpacingMark)
AddCharMap ((char) i, 1, 1);
fillIndex [0x1] = 0x8D;
// syriac dotted nonspacing marks (1)
AddCharMap ('\u0740', 0x1, 1);
AddCharMap ('\u0741', 0x1, 1);
AddCharMap ('\u0742', 0x1, 1);
// syriac oblique nonspacing marks
AddCharMap ('\u0747', 0x1, 1);
AddCharMap ('\u0748', 0x1, 1);
// syriac dotted nonspacing marks (2)
fillIndex [0x1] = 0x94; // this reset is mandatory
AddCharMap ('\u0732', 0x1, 1);
AddCharMap ('\u0735', 0x1, 1);
AddCharMap ('\u0738', 0x1, 1);
AddCharMap ('\u0739', 0x1, 1);
AddCharMap ('\u073C', 0x1, 1);
// SPECIAL CASES: superscripts
AddCharMap ('\u073F', 0x1, 1);
AddCharMap ('\u0711', 0x1, 1);
// syriac "DOTS"
for (int i = 0x0743; i <= 0x0746; i++)
AddCharMap ((char) i, 0x1, 1);
for (int i = 0x0730; i <= 0x0780; i++)
if (!map [i].Defined &&
Char.GetUnicodeCategory ((char) i) ==
UnicodeCategory.NonSpacingMark)
AddCharMap ((char) i, 0x1, 1);
// LAMESPEC: It should not stop at '\u20E1'. There are
// a few more characters (that however results in
// overflow of level 2 unless we start before 0xDD).
fillIndex [0x1] = 0xDD;
for (int i = 0x20D0; i <= 0x20DC; i++)
AddCharMap ((char) i, 0x1, 1);
fillIndex [0x1] = 0xEC;
for (int i = 0x20DD; i <= 0x20E1; i++)
AddCharMap ((char) i, 0x1, 1);
fillIndex [0x1] = 0x4;
AddCharMap ('\u0CD5', 0x1, 1);
AddCharMap ('\u0CD6', 0x1, 1);
AddCharMap ('\u093C', 0x1, 1);
for (int i = 0x302A; i <= 0x302D; i++)
AddCharMap ((char) i, 0x1, 1);
AddCharMap ('\u0C55', 0x1, 1);
AddCharMap ('\u0C56', 0x1, 1);
fillIndex [0x1] = 0x50; // I wonder how they are sorted
for (int i = 0x02D4; i <= 0x02D7; i++)
AddCharMap ((char) i, 0x1, 1);
// They are not part of Nonspacing marks, but have
// only diacritical weight.
for (int i = 0x3099; i <= 0x309C; i++)
map [i] = new CharMapEntry (1, 1, 1);
map [0xFF9E] = new CharMapEntry (1, 1, 1);
map [0xFF9F] = new CharMapEntry (1, 1, 2);
map [0x309D] = new CharMapEntry (0xFF, 0xFF, 1);
map [0x309E] = new CharMapEntry (0xFF, 0xFF, 1);
for (int i = 0x30FC; i <= 0x30FE; i++)
map [i] = new CharMapEntry (0xFF, 0xFF, 1);
fillIndex [0x1] = 0xA;
for (int i = 0x0951; i <= 0x0954; i++)
AddCharMap ((char) i, 0x1, 2);
#endregion
#region Whitespaces // 07 03 -
fillIndex [0x7] = 0x2;
AddCharMap (' ', 0x7, 2);
AddCharMap ('\u00A0', 0x7, 1);
for (int i = 9; i <= 0xD; i++)
AddCharMap ((char) i, 0x7, 1);
for (int i = 0x2000; i <= 0x200B; i++)
AddCharMap ((char) i, 0x7, 1);
fillIndex [0x7] = 0x17;
AddCharMapGroup ('\u2028', 0x7, 1, 0);
AddCharMapGroup ('\u2029', 0x7, 1, 0);
// Characters which used to represent layout control.
// LAMESPEC: Windows developers seem to have thought
// that those characters are kind of whitespaces,
// while they aren't.
AddCharMap ('\u2422', 0x7, 1, 0); // blank symbol
AddCharMap ('\u2423', 0x7, 1, 0); // open box
#endregion
// category 09 - continued symbols from 08
fillIndex [0x9] = 2;
// misc tech mark
for (int cp = 0x2300; cp <= 0x237A; cp++)
AddCharMap ((char) cp, 0x9, 1, 0);
// arrows
byte [] arrowLv2 = new byte [] {0, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3};
foreach (DictionaryEntry de in arrowValues) {
int idx = (int) de.Value;
int cp = (int) de.Key;
if (map [cp].Defined)
continue;
fillIndex [0x9] = (byte) (0xD8 + idx);
AddCharMapGroup ((char) cp, 0x9, 0, arrowLv2 [idx]);
arrowLv2 [idx]++;
}
// boxes
byte [] boxLv2 = new byte [128];
// 0-63 will be used for those offsets are positive,
// and 64-127 are for negative ones.
for (int i = 0; i < boxLv2.Length; i++)
boxLv2 [i] = 3;
foreach (DictionaryEntry de in boxValues) {
int cp = (int) de.Key;
int off = (int) de.Value;
if (map [cp].Defined)
continue;
if (off < 0) {
fillIndex [0x9] = (byte) (0xE5 + off);
AddCharMapGroup ((char) cp, 0x9, 0, boxLv2 [128 + off]++);
}
else {
fillIndex [0x9] = (byte) (0xE5 + off);
AddCharMapGroup ((char) cp, 0x9, 0, boxLv2 [off]++);
}
}
// Some special characters (slanted)
fillIndex [0x9] = 0xF4;
AddCharMap ('\u2571', 0x9, 3);
AddCharMap ('\u2572', 0x9, 3);
AddCharMap ('\u2573', 0x9, 3);
// FIXME: implement 0A
#region Symbols
fillIndex [0xA] = 2;
// byte currency symbols
for (int cp = 0; cp < 0x100; cp++) {
uc = Char.GetUnicodeCategory ((char) cp);
if (!IsIgnorable (cp) &&
uc == UnicodeCategory.CurrencySymbol &&
cp != '$')
AddCharMapGroup ((char) cp, 0xA, 1, 0);
}
// byte other symbols
for (int cp = 0; cp < 0x100; cp++) {
if (cp == 0xA6)
continue; // SPECIAL: skip FIXME: why?
uc = Char.GetUnicodeCategory ((char) cp);
if (!IsIgnorable (cp) &&
uc == UnicodeCategory.OtherSymbol ||
cp == '\u00AC' || cp == '\u00B5' || cp == '\u00B7')
AddCharMapGroup ((char) cp, 0xA, 1, 0);
}
// U+30FB here
AddCharMapGroup ('\u30FB', 0xA, 1, 0);
for (int cp = 0x2020; cp <= 0x2031; cp++)
if (Char.IsPunctuation ((char) cp))
AddCharMap ((char) cp, 0xA, 1, 0);
// SPECIAL CASES: why?
