qasm/ciderpress/nufxlib/Charset.c
2019-11-13 15:45:39 -08:00

554 lines
21 KiB
C

/*
* NuFX archive manipulation library
* Copyright (C) 2014 by Andy McFadden, All Rights Reserved.
* This is free software; you can redistribute it and/or modify it under the
* terms of the BSD License, see the file COPYING-LIB.
*
* Miscellaneous NufxLib utility functions.
*/
#include "NufxLibPriv.h"
/*
* Convert Mac OS Roman to Unicode. Mapping comes from:
*
* http://www.unicode.org/Public/MAPPINGS/VENDORS/APPLE/ROMAN.TXT
*
* We use the "Control Pictures" block for the control characters
* (0x00-0x1f, 0x7f --> 0x2400-0x241f, 0x2421). This is a bit nicer
* than embedding control characters in filenames.
*/
static const uint16_t gMORToUnicode[256] = {
/*0x00*/ 0x2400, // [control] NULL
/*0x01*/ 0x2401, // [control] START OF HEADING
/*0x02*/ 0x2402, // [control] START OF TEXT
/*0x03*/ 0x2403, // [control] END OF TEXT
/*0x04*/ 0x2404, // [control] END OF TRANSMISSION
/*0x05*/ 0x2405, // [control] ENQUIRY
/*0x06*/ 0x2406, // [control] ACKNOWLEDGE
/*0x07*/ 0x2407, // [control] BELL
/*0x08*/ 0x2408, // [control] BACKSPACE
/*0x09*/ 0x2409, // [control] HORIZONTAL TABULATION
/*0x0a*/ 0x240a, // [control] LINE FEED
/*0x0b*/ 0x240b, // [control] VERTICAL TABULATION
/*0x0c*/ 0x240c, // [control] FORM FEED
/*0x0d*/ 0x240d, // [control] CARRIAGE RETURN
/*0x0e*/ 0x240e, // [control] SHIFT OUT
/*0x0f*/ 0x240f, // [control] SHIFT IN
/*0x10*/ 0x2410, // [control] DATA LINK ESCAPE
/*0x11*/ 0x2411, // [control] DEVICE CONTROL ONE
/*0x12*/ 0x2412, // [control] DEVICE CONTROL TWO
/*0x13*/ 0x2413, // [control] DEVICE CONTROL THREE
/*0x14*/ 0x2414, // [control] DEVICE CONTROL FOUR
/*0x15*/ 0x2415, // [control] NEGATIVE ACKNOWLEDGE
/*0x16*/ 0x2416, // [control] SYNCHRONOUS IDLE
/*0x17*/ 0x2417, // [control] END OF TRANSMISSION BLOCK
/*0x18*/ 0x2418, // [control] CANCEL
/*0x19*/ 0x2419, // [control] END OF MEDIUM
/*0x1a*/ 0x241a, // [control] SUBSTITUTE
/*0x1b*/ 0x241b, // [control] ESCAPE
/*0x1c*/ 0x241c, // [control] FILE SEPARATOR
/*0x1d*/ 0x241d, // [control] GROUP SEPARATOR
/*0x1e*/ 0x241e, // [control] RECORD SEPARATOR
/*0x1f*/ 0x241f, // [control] UNIT SEPARATOR
/*0x20*/ 0x0020, // SPACE
/*0x21*/ 0x0021, // EXCLAMATION MARK
/*0x22*/ 0x0022, // QUOTATION MARK
/*0x23*/ 0x0023, // NUMBER SIGN
/*0x24*/ 0x0024, // DOLLAR SIGN
/*0x25*/ 0x0025, // PERCENT SIGN
/*0x26*/ 0x0026, // AMPERSAND