AddCharMap ('\u203B', 0xA, 1, 0);
AddCharMap ('\u2040', 0xA, 1, 0);
AddCharMap ('\u2041', 0xA, 1, 0);
AddCharMap ('\u2042', 0xA, 1, 0);
for (int cp = 0x20A0; cp <= 0x20AB; cp++)
AddCharMap ((char) cp, 0xA, 1, 0);
// 3004 is skipped at first...
for (int cp = 0x3010; cp <= 0x3040; cp++)
if (Char.IsSymbol ((char) cp))
AddCharMap ((char) cp, 0xA, 1, 0);
// SPECIAL CASES: added here
AddCharMap ('\u3004', 0xA, 1, 0);
AddCharMap ('\u327F', 0xA, 1, 0);
for (int cp = 0x2600; cp <= 0x2613; cp++)
AddCharMap ((char) cp, 0xA, 1, 0);
// Dingbats
for (int cp = 0x2620; cp <= 0x2770; cp++)
if (Char.IsSymbol ((char) cp))
AddCharMap ((char) cp, 0xA, 1, 0);
// OCR
for (int i = 0x2440; i < 0x2460; i++)
AddCharMap ((char) i, 0xA, 1, 0);
// SPECIAL CASES: why?
AddCharMap ('\u0E3F', 0xA, 1, 0);
AddCharMap ('\u2117', 0xA, 1, 0);
AddCharMap ('\u20AC', 0xA, 1, 0);
#endregion
#region Numbers // 0C 02 - 0C E1
fillIndex [0xC] = 2;
// 9F8 : Bengali "one less than the denominator"
AddCharMap ('\u09F8', 0xC, 1, 0x3C);
ArrayList numbers = new ArrayList ();
for (int i = 0; i < 65536; i++)
if (!IsIgnorable (i) &&
Char.IsNumber ((char) i) &&
(i < 0x3190 || 0x32C0 < i)) // they are CJK characters
numbers.Add (i);
ArrayList numberValues = new ArrayList ();
foreach (int i in numbers)
numberValues.Add (new DictionaryEntry (i, decimalValue [(char) i]));
// SPECIAL CASE: Cyrillic Thousand sign
numberValues.Add (new DictionaryEntry (0x0482, 1000m));
numberValues.Sort (DecimalDictionaryValueComparer.Instance);
//foreach (DictionaryEntry de in numberValues)
//Console.Error.WriteLine ("****** number {0:X04} : {1} {2}", de.Key, de.Value, decompType [(int) de.Key]);
// FIXME: fillIndex adjustment lines are too
// complicated. It must be simpler.
decimal prevValue = -1;
foreach (DictionaryEntry de in numberValues) {
int cp = (int) de.Key;
decimal currValue = (decimal) de.Value;
bool addnew = false;
if (prevValue < currValue &&
prevValue - (int) prevValue == 0 &&
prevValue >= 1) {
addnew = true;
// Process Hangzhou and Roman numbers
// There are some SPECIAL cases.
if (currValue != 4) // no increment for 4
fillIndex [0xC]++;
int xcp;
if (currValue <= 13) {
if (currValue == 4)
fillIndex [0xC]++;
// SPECIAL CASE
if (currValue == 11)
AddCharMap ('\u0BF0', 0xC, 1);
xcp = (int) prevValue + 0x2160 - 1;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
xcp = (int) prevValue + 0x2170 - 1;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
fillIndex [0xC]++;
}
if (currValue < 12)
fillIndex [0xC]++;
if (currValue <= 10) {
xcp = (int) prevValue + 0x3021 - 1;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
fillIndex [0xC]++;
}
}
if (prevValue < currValue)
prevValue = currValue;
if (map [cp].Defined)
continue;
// HangZhou and Roman are add later
// (code is above)
if (0x3021 <= cp && cp < 0x302A
|| 0x2160 <= cp && cp < 0x216C
|| 0x2170 <= cp && cp < 0x217C)
continue;
if (cp == 0x215B) // FIXME: why?
fillIndex [0xC] += 2;
else if (cp == 0x3021) // FIXME: why?
fillIndex [0xC]++;
if (addnew || cp <= '9') {
int mod = (int) currValue - 1;
int xcp;
if (1 <= currValue && currValue <= 11) {
xcp = mod + 0x2776;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
xcp = mod + 0x2780;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
xcp = mod + 0x278A;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
}
if (1 <= currValue && currValue <= 20) {
xcp = mod + 0x2460;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
xcp = mod + 0x2474;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
xcp = mod + 0x2488;
AddCharMap ((char) xcp, 0xC, 0, diacritical [xcp]);
}
}
if (addnew && currValue >= 10 && currValue < 13 || cp == 0x09F9)
fillIndex [0xC]++;
AddCharMapGroup ((char) cp, 0xC, 0, diacritical [cp], true);
switch (cp) {
// Maybe Bengali digit numbers do not increase
// indexes, but 0x09E6 does.
case 0x09E7: case 0x09E8: case 0x09E9:
case 0x09EA:
// SPECIAL CASES
case 0x0BF0: case 0x2180: case 0x2181:
break;
// SPECIAL CASE
case 0x0BF1:
fillIndex [0xC]++;
break;
default:
if (currValue < 11 || currValue == 1000)
fillIndex [0xC]++;
break;
}
// Add special cases that are not regarded as
// numbers in UnicodeCategory speak.
if (cp == '5') {
// TONE FIVE
AddCharMapGroup ('\u01BD', 0xC, 0, 0);
AddCharMapGroup ('\u01BC', 0xC, 1, 0);
}
else if (cp == '2' || cp == '6') // FIXME: why?
fillIndex [0xC]++;
}
// 221E: infinity
fillIndex [0xC] = 0xFF;
AddCharMap ('\u221E', 0xC, 1);
#endregion
#region Letters and NonSpacing Marks (general)
// ASCII Latin alphabets
for (int i = 0; i < alphabets.Length; i++)
AddAlphaMap (alphabets [i], 0xE, alphaWeights [i]);
// non-ASCII Latin alphabets
// FIXME: there is no such characters that are placed
// *after* "alphabets" array items. This is nothing
// more than a hack that creates dummy weight for
// primary characters.
for (int i = 0x0080; i < 0x0300; i++) {
if (!Char.IsLetter ((char) i))
continue;
// For those Latin Letters which has NFKD are
// not added as independent primary character.
if (decompIndex [i] != 0)
continue;
// SPECIAL CASES:
// 1.some alphabets have primarily
// equivalent ASCII alphabets.
// 2.some have independent primary weights,
// but inside a-to-z range.
// 3.there are some expanded characters that
// are not part of Unicode Standard NFKD.
// 4. some characters are letter in IsLetter
// but not in sortkeys (maybe unicode version
// difference caused it).
switch (i) {
// 1. skipping them does not make sense
// case 0xD0: case 0xF0: case 0x131: case 0x138:
// case 0x184: case 0x185: case 0x186: case 0x189:
// case 0x18D: case 0x18E: case 0x18F: case 0x190:
// case 0x194: case 0x195: case 0x196: case 0x19A:
// case 0x19B: case 0x19C:
// 2. skipping them does not make sense
// case 0x14A: // Ng
// case 0x14B: // ng
// 3.
case 0xC6: // AE
case 0xE6: // ae
case 0xDE: // Icelandic Thorn
case 0xFE: // Icelandic Thorn
case 0xDF: // German ss
case 0xFF: // German ss
// 4.
case 0x1C0: case 0x1C1: case 0x1C2: case 0x1C3:
// not classified yet
// case 0x1A6: case 0x1A7: case 0x1A8: case 0x1A9:
// case 0x1AA: case 0x1B1: case 0x1B7: case 0x1B8:
// case 0x1B9: case 0x1BA: case 0x1BB: case 0x1BF:
// case 0x1DD:
continue;
}
AddCharMapGroup ((char) i, 0xE, 1, 0);
}
// IPA extensions
// FIXME: this results in not equivalent values to
// Windows, but is safer for comparison.
char [] ipaArray = new char [0x300 - 0x250 + 0x20];
for (int i = 0x40; i < 0x60; i++)
if (Char.IsLetter ((char) i))
ipaArray [i - 0x40] = (char) (i);
for (int i = 0x250; i < 0x300; i++)
if (Char.IsLetter ((char) i))
ipaArray [i - 0x250 + 0x20] = (char) i;
Array.Sort (ipaArray, UCAComparer.Instance);
int targetASCII = 0;
byte latinDiacritical = 0x7B;
foreach (char c in ipaArray) {
if (c <= 'Z') {
targetASCII = c;
latinDiacritical = 0x7B;
}
else
map [(int) c] = new CharMapEntry (
0xE,
map [targetASCII].Level1,
latinDiacritical++);
}
// Greek and Coptic
// FIXME: this is (mysterious and) incomplete.
for (int i = 0x0380; i < 0x0400; i++)
if (diacritical [i] == 0 &&
decompLength [i] == 1 &&
decompType [i] == DecompositionCompat)
diacritical [i] = 3;
fillIndex [0xF] = 2;
for (int i = 0x0391; i < 0x03AA; i++)
if (i != 0x03A2)
AddCharMap ((char) i, 0xF, 1,
diacritical [i]);
fillIndex [0xF] = 2;
for (int i = 0x03B1; i < 0x03CA; i++)
if (i != 0x03C2)
AddCharMap ((char) i, 0xF, 1,
diacritical [i]);
// Final Sigma
map [0x03C2] = new CharMapEntry (0xF,
map [0x03C3].Level1, map [0x03C3].Level2);
fillIndex [0xF] = 0x40;
for (int i = 0x03DA; i < 0x03F0; i++)
AddCharMap ((char) i, 0xF,
(byte) (i % 2 == 0 ? 0 : 2),
diacritical [i]);
// NFKD
for (int i = 0x0386; i <= 0x0400; i++)
FillLetterNFKD (i, true, true);
// Cyrillic.
// Cyrillic letters are sorted like Latin letters i.e.
// containing culture-specific letters between the
// standard Cyrillic sequence.
//
// We can't use UCA here; it has different sorting.
char [] orderedCyrillic = new char [] {
'\u0430', '\u0431', '\u0432', '\u0433', '\u0434',
'\u0452', // DJE for Serbocroatian
'\u0435',
'\u0454', // IE for Ukrainian
'\u0436', '\u0437',
'\u0455', // DZE
'\u0438',
'\u0456', // Byelorussian-Ukrainian I
'\u0457', // YI
'\u0439',
'\u0458', // JE
'\u043A', '\u043B',
'\u0459', // LJE
'\u043C', '\u043D',
'\u045A', // NJE
'\u043E',
// 4E9 goes here.
'\u043F', '\u0440', '\u0441', '\u0442',
'\u045B', // TSHE for Serbocroatian
'\u0443',
'\u045E', // Short U for Byelorussian
'\u04B1', // Straight U w/ stroke (diacritical!)
'\u0444', '\u0445', '\u0446', '\u0447',
'\u045F', // DZHE
'\u0448', '\u0449', '\u044A', '\u044B', '\u044C',
'\u044D', '\u044E', '\u044F'};
// For some characters here is a map to basic cyrillic
// letters. See UnicodeData.txt character names for
// the sources. Here I simply declare an equiv. array.
// The content characters are map from U+490(,491),
// skipping small letters.
char [] cymap_src = new char [] {
'\u0433', '\u0433', '\u0433', '\u0436',
'\u0437', '\u043A', '\u043A', '\u043A',
'\u043A', '\u043D', '\u043D', '\u043F',
'\u0445', '\u0441', '\u0442', '\u0443',
'\u0443', '\u0445', '\u0446', '\u0447',
'\u0447', '\u0432', '\u0435', '\u0435',
'\u0406', '\u0436', '\u043A', '\u043D',
'\u0447', '\u0435'};
fillIndex [0x10] = 0x8D;
for (int i = 0x0460; i < 0x0481; i++) {
if (Char.IsLetter ((char) i)) {
if (i == 0x0476)
// U+476/477 have the same
// primary weight as U+474/475.
fillIndex [0x10] -= 3;
AddLetterMap ((char) i, 0x10, 3);
}
}
fillIndex [0x10] = 0x6;
for (int i = 0; i < orderedCyrillic.Length; i++) {
char c = Char.ToUpper (orderedCyrillic [i], CultureInfo.InvariantCulture);
if (!IsIgnorable ((int) c) &&
Char.IsLetter (c) &&
!map [c].Defined) {
AddLetterMap (c, 0x10, 0);
fillIndex [0x10] += 3;
}
}
// NFKD
for (int i = 0x0401; i <= 0x045F; i++)
FillLetterNFKD (i, false, false);
for (int i = 0; i < cymap_src.Length; i++) {
char c = cymap_src [i];
fillIndex [0x10] = map [c].Level1;
int c2 = 0x0490 + i * 2;
AddLetterMapCore ((char) c2, 0x10, 0, diacritical [c2], false);
}
// Armenian
fillIndex [0x11] = 0x3;
fillIndex [0x1] = 0x98;
for (int i = 0x0531; i < 0x0586; i++) {
if (i == 0x0559 || i == 0x55A)
AddCharMap ((char) i, 1, 1);
if (Char.IsLetter ((char) i))
AddLetterMap ((char) i, 0x11, 1);
}
// Hebrew
// -Letters
fillIndex [0x12] = 0x2;
for (int i = 0x05D0; i < 0x05FF; i++)
if (Char.IsLetter ((char) i)) {
if (isUppercase [i]) {
fillIndex [0x12]--;
AddLetterMap ((char) i, 0x12, 2);
}
else
AddLetterMap ((char) i, 0x12, 1);
}
// -Accents
fillIndex [0x1] = 0x3;
for (int i = 0x0591; i <= 0x05C2; i++) {
if (i == 0x05A3 || i == 0x05BB)
fillIndex [0x1]++;
if (i != 0x05BE)
AddCharMap ((char) i, 0x1, 1);
}
// Arabic
fillIndex [0x1] = 0x8E;
fillIndex [0x13] = 0x3;
for (int i = 0x0621; i <= 0x064A; i++) {
// Abjad
if (Char.GetUnicodeCategory ((char) i)
!= UnicodeCategory.OtherLetter) {
// FIXME: arabic nonspacing marks are
// in different order.