/*0x27*/ 0x0027, // APOSTROPHE
/*0x28*/ 0x0028, // LEFT PARENTHESIS
/*0x29*/ 0x0029, // RIGHT PARENTHESIS
/*0x2A*/ 0x002A, // ASTERISK
/*0x2B*/ 0x002B, // PLUS SIGN
/*0x2C*/ 0x002C, // COMMA
/*0x2D*/ 0x002D, // HYPHEN-MINUS
/*0x2E*/ 0x002E, // FULL STOP
/*0x2F*/ 0x002F, // SOLIDUS
/*0x30*/ 0x0030, // DIGIT ZERO
/*0x31*/ 0x0031, // DIGIT ONE
/*0x32*/ 0x0032, // DIGIT TWO
/*0x33*/ 0x0033, // DIGIT THREE
/*0x34*/ 0x0034, // DIGIT FOUR
/*0x35*/ 0x0035, // DIGIT FIVE
/*0x36*/ 0x0036, // DIGIT SIX
/*0x37*/ 0x0037, // DIGIT SEVEN
/*0x38*/ 0x0038, // DIGIT EIGHT
/*0x39*/ 0x0039, // DIGIT NINE
/*0x3A*/ 0x003A, // COLON
/*0x3B*/ 0x003B, // SEMICOLON
/*0x3C*/ 0x003C, // LESS-THAN SIGN
/*0x3D*/ 0x003D, // EQUALS SIGN
/*0x3E*/ 0x003E, // GREATER-THAN SIGN
/*0x3F*/ 0x003F, // QUESTION MARK
/*0x40*/ 0x0040, // COMMERCIAL AT
/*0x41*/ 0x0041, // LATIN CAPITAL LETTER A
/*0x42*/ 0x0042, // LATIN CAPITAL LETTER B
/*0x43*/ 0x0043, // LATIN CAPITAL LETTER C
/*0x44*/ 0x0044, // LATIN CAPITAL LETTER D
/*0x45*/ 0x0045, // LATIN CAPITAL LETTER E
/*0x46*/ 0x0046, // LATIN CAPITAL LETTER F
/*0x47*/ 0x0047, // LATIN CAPITAL LETTER G
/*0x48*/ 0x0048, // LATIN CAPITAL LETTER H
/*0x49*/ 0x0049, // LATIN CAPITAL LETTER I
/*0x4A*/ 0x004A, // LATIN CAPITAL LETTER J
/*0x4B*/ 0x004B, // LATIN CAPITAL LETTER K
/*0x4C*/ 0x004C, // LATIN CAPITAL LETTER L
/*0x4D*/ 0x004D, // LATIN CAPITAL LETTER M
/*0x4E*/ 0x004E, // LATIN CAPITAL LETTER N
/*0x4F*/ 0x004F, // LATIN CAPITAL LETTER O
/*0x50*/ 0x0050, // LATIN CAPITAL LETTER P
/*0x51*/ 0x0051, // LATIN CAPITAL LETTER Q
/*0x52*/ 0x0052, // LATIN CAPITAL LETTER R
/*0x53*/ 0x0053, // LATIN CAPITAL LETTER S
/*0x54*/ 0x0054, // LATIN CAPITAL LETTER T
/*0x55*/ 0x0055, // LATIN CAPITAL LETTER U
/*0x56*/ 0x0056, // LATIN CAPITAL LETTER V
/*0x57*/ 0x0057, // LATIN CAPITAL LETTER W
/*0x58*/ 0x0058, // LATIN CAPITAL LETTER X
/*0x59*/ 0x0059, // LATIN CAPITAL LETTER Y
/*0x5A*/ 0x005A, // LATIN CAPITAL LETTER Z
/*0x5B*/ 0x005B, // LEFT SQUARE BRACKET
/*0x5C*/ 0x005C, // REVERSE SOLIDUS
/*0x5D*/ 0x005D, // RIGHT SQUARE BRACKET
/*0x5E*/ 0x005E, // CIRCUMFLEX ACCENT
/*0x5F*/ 0x005F, // LOW LINE
/*0x60*/ 0x0060, // GRAVE ACCENT
/*0x61*/ 0x0061, // LATIN SMALL LETTER A
/*0x62*/ 0x0062, // LATIN SMALL LETTER B
/*0x63*/ 0x0063, // LATIN SMALL LETTER C
/*0x64*/ 0x0064, // LATIN SMALL LETTER D
/*0x65*/ 0x0065, // LATIN SMALL LETTER E
/*0x66*/ 0x0066, // LATIN SMALL LETTER F