AddCharMap ((char) i, 0x1, 1);
continue;
}
// map [i] = new CharMapEntry (0x13,
// (byte) arabicLetterPrimaryValues [i], 1);
fillIndex [0x13] =
(byte) arabicLetterPrimaryValues [i];
byte formDiacritical = 8; // default
// SPECIAL CASES:
switch (i) {
case 0x0622: formDiacritical = 9; break;
case 0x0623: formDiacritical = 0xA; break;
case 0x0624: formDiacritical = 5; break;
case 0x0625: formDiacritical = 0xB; break;
case 0x0626: formDiacritical = 7; break;
case 0x0649: formDiacritical = 5; break;
case 0x064A: formDiacritical = 7; break;
}
// AddLetterMapCore ((char) i, 0x13, 1, formDiacritical, false);
AddArabicCharMap ((char) i, 0x13, 1, formDiacritical);
}
for (int i = 0x0670; i < 0x0673; i++)
map [i] = new CharMapEntry (0x13, 0xB, (byte) (0xC + i - 0x670));
fillIndex [0x13] = 0x84;
for (int i = 0x0674; i < 0x06D6; i++)
if (Char.IsLetter ((char) i))
AddLetterMapCore ((char) i, 0x13, 1, 0, false);
// Devanagari
// FIXME: this could be fixed in more decent way
for (int i = 0x0958; i <= 0x095F; i++)
diacritical [i] = 8;
// FIXME: it does seem straight codepoint mapping.
fillIndex [0x14] = 04;
for (int i = 0x0901; i < 0x0905; i++)
if (!IsIgnorable (i))
AddLetterMap ((char) i, 0x14, 2);
fillIndex [0x14] = 0xB;
for (int i = 0x0905; i < 0x093A; i++) {
if (i == 0x0928)
AddCharMap ('\u0929', 0x14, 0, 8);
if (i == 0x0930)
AddCharMap ('\u0931', 0x14, 0, 8);
if (i == 0x0933)
AddCharMap ('\u0934', 0x14, 0, 8);
if (Char.IsLetter ((char) i))
AddLetterMap ((char) i, 0x14, 4);
if (i == 0x090B)
AddCharMap ('\u0960', 0x14, 4);
if (i == 0x090C)
AddCharMap ('\u0961', 0x14, 4);
}
fillIndex [0x14] = 0xDA;
for (int i = 0x093E; i < 0x0945; i++)
if (!IsIgnorable (i))
AddLetterMap ((char) i, 0x14, 2);
fillIndex [0x14] = 0xEC;
for (int i = 0x0945; i < 0x094F; i++)
if (!IsIgnorable (i))
AddLetterMap ((char) i, 0x14, 2);
// Bengali
// -Letters
fillIndex [0x15] = 02;
for (int i = 0x0980; i < 0x9FF; i++) {
if (IsIgnorable (i))
continue;
if (i == 0x09E0)
fillIndex [0x15] = 0x3B;
switch (Char.GetUnicodeCategory ((char) i)) {
case UnicodeCategory.NonSpacingMark:
case UnicodeCategory.DecimalDigitNumber:
case UnicodeCategory.OtherNumber:
continue;
}
AddLetterMap ((char) i, 0x15, 1);
}
// -Signs
fillIndex [0x1] = 0x3;
for (int i = 0x0981; i < 0x0A00; i++)
if (Char.GetUnicodeCategory ((char) i) ==
UnicodeCategory.NonSpacingMark)
AddCharMap ((char) i, 0x1, 1);
// Gurmukhi. orderedGurmukhi is from UCA
// FIXME: it does not look equivalent to UCA.
fillIndex [0x16] = 04;
fillIndex [0x1] = 3;
for (int i = 0; i < orderedGurmukhi.Length; i++) {
char c = orderedGurmukhi [i];
if (IsIgnorable ((int) c))
continue;
if (IsIgnorableNonSpacing (c)) {
AddLetterMap (c, 0x1, 1);
continue;
}
if (c == '\u0A3C' || c == '\u0A4D' ||
'\u0A66' <= c && c <= '\u0A71')
continue;
// SPECIAL CASES
byte shift = 4;
switch (c) {
case '\u0A33': case '\u0A36': case '\u0A16':
case '\u0A17': case '\u0A5B': case '\u0A5E':
shift = 0;
break;
}
if (c == '\u0A3E') // Skip
fillIndex [0x16] = 0xC0;
AddLetterMap (c, 0x16, shift);
}
// Gujarati. orderedGujarati is from UCA
fillIndex [0x17] = 0x4;
// nonspacing marks
map [0x0A4D] = new CharMapEntry (1, 0, 0x3);
map [0x0ABD] = new CharMapEntry (1, 0, 0x3);
map [0x0A3C] = new CharMapEntry (1, 0, 0x4);
map [0x0A71] = new CharMapEntry (1, 0, 0x6);
map [0x0ABC] = new CharMapEntry (1, 0, 0xB);
map [0x0A70] = new CharMapEntry (1, 0, 0xE);
// letters go first.
for (int i = 0; i < orderedGujarati.Length; i++) {
// SPECIAL CASE
char c = orderedGujarati [i];
if (Char.IsLetter (c)) {
// SPECIAL CASES
if (c == '\u0AB3' || c == '\u0A32')
continue;
if (c == '\u0A33') {
AddCharMap ('\u0A32', 0x17, 0);
AddCharMap ('\u0A33', 0x17, 4, 4);
continue;
}
if (c == '\u0A8B')
AddCharMap ('\u0AE0', 0x17, 0, 5);
AddCharMap (c, 0x17, 4);
if (c == '\u0AB9')
AddCharMap ('\u0AB3', 0x17, 6);
}
}
// non-letters
byte gujaratiShift = 4;
fillIndex [0x17] = 0xC0;
for (int i = 0; i < orderedGujarati.Length; i++) {
char c = orderedGujarati [i];
if (fillIndex [0x17] == 0xCC)
gujaratiShift = 3;
if (!Char.IsLetter (c)) {
// SPECIAL CASES
if (c == '\u0A82')
AddCharMap ('\u0A81', 0x17, 2);
if (c == '\u0AC2')
fillIndex [0x17]++;
AddLetterMap (c, 0x17, gujaratiShift);
}
}
// Oriya
fillIndex [0x1] = 03;
fillIndex [0x18] = 02;
for (int i = 0x0B00; i < 0x0B7F; i++) {
switch (Char.GetUnicodeCategory ((char) i)) {
case UnicodeCategory.NonSpacingMark:
case UnicodeCategory.DecimalDigitNumber:
AddLetterMap ((char) i, 0x1, 1);
continue;
}
AddLetterMapCore ((char) i, 0x18, 1, 0, true);
}
// Tamil
fillIndex [0x19] = 2;
AddCharMap ('\u0BD7', 0x19, 0);
fillIndex [0x19] = 0xA;
// vowels
for (int i = 0x0B82; i <= 0x0B94; i++)
if (!IsIgnorable ((char) i))
AddCharMap ((char) i, 0x19, 2);
// special vowel
fillIndex [0x19] = 0x28;
// The array for Tamil consonants is a constant.
// Windows have almost similar sequence to TAM from
// tamilnet but a bit different in Grantha.
for (int i = 0; i < orderedTamilConsonants.Length; i++)
AddLetterMap (orderedTamilConsonants [i], 0x19, 4);
// combining marks
fillIndex [0x19] = 0x82;
for (int i = 0x0BBE; i < 0x0BCD; i++)
if (Char.GetUnicodeCategory ((char) i) ==
UnicodeCategory.SpacingCombiningMark
|| i == 0x0BC0)
AddLetterMap ((char) i, 0x19, 2);
// Telugu
fillIndex [0x1A] = 0x4;
for (int i = 0x0C00; i < 0x0C62; i++) {
if (i == 0x0C55 || i == 0x0C56)
continue; // skip
AddCharMap ((char) i, 0x1A, 3);
char supp = (i == 0x0C0B) ? '\u0C60':
i == 0x0C0C ? '\u0C61' : char.MinValue;
if (supp == char.MinValue)
continue;
AddCharMap (supp, 0x1A, 3);
}
// Kannada
fillIndex [0x1B] = 4;
for (int i = 0x0C80; i < 0x0CE5; i++) {
if (i == 0x0CD5 || i == 0x0CD6)
continue; // ignore
if (i == 0x0CB1 || i == 0x0CB3 || i == 0x0CDE)
continue; // shift after 0xCB9
AddCharMap ((char) i, 0x1B, 3);
if (i == 0x0CB9) {
// SPECIAL CASES: but why?
AddCharMap ('\u0CB1', 0x1B, 3); // RRA
AddCharMap ('\u0CB3', 0x1B, 3); // LLA
AddCharMap ('\u0CDE', 0x1B, 3); // FA
}
if (i == 0x0CB2)
AddCharMap ('\u0CE1', 0x1B, 3); // vocalic LL
}
// Malayalam
fillIndex [0x1C] = 2;
fillIndex [0x1] = 3;
for (int i = 0x0D02; i < 0x0D61; i++) {
// FIXME: I avoided MSCompatUnicodeTable usage
// here (it results in recursion). So check if
// using NonSpacingMark makes sense or not.
if (Char.GetUnicodeCategory ((char) i) != UnicodeCategory.NonSpacingMark)
// if (!MSCompatUnicodeTable.IsIgnorable ((char) i))
AddCharMap ((char) i, 0x1C, 1);
else if (!IsIgnorable ((char) i))
AddCharMap ((char) i, 1, 1);
}
// Thai ... note that it breaks 0x1E wall after E2B!
// Also, all Thai characters have level 2 value 3.
fillIndex [0x1E] = 2;
fillIndex [0x1] = 3;
for (int i = 0xE40; i <= 0xE44; i++)
AddCharMap ((char) i, 0x1E, 1, 3);
for (int i = 0xE01; i < 0xE2B; i++)
AddCharMap ((char) i, 0x1E, 6, 3);
fillIndex [0x1F] = 5;
for (int i = 0xE2B; i < 0xE30; i++)
AddCharMap ((char) i, 0x1F, 6, 3);
fillIndex [0x1F] = 0x1E;
for (int i = 0xE30; i < 0xE3B; i++)
AddCharMap ((char) i, 0x1F, 1, 3);
// some Thai characters remains.
char [] specialThai = new char [] {'\u0E45', '\u0E46',
'\u0E4E', '\u0E4F', '\u0E5A', '\u0E5B'};
foreach (char c in specialThai)
AddCharMap (c, 0x1F, 1, 3);
for (int i = 0xE00; i < 0xE80; i++)
if (Char.GetUnicodeCategory ((char) i) ==
UnicodeCategory.NonSpacingMark)
AddCharMap ((char) i, 1, 1);
// Lao
fillIndex [0x1F] = 2;
fillIndex [0x1] = 3;
for (int i = 0xE80; i < 0xEDF; i++) {
if (IsIgnorable ((char) i))
continue;
else if (Char.IsLetter ((char) i))
AddCharMap ((char) i, 0x1F, 1);
else if (Char.GetUnicodeCategory ((char) i) ==
UnicodeCategory.NonSpacingMark)
AddCharMap ((char) i, 1, 1);
}
// Georgian. orderedGeorgian is from UCA DUCET.
fillIndex [0x21] = 5;
for (int i = 0; i < orderedGeorgian.Length; i++) {
char c = orderedGeorgian [i];
if (map [(int) c].Defined)
continue;
AddCharMap (c, 0x21, 0);
if (c < '\u10F6')
AddCharMap ((char) (c - 0x30), 0x21, 0);
fillIndex [0x21] += 5;
}
// Japanese Kana.
fillIndex [0x22] = 2;
int kanaOffset = 0x3041;
byte [] kanaLines = new byte [] {2, 2, 2, 2, 1, 3, 1, 2, 1};
for (int gyo = 0; gyo < 9; gyo++) {
for (int dan = 0; dan < 5; dan++) {
if (gyo == 7 && dan % 2 == 1) {
// 'ya'-gyo
fillIndex [0x22]++;
kanaOffset -= 2; // There is no space for yi and ye.
continue;
}
int cp = kanaOffset + dan * kanaLines [gyo];
// small lines (a-gyo, ya-gyo)
if (gyo == 0 || gyo == 7) {
AddKanaMap (cp, 1); // small
AddKanaMap (cp + 1, 1);
}
else
AddKanaMap (cp, kanaLines [gyo]);
fillIndex [0x22]++;
if (cp == 0x30AB) {
// add small 'ka' (before normal one)
AddKanaMap (0x30F5, 1);
kanaOffset++;
}
if (cp == 0x30B1) {
// add small 'ke' (before normal one)
AddKanaMap (0x30F6, 1);
kanaOffset++;
}
if (cp == 0x3061) {
// add small 'Tsu' (before normal one)
AddKanaMap (0x3063, 1);
kanaOffset++;
}
}
fillIndex [0x22] += 3;
kanaOffset += 5 * kanaLines [gyo];
}
// Wa-gyo is almost special, so I just manually add.
AddLetterMap ((char) 0x308E, 0x22, 0);
AddLetterMap ((char) (0x308E + 0x60), 0x22, 0);
AddLetterMap ((char) 0x308F, 0x22, 0);
AddLetterMap ((char) (0x308F + 0x60), 0x22, 0);
fillIndex [0x22]++;
AddLetterMap ((char) 0x3090, 0x22, 0);
AddLetterMap ((char) (0x3090 + 0x60), 0x22, 0);
fillIndex [0x22] += 2;
// no "Wu" in Japanese.