/*0x67*/ 0x0067, // LATIN SMALL LETTER G
/*0x68*/ 0x0068, // LATIN SMALL LETTER H
/*0x69*/ 0x0069, // LATIN SMALL LETTER I
/*0x6A*/ 0x006A, // LATIN SMALL LETTER J
/*0x6B*/ 0x006B, // LATIN SMALL LETTER K
/*0x6C*/ 0x006C, // LATIN SMALL LETTER L
/*0x6D*/ 0x006D, // LATIN SMALL LETTER M
/*0x6E*/ 0x006E, // LATIN SMALL LETTER N
/*0x6F*/ 0x006F, // LATIN SMALL LETTER O
/*0x70*/ 0x0070, // LATIN SMALL LETTER P
/*0x71*/ 0x0071, // LATIN SMALL LETTER Q
/*0x72*/ 0x0072, // LATIN SMALL LETTER R
/*0x73*/ 0x0073, // LATIN SMALL LETTER S
/*0x74*/ 0x0074, // LATIN SMALL LETTER T
/*0x75*/ 0x0075, // LATIN SMALL LETTER U
/*0x76*/ 0x0076, // LATIN SMALL LETTER V
/*0x77*/ 0x0077, // LATIN SMALL LETTER W
/*0x78*/ 0x0078, // LATIN SMALL LETTER X
/*0x79*/ 0x0079, // LATIN SMALL LETTER Y
/*0x7A*/ 0x007A, // LATIN SMALL LETTER Z
/*0x7B*/ 0x007B, // LEFT CURLY BRACKET
/*0x7C*/ 0x007C, // VERTICAL LINE
/*0x7D*/ 0x007D, // RIGHT CURLY BRACKET
/*0x7E*/ 0x007E, // TILDE
/*0x7f*/ 0x2421, // [control] DELETE
/*0x80*/ 0x00C4, // LATIN CAPITAL LETTER A WITH DIAERESIS
/*0x81*/ 0x00C5, // LATIN CAPITAL LETTER A WITH RING ABOVE
/*0x82*/ 0x00C7, // LATIN CAPITAL LETTER C WITH CEDILLA
/*0x83*/ 0x00C9, // LATIN CAPITAL LETTER E WITH ACUTE
/*0x84*/ 0x00D1, // LATIN CAPITAL LETTER N WITH TILDE
/*0x85*/ 0x00D6, // LATIN CAPITAL LETTER O WITH DIAERESIS
/*0x86*/ 0x00DC, // LATIN CAPITAL LETTER U WITH DIAERESIS
/*0x87*/ 0x00E1, // LATIN SMALL LETTER A WITH ACUTE
/*0x88*/ 0x00E0, // LATIN SMALL LETTER A WITH GRAVE
/*0x89*/ 0x00E2, // LATIN SMALL LETTER A WITH CIRCUMFLEX
/*0x8A*/ 0x00E4, // LATIN SMALL LETTER A WITH DIAERESIS
/*0x8B*/ 0x00E3, // LATIN SMALL LETTER A WITH TILDE
/*0x8C*/ 0x00E5, // LATIN SMALL LETTER A WITH RING ABOVE
/*0x8D*/ 0x00E7, // LATIN SMALL LETTER C WITH CEDILLA
/*0x8E*/ 0x00E9, // LATIN SMALL LETTER E WITH ACUTE
/*0x8F*/ 0x00E8, // LATIN SMALL LETTER E WITH GRAVE
/*0x90*/ 0x00EA, // LATIN SMALL LETTER E WITH CIRCUMFLEX
/*0x91*/ 0x00EB, // LATIN SMALL LETTER E WITH DIAERESIS
/*0x92*/ 0x00ED, // LATIN SMALL LETTER I WITH ACUTE
/*0x93*/ 0x00EC, // LATIN SMALL LETTER I WITH GRAVE
/*0x94*/ 0x00EE, // LATIN SMALL LETTER I WITH CIRCUMFLEX
/*0x95*/ 0x00EF, // LATIN SMALL LETTER I WITH DIAERESIS
/*0x96*/ 0x00F1, // LATIN SMALL LETTER N WITH TILDE
/*0x97*/ 0x00F3, // LATIN SMALL LETTER O WITH ACUTE
/*0x98*/ 0x00F2, // LATIN SMALL LETTER O WITH GRAVE
/*0x99*/ 0x00F4, // LATIN SMALL LETTER O WITH CIRCUMFLEX
/*0x9A*/ 0x00F6, // LATIN SMALL LETTER O WITH DIAERESIS
/*0x9B*/ 0x00F5, // LATIN SMALL LETTER O WITH TILDE
/*0x9C*/ 0x00FA, // LATIN SMALL LETTER U WITH ACUTE
/*0x9D*/ 0x00F9, // LATIN SMALL LETTER U WITH GRAVE
/*0x9E*/ 0x00FB, // LATIN SMALL LETTER U WITH CIRCUMFLEX
/*0x9F*/ 0x00FC, // LATIN SMALL LETTER U WITH DIAERESIS
/*0xA0*/ 0x2020, // DAGGER
/*0xA1*/ 0x00B0, // DEGREE SIGN
/*0xA2*/ 0x00A2, // CENT SIGN
/*0xA3*/ 0x00A3, // POUND SIGN
/*0xA4*/ 0x00A7, // SECTION SIGN
/*0xA5*/ 0x2022, // BULLET
/*0xA6*/ 0x00B6, // PILCROW SIGN
/*0xA7*/ 0x00DF, // LATIN SMALL LETTER SHARP S
/*0xA8*/ 0x00AE, // REGISTERED SIGN
/*0xA9*/ 0x00A9, // COPYRIGHT SIGN
/*0xAA*/ 0x2122, // TRADE MARK SIGN
/*0xAB*/ 0x00B4, // ACUTE ACCENT
/*0xAC*/ 0x00A8, // DIAERESIS
/*0xAD*/ 0x2260, // NOT EQUAL TO
/*0xAE*/ 0x00C6, // LATIN CAPITAL LETTER AE
/*0xAF*/ 0x00D8, // LATIN CAPITAL LETTER O WITH STROKE
/*0xB0*/ 0x221E, // INFINITY
/*0xB1*/ 0x00B1, // PLUS-MINUS SIGN
/*0xB2*/ 0x2264, // LESS-THAN OR EQUAL TO
/*0xB3*/ 0x2265, // GREATER-THAN OR EQUAL TO
/*0xB4*/ 0x00A5, // YEN SIGN
/*0xB5*/ 0x00B5, // MICRO SIGN
/*0xB6*/ 0x2202, // PARTIAL DIFFERENTIAL
/*0xB7*/ 0x2211, // N-ARY SUMMATION
/*0xB8*/ 0x220F, // N-ARY PRODUCT
/*0xB9*/ 0x03C0, // GREEK SMALL LETTER PI
/*0xBA*/ 0x222B, // INTEGRAL
/*0xBB*/ 0x00AA, // FEMININE ORDINAL INDICATOR
/*0xBC*/ 0x00BA, // MASCULINE ORDINAL INDICATOR
/*0xBD*/ 0x03A9, // GREEK CAPITAL LETTER OMEGA
/*0xBE*/ 0x00E6, // LATIN SMALL LETTER AE
/*0xBF*/ 0x00F8, // LATIN SMALL LETTER O WITH STROKE
/*0xC0*/ 0x00BF, // INVERTED QUESTION MARK
/*0xC1*/ 0x00A1, // INVERTED EXCLAMATION MARK
/*0xC2*/ 0x00AC, // NOT SIGN
/*0xC3*/ 0x221A, // SQUARE ROOT
/*0xC4*/ 0x0192, // LATIN SMALL LETTER F WITH HOOK
/*0xC5*/ 0x2248, // ALMOST EQUAL TO
/*0xC6*/ 0x2206, // INCREMENT
/*0xC7*/ 0x00AB, // LEFT-POINTING DOUBLE ANGLE QUOTATION MARK
/*0xC8*/ 0x00BB, // RIGHT-POINTING DOUBLE ANGLE QUOTATION MARK
/*0xC9*/ 0x2026, // HORIZONTAL ELLIPSIS
/*0xCA*/ 0x00A0, // NO-BREAK SPACE
/*0xCB*/ 0x00C0, // LATIN CAPITAL LETTER A WITH GRAVE
/*0xCC*/ 0x00C3, // LATIN CAPITAL LETTER A WITH TILDE
/*0xCD*/ 0x00D5, // LATIN CAPITAL LETTER O WITH TILDE
/*0xCE*/ 0x0152, // LATIN CAPITAL LIGATURE OE
/*0xCF*/ 0x0153, // LATIN SMALL LIGATURE OE
/*0xD0*/ 0x2013, // EN DASH
/*0xD1*/ 0x2014, // EM DASH
/*0xD2*/ 0x201C, // LEFT DOUBLE QUOTATION MARK
/*0xD3*/ 0x201D, // RIGHT DOUBLE QUOTATION MARK