AddLetterMap ((char) 0x3091, 0x22, 0);
AddLetterMap ((char) (0x3091 + 0x60), 0x22, 0);
fillIndex [0x22]++;
AddLetterMap ((char) 0x3092, 0x22, 0);
AddLetterMap ((char) (0x3092 + 0x60), 0x22, 0);
// Nn
fillIndex [0x22] = 0x80;
AddLetterMap ((char) 0x3093, 0x22, 0);
AddLetterMap ((char) (0x3093 + 0x60), 0x22, 0);
map [0x3094] = new CharMapEntry (map [0x30A6].Category,
map [0x30A6].Level1, 3);// voiced hiragana U
map [0x30F4] = new CharMapEntry (map [0x30A6].Category,
map [0x30A6].Level1, 3);// voiced katakana U
map [0x30F5] = new CharMapEntry (map [0x30AB].Category,
map [0x30AB].Level1, 0);// small katakana Ka
map [0x30F6] = new CharMapEntry (map [0x30B1].Category,
map [0x30B1].Level1, 0);// small katakana Ke
// voiced Wa lines
for (int i = 0x30F7; i < 0x30FB; i++)
map [i] = new CharMapEntry (map [i - 8].Category,
map [i - 8].Level1,
3);
// JIS Japanese square chars.
fillIndex [0x22] = 0x97;
jisJapanese.Sort (JISComparer.Instance);
foreach (JISCharacter j in jisJapanese)
if (0x3300 <= j.CP && j.CP <= 0x3357)
AddCharMap ((char) j.CP, 0x22, 1);
// non-JIS Japanese square chars.
nonJisJapanese.Sort (NonJISComparer.Instance);
foreach (NonJISCharacter j in nonJisJapanese)
AddCharMap ((char) j.CP, 0x22, 1);
// Bopomofo
fillIndex [0x23] = 0x02;
for (int i = 0x3105; i <= 0x312C; i++)
AddCharMap ((char) i, 0x23, 1);
// Estrangela: ancient Syriac
fillIndex [0x24] = 0x0B;
// FIXME: is 0x71E really alternative form?
ArrayList syriacAlternatives = new ArrayList (
new int [] {0x714, 0x716, 0x71C, 0x71E, 0x724, 0x727});
for (int i = 0x0710; i <= 0x072C; i++) {
if (i == 0x0711) // NonSpacingMark
continue;
if (syriacAlternatives.Contains (i))
continue;
AddCharMap ((char) i, 0x24, 4);
// FIXME: why?
if (i == 0x721)
fillIndex [0x24]++;
}
foreach (int cp in syriacAlternatives)
map [cp] = new CharMapEntry (0x24,
(byte) (map [cp - 1].Level1 + 2),
0);
// FIXME: Syriac NonSpacingMark should go here.
// Thaana
// FIXME: it turned out that it does not look like UCA
fillIndex [0x24] = 0x6E;
fillIndex [0x1] = 0xAC;
for (int i = 0; i < orderedThaana.Length; i++) {
char c = orderedThaana [i];
if (IsIgnorableNonSpacing ((int) c))
AddCharMap (c, 1, 1);
AddCharMap (c, 0x24, 2);
if (c == '\u0782') // SPECIAL CASE: why?
fillIndex [0x24] += 2;
}
#endregion
// FIXME: Add more culture-specific letters (that are
// not supported in Windows collation) here.
// Surrogate ... they are computed.
#region Hangul
// Hangul.
//
// Unlike UCA Windows Hangul sequence mixes Jongseong
// with Choseong sequence as well as Jungseong,
// adjusted to have the same primary weight for the
// same base character. So it is impossible to compute
// those sort keys.
//
// Here I introduce an ordered sequence of mixed
// 'commands' and 'characters' that is similar to
// LDML text:
// - ',' increases primary weight.
// - [A B] means a range, increasing index
// - {A B} means a range, without increasing index
// - '=' is no operation (it means the characters
// of both sides have the same weight).
// - '>' inserts a Hangul Syllable block that
// contains 0x251 characters.
// - '<' decreases the index
// - '0'-'9' means skip count
// - whitespaces are ignored
//
string hangulSequence =
"\u1100=\u11A8 > \u1101=\u11A9 >"
+ "\u11C3, \u11AA, \u11C4, \u1102=\u11AB >"
+ "<{\u1113 \u1116}, \u3165,"
+ "\u11C5, \u11C6=\u3166,, \u11C7, \u11C8,"
+ "\u11AC, \u11C9, \u11AD, \u1103=\u11AE >"
+ "<\u1117, \u11CA, \u1104, \u11CB > \u1105=\u11AF >"
+ "<{\u1118 \u111B}, \u11B0, [\u11CC \u11D0], \u11B1,"
+ "[\u11D1 \u11D2], \u11B2,"
+ "[\u11D3 \u11D5], \u11B3,"
+ "[\u11D6 \u11D7], \u11B4, \u11B5,"
+ "\u11B6=\u11D8, \u3140,, \u11D9, \u1106=\u11B7 >"
+ "<{\u111C \u111D}, [\u11DA \u11E2], \u1107=\u11B8 >"
+ "<{\u111E \u1120}, \u3172,, \u3173, \u11E3, \u1108 >"
+ "<{\u1121 \u112C}, \u3144 \u11B9, \u3174, \u3175,,,, "
+ "\u3176,, \u3177, [\u11E4 \u11E6] \u3178,"
+ "\u3179, \u1109=\u11BA,,, \u3214=\u3274 <>"
+ "<{\u112D \u1133}, \u11E7 \u317A, \u317B, \u317C "
+ "[\u11E8 \u11E9],, \u11EA \u317D,, \u110A=\u11BB,,, >"
+ "<{\u1134 \u1140}, \u317E,,,,,,, \u11EB,"
+ "\u110B=\u11BC, [\u1161 \u11A2], \u1160 >"
+ "<{\u1141 \u114C}, \u3180=\u11EE, \u11EC, \u11ED,,,,, "
+ "\u11F1,, \u11F2,,,"
+ "\u11EF,,, \u3181=\u11F0, \u110C=\u11BD,, >"
+ "<\u114D, \u110D,, >"
+ "<{\u114E \u1151},, \u110E=\u11BE,, >"
+ "<{\u1152 \u1155},,, \u110F=\u11BF >"
+ "\u1110=\u11C0 > \u1111=\u11C1 >"
+ "<\u1156=\u1157, \u11F3, \u11F4, \u1112=\u11C2 >"
+ "<\u1158=\u1159=\u115F, \u3185, \u11F9,"
+ "[\u11F5 \u11F8]"
;
byte hangulCat = 0x52;
fillIndex [hangulCat] = 0x2;
int syllableBlock = 0;
for (int n = 0; n < hangulSequence.Length; n++) {
char c = hangulSequence [n];
int start, end;
if (Char.IsWhiteSpace (c))
continue;
switch (c) {
case '=':
break; // NOP
case ',':
IncrementSequentialIndex (ref hangulCat);
break;
case '<':
if (fillIndex [hangulCat] == 2)
throw new Exception ("FIXME: handle it correctly (yes it is hacky, it is really unfortunate).");
fillIndex [hangulCat]--;
break;
case '>':
IncrementSequentialIndex (ref hangulCat);
for (int l = 0; l < 0x15; l++)
for (int v = 0; v < 0x1C; v++) {
AddCharMap (
(char) (0xAC00 + syllableBlock * 0x1C * 0x15 + l * 0x1C + v), hangulCat, 0);
IncrementSequentialIndex (ref hangulCat);
}
syllableBlock++;
break;
case '[':
start = hangulSequence [n + 1];
end = hangulSequence [n + 3];
for (int i = start; i <= end; i++) {
AddCharMap ((char) i, hangulCat, 0);
if (end > i)
IncrementSequentialIndex (ref hangulCat);
}
n += 4; // consumes 5 characters for this operation
break;
case '{':
start = hangulSequence [n + 1];
end = hangulSequence [n + 3];
for (int i = start; i <= end; i++)
AddCharMap ((char) i, hangulCat, 0);
n += 4; // consumes 5 characters for this operation
break;
default:
AddCharMap (c, hangulCat, 0);
break;
}
}
// Some Jamo NFKD.