/*0xD4*/ 0x2018, // LEFT SINGLE QUOTATION MARK
/*0xD5*/ 0x2019, // RIGHT SINGLE QUOTATION MARK
/*0xD6*/ 0x00F7, // DIVISION SIGN
/*0xD7*/ 0x25CA, // LOZENGE
/*0xD8*/ 0x00FF, // LATIN SMALL LETTER Y WITH DIAERESIS
/*0xD9*/ 0x0178, // LATIN CAPITAL LETTER Y WITH DIAERESIS
/*0xDA*/ 0x2044, // FRACTION SLASH
/*0xDB*/ 0x00A4, // CURRENCY SIGN (was EURO SIGN)
/*0xDC*/ 0x2039, // SINGLE LEFT-POINTING ANGLE QUOTATION MARK
/*0xDD*/ 0x203A, // SINGLE RIGHT-POINTING ANGLE QUOTATION MARK
/*0xDE*/ 0xFB01, // LATIN SMALL LIGATURE FI
/*0xDF*/ 0xFB02, // LATIN SMALL LIGATURE FL
/*0xE0*/ 0x2021, // DOUBLE DAGGER
/*0xE1*/ 0x00B7, // MIDDLE DOT
/*0xE2*/ 0x201A, // SINGLE LOW-9 QUOTATION MARK
/*0xE3*/ 0x201E, // DOUBLE LOW-9 QUOTATION MARK
/*0xE4*/ 0x2030, // PER MILLE SIGN
/*0xE5*/ 0x00C2, // LATIN CAPITAL LETTER A WITH CIRCUMFLEX
/*0xE6*/ 0x00CA, // LATIN CAPITAL LETTER E WITH CIRCUMFLEX
/*0xE7*/ 0x00C1, // LATIN CAPITAL LETTER A WITH ACUTE
/*0xE8*/ 0x00CB, // LATIN CAPITAL LETTER E WITH DIAERESIS
/*0xE9*/ 0x00C8, // LATIN CAPITAL LETTER E WITH GRAVE
/*0xEA*/ 0x00CD, // LATIN CAPITAL LETTER I WITH ACUTE
/*0xEB*/ 0x00CE, // LATIN CAPITAL LETTER I WITH CIRCUMFLEX
/*0xEC*/ 0x00CF, // LATIN CAPITAL LETTER I WITH DIAERESIS
/*0xED*/ 0x00CC, // LATIN CAPITAL LETTER I WITH GRAVE
/*0xEE*/ 0x00D3, // LATIN CAPITAL LETTER O WITH ACUTE
/*0xEF*/ 0x00D4, // LATIN CAPITAL LETTER O WITH CIRCUMFLEX
/*0xF0*/ 0xF8FF, // Apple logo
/*0xF1*/ 0x00D2, // LATIN CAPITAL LETTER O WITH GRAVE
/*0xF2*/ 0x00DA, // LATIN CAPITAL LETTER U WITH ACUTE
/*0xF3*/ 0x00DB, // LATIN CAPITAL LETTER U WITH CIRCUMFLEX
/*0xF4*/ 0x00D9, // LATIN CAPITAL LETTER U WITH GRAVE
/*0xF5*/ 0x0131, // LATIN SMALL LETTER DOTLESS I
/*0xF6*/ 0x02C6, // MODIFIER LETTER CIRCUMFLEX ACCENT
/*0xF7*/ 0x02DC, // SMALL TILDE
/*0xF8*/ 0x00AF, // MACRON
/*0xF9*/ 0x02D8, // BREVE
/*0xFA*/ 0x02D9, // DOT ABOVE
/*0xFB*/ 0x02DA, // RING ABOVE
/*0xFC*/ 0x00B8, // CEDILLA
/*0xFD*/ 0x02DD, // DOUBLE ACUTE ACCENT
/*0xFE*/ 0x02DB, // OGONEK
/*0xFF*/ 0x02C7 // CARON
};
/*
* Static table, populated on first use. Provides the inverse map.
*
* An entry with 0x00 indicates no conversion. That's incorrect for
* the entry for '\0', but since we're operating on null-terminated
* strings that's never valid anyway. (It's possible for a filename
* to contain 0x2400, but that would translate to 0x00, which we don't
* allow; so it makes more sense to treat it as illegal.)