for (int i = 0x3200; i < 0x3300; i++) {
if (IsIgnorable (i) || map [i].Defined)
continue;
int ch = 0;
// w/ bracket
if (decompLength [i] == 4 &&
decompValues [decompIndex [i]] == '(')
ch = decompIndex [i] + 1;
// circled
else if (decompLength [i] == 2 &&
decompValues [decompIndex [i] + 1] == '\u1161')
ch = decompIndex [i];
else if (decompLength [i] == 1)
ch = decompIndex [i];
else
continue;
ch = decompValues [ch];
if (ch < 0x1100 || 0x1200 < ch &&
ch < 0xAC00 || 0xD800 < ch)
continue;
// SPECIAL CASE ?
int offset = i < 0x3260 ? 1 : 0;
if (0x326E <= i && i <= 0x3273)
offset = 1;
map [i] = new CharMapEntry (map [ch].Category,
(byte) (map [ch].Level1 + offset),
map [ch].Level2);
// Console.Error.WriteLine ("Jamo {0:X04} -> {1:X04}", i, decompValues [decompIndex [i] + 1]);
}
#endregion
// Letterlike characters and CJK compatibility square
sortableCharNames.Sort (StringDictionaryValueComparer.Instance);
int [] counts = new int ['Z' - 'A' + 1];
char [] namedChars = new char [sortableCharNames.Count];
int nCharNames = 0;
foreach (DictionaryEntry de in sortableCharNames) {
counts [((string) de.Value) [0] - 'A']++;
namedChars [nCharNames++] = (char) ((int) de.Key);
}
nCharNames = 0; // reset
for (int a = 0; a < counts.Length; a++) {
fillIndex [0xE] = (byte) (alphaWeights [a + 1] - counts [a]);
for (int i = 0; i < counts [a]; i++)
//Console.Error.WriteLine ("---- {0:X04} : {1:x02} / {2} {3}", (int) namedChars [nCharNames], fillIndex [0xE], ((DictionaryEntry) sortableCharNames [nCharNames]).Value, Char.GetUnicodeCategory (namedChars [nCharNames]));
AddCharMap (namedChars [nCharNames++], 0xE, 1);
}
// CJK unified ideograph.
byte cjkCat = 0x9E;
fillIndex [cjkCat] = 0x2;
for (int cp = 0x4E00; cp <= 0x9FBB; cp++)
if (!IsIgnorable (cp))
AddCharMapGroupCJK ((char) cp, ref cjkCat);
// CJK Extensions goes here.
// LAMESPEC: With this Windows style CJK layout, it is
// impossible to add more CJK ideograph i.e. 0x9FA6-
// 0x9FBB can never be added w/o breaking compat.
for (int cp = 0xF900; cp <= 0xFA2D; cp++)
if (!IsIgnorable (cp))
AddCharMapGroupCJK ((char) cp, ref cjkCat);
// PrivateUse ... computed.
// remaining Surrogate ... computed.
#region 07 - ASCII non-alphanumeric + 3001, 3002 // 07
// non-alphanumeric ASCII except for: + - < = > '
for (int i = 0x21; i < 0x7F; i++) {
// SPECIAL CASE: 02C6 looks regarded as
// equivalent to '^', which does not conform
// to Unicode standard character database.
if (i == 0x005B)
AddCharMap ('\u2045', 0x7, 0, 0x1C);
if (i == 0x005D)
AddCharMap ('\u2046', 0x7, 0, 0x1C);
if (i == 0x005E)
AddCharMap ('\u02C6', 0x7, 0, 3);
if (i == 0x0060)
AddCharMap ('\u02CB', 0x7, 0, 3);
if (Char.IsLetterOrDigit ((char) i)
|| "+-<=>'".IndexOf ((char) i) >= 0)
continue; // they are not added here.
AddCharMapGroup2 ((char) i, 0x7, 1, 0);
// Insert 3001 after ',' and 3002 after '.'
if (i == 0x2C)
AddCharMapGroup2 ('\u3001', 0x7, 1, 0);
else if (i == 0x2E)
AddCharMapGroup2 ('\u3002', 0x7, 1, 0);
else if (i == 0x3A)
AddCharMap ('\uFE30', 0x7, 1, 0);
}
#endregion
#region 07 - Punctuations and something else
for (int i = 0xA0; i < char.MaxValue; i++) {
if (IsIgnorable (i))
continue;
// FIXME: actually those reset should not be
// done but here I put for easy goal.
if (i == 0x05C3)
fillIndex [0x7]++;
if (i == 0x0700)
fillIndex [0x7] = 0xE2;
if (i == 0x2016)
fillIndex [0x7] = 0x77;
if (i == 0x3008)
fillIndex [0x7] = 0x93;
if (0x02C8 <= i && i <= 0x02CD)
continue; // nonspacing marks
// SPECIAL CASE: maybe they could be allocated
// dummy NFKD mapping and no special processing
// would be required here.
if (i == 0x00AF)
AddCharMap ('\u02C9', 0x7, 0, 3);
if (i == 0x00B4)
AddCharMap ('\u02CA', 0x7, 0, 3);
if (i == 0x02C7)
AddCharMap ('\u02D8', 0x7, 0, 3);
// SPECIAL CASES:
switch (i) {
case 0xAB: // 08
case 0xB7: // 0A
case 0xBB: // 08
case 0x02B9: // 01
case 0x02BA: // 01
case 0x2329: // 09
case 0x232A: // 09
continue;
}
switch (Char.GetUnicodeCategory ((char) i)) {
case UnicodeCategory.OtherPunctuation:
case UnicodeCategory.ClosePunctuation:
case UnicodeCategory.OpenPunctuation:
case UnicodeCategory.ConnectorPunctuation:
case UnicodeCategory.InitialQuotePunctuation:
case UnicodeCategory.FinalQuotePunctuation:
case UnicodeCategory.ModifierSymbol:
// SPECIAL CASES: // 0xA
if (0x2020 <= i && i <= 0x2031)
continue;
if (i == 0x3003) // added later
continue;
AddCharMapGroup2 ((char) i, 0x7, 1, 0);
break;
default:
if (i == 0xA6 || i == 0x1C3 || i == 0x037A) // SPECIAL CASE. FIXME: why?
goto case UnicodeCategory.OtherPunctuation;
break;
}
}
// Control pictures
// FIXME: it should not need to reset level 1, but
// it's for easy goal.
fillIndex [0x7] = 0xB6;
for (int i = 0x2400; i <= 0x2424; i++)
AddCharMap ((char) i, 0x7, 1, 0);
// FIXME: what are they?
AddCharMap ('\u3003', 0x7, 1);
AddCharMap ('\u3006', 0x7, 1);
AddCharMap ('\u02D0', 0x7, 1);
AddCharMap ('\u10FB', 0x7, 1);
AddCharMap ('\u0950', 0x7, 1);
AddCharMap ('\u093D', 0x7, 1);
AddCharMap ('\u0964', 0x7, 1);
AddCharMap ('\u0965', 0x7, 1);
AddCharMap ('\u0970', 0x7, 1);
#endregion
#region category 08 - symbols
fillIndex [0x8] = 2;
// Here Windows mapping is not straightforward. It is
// not based on computation but seems manual sorting.
AddCharMapGroup ('+', 0x8, 1, 0); // plus
AddCharMapGroup ('\u2212', 0x8, 1); // minus
AddCharMapGroup ('\u229D', 0x8, 1); // minus
AddCharMapGroup ('\u2297', 0x8, 1); // mul
AddCharMapGroup ('\u2044', 0x8, 1); // div
AddCharMapGroup ('\u2215', 0x8, 0); // div
AddCharMapGroup ('\u2298', 0x8, 1); // div slash
AddCharMapGroup ('\u2217', 0x8, 0); // mul
AddCharMapGroup ('\u229B', 0x8, 1); // asterisk oper
AddCharMapGroup ('\u2218', 0x8, 0); // ring
AddCharMapGroup ('\u229A', 0x8, 1); // ring
AddCharMapGroup ('\u2219', 0x8, 0); // bullet
AddCharMapGroup ('\u2299', 0x8, 1); // dot oper
AddCharMapGroup ('\u2213', 0x8, 1); // minus-or-plus
AddCharMapGroup ('\u003C', 0x8, 1); // <
AddCharMapGroup ('\u227A', 0x8, 1); // precedes relation
AddCharMapGroup ('\u22B0', 0x8, 1); // precedes under relation
for (int cp = 0; cp < 0x2300; cp++) {
if (cp == 0xAC) // SPECIAL CASE: skip
continue;
if (cp == 0x200) {
cp = 0x2200; // skip to 2200
fillIndex [0x8] = 0x21;
}
if (cp == 0x2295)
fillIndex [0x8] = 0x3;
if (cp == 0x22A2)
fillIndex [0x8] = 0xAB;
if (cp == 0x22B2)
fillIndex [0x8] = 0xB9;
if (!map [cp].Defined &&
// Char.GetUnicodeCategory ((char) cp) ==
// UnicodeCategory.MathSymbol)
Char.IsSymbol ((char) cp))
AddCharMapGroup ((char) cp, 0x8, 1);
// SPECIAL CASES: no idea why Windows sorts as such
switch (cp) {
case 0x3E:
AddCharMap ('\u227B', 0x8, 1, 0);
AddCharMap ('\u22B1', 0x8, 1, 0);
break;
case 0xB1:
AddCharMapGroup ('\u00AB', 0x8, 1);
AddCharMapGroup ('\u226A', 0x8, 1);
AddCharMapGroup ('\u00BB', 0x8, 1);
AddCharMapGroup ('\u226B', 0x8, 1);
break;
case 0xF7:
AddCharMap ('\u01C0', 0x8, 1, 0);
AddCharMap ('\u01C1', 0x8, 1, 0);
AddCharMap ('\u01C2', 0x8, 1, 0);
break;
}
}
#endregion
#region Hack!
// Characters w/ diacritical marks (NFKD)
for (int i = 0; i <= char.MaxValue; i++) {
if (map [i].Defined || IsIgnorable (i))
continue;
if (decompIndex [i] == 0)
continue;
int start = decompIndex [i];
int primaryChar = decompValues [start];
int secondary = diacritical [i];
bool skip = false;
int length = decompLength [i];
// special processing for parenthesized ones.
if (length == 3 &&
decompValues [start] == '(' &&
decompValues [start + 2] == ')') {
primaryChar = decompValues [start + 1];
length = 1;
}
if (map [primaryChar].Level1 == 0)
continue;
for (int l = 1; l < length; l++) {
int c = decompValues [start + l];
if (map [c].Level1 != 0)
skip = true;
secondary += diacritical [c];
}
if (skip)
continue;
map [i] = new CharMapEntry (
map [primaryChar].Category,
map [primaryChar].Level1,
(byte) secondary);
}
// Diacritical weight adjustment
// Arabic Hamzah
diacritical [0x624] = 0x5;
diacritical [0x626] = 0x7;
diacritical [0x622] = 0x9;
diacritical [0x623] = 0xA;
diacritical [0x625] = 0xB;
diacritical [0x649] = 0x5; // 'alif maqs.uurah
diacritical [0x64A] = 0x7; // Yaa'
for (int i = 0; i < char.MaxValue; i++) {
byte mod = 0;
byte cat = map [i].Category;
switch (cat) {
case 0xE: // Latin diacritics
case 0x22: // Japanese: circled characters
mod = diacritical [i];
break;
case 0x13: // Arabic
if (i == 0x0621)
break; // 0
if (diacritical [i] == 0 && decompLength [i] != 0)
diacritical [i] = map [decompValues [decompIndex [i]]].Level2;
if (diacritical [i] == 0 && i >= 0xFE8D)
mod = 0x8; // default for arabic
break;
}
if (0x52 <= cat && cat <= 0x7F) // Hangul
mod = diacritical [i];
if (mod > 0)
map [i] = new CharMapEntry (
cat, map [i].Level1, mod);
}
// FIXME: this is halfly hack but those NonSpacingMark
// characters and still undefined are likely to
// be nonspacing.
for (int i = 0; i < char.MaxValue; i++) {
if (map [i].Defined ||
IsIgnorable (i))
continue;
switch (i) {
// SPECIAL CASES.
case 0x02B9:
case 0x02BA:
break;
default:
if (Char.GetUnicodeCategory ((char) i) !=
UnicodeCategory.NonSpacingMark)
continue;
break;
}
if (diacritical [i] != 0)
map [i] = new CharMapEntry (1, 1, diacritical [i]);
else
AddCharMap ((char) i, 1, 1);
}
#endregion
}