*/
static uint8_t gUnicodeToMOR[65536] = { 0xff /*indicates not initialized*/ };
static void Nu_GenerateUnicodeToMOR(void)
{
memset(gUnicodeToMOR, 0, sizeof(gUnicodeToMOR));
int i;
for (i = 0; i < 256; i++) {
int codePoint = gMORToUnicode[i];
Assert(codePoint >= 0 && codePoint < 65536);
gUnicodeToMOR[codePoint] = i;
}
}
/*
* Converts stringMOR to Unicode, storing the output in bufUNI until it's
* full. Null termination is guaranteed. If the buffer size is zero or
* bufUNI is NULL, no string data is returned.
*
* Returns the number of bytes required to represent stringMOR in Unicode.
*/
size_t Nu_ConvertMORToUNI(const char* stringMOR, UNICHAR* bufUNI,
size_t bufSize)
{
Assert(stringMOR != 0);
#ifdef _WIN32
/* place-holder if we're not using UTF-16 yet */
Assert(sizeof(UNICHAR) == 1);
size_t morLen = strlen(stringMOR) + 1;
if (bufUNI != NULL && bufSize != 0) {
size_t copyLen = morLen < bufSize ? morLen : bufSize;
memcpy(bufUNI, stringMOR, copyLen);
bufUNI[bufSize-1] = '\0';
}
return morLen;
#else
/*
* Convert Mac OS Roman to UTF-8. We only output full code points,
* so if only the first byte of a UTF-8 sequence will fit we just
* stop early.
*/
size_t uniLen = 0;
Boolean doOutput = (bufUNI != NULL);
while (*stringMOR != '\0') {
// ASCII values just "convert" to themselves in this table
uint16_t us = gMORToUnicode[(uint8_t)*stringMOR];
if (us < 0x80) {
// single byte, no conversion
if (uniLen+1 >= bufSize) {
doOutput = false;
}
if (doOutput) {
bufUNI[uniLen] = (char) us;
}
uniLen++;
} else if (us < 0x7ff) {
// two bytes
if (uniLen+2 >= bufSize) {
doOutput = false;
}
if (doOutput) {
bufUNI[uniLen] = (us >> 6) | 0xc0;
bufUNI[uniLen+1] = (us & 0x3f) | 0x80;
}
uniLen += 2;
} else {
// three bytes
if (uniLen+3 >= bufSize) {
doOutput = false;
}
if (doOutput) {
bufUNI[uniLen] = (us >> 12) | 0xe0;
bufUNI[uniLen+1] = ((us >> 6) & 0x3f) | 0x80;
bufUNI[uniLen+2] = (us & 0x3f) | 0x80;
}
uniLen += 3;
}
stringMOR++;
}
// null-terminate
if (doOutput && uniLen < bufSize) {
bufUNI[uniLen] = '\0';
}
uniLen++;
return uniLen;
#endif
}
/*
* Decode a single Unicode code point from a UTF-8 string. This will
* consume 1 to 4 bytes. If an error is detected, only one byte is
* consumed, and the code point value will be 0xDCnn (invalid).
*
* cf. http://en.wikipedia.org/wiki/UTF-8#Sample_code
*/
static uint32_t Nu_DecodeUTF8(const char** pStr)
{
const uint8_t* str = (const uint8_t*) *pStr;
uint32_t codePoint;
uint32_t uc1, uc2, uc3, uc4;
uc1 = *str++;
if (uc1 < 0x80) {
// single byte
codePoint = uc1;
} else if (uc1 < 0xc2) {
// illegal: continuation or overlong 2-byte sequence
goto fail;
} else if (uc1 < 0xe0) {
// 2-byte sequence
uc2 = *str++;
if ((uc2 & 0xc0) != 0x80) {
goto fail; // not a continuation
}
codePoint = (uc1 << 6) + uc2 - 0x3080;
} else if (uc1 < 0xf0) {
// 3-byte sequence */
uc2 = *str++;
if ((uc2 & 0xc0) != 0x80) {
goto fail; // not a continuation
}
if (uc1 == 0xe0 && uc2 < 0xa0) {
goto fail; // overlong
}
uc3 = *str++;
if ((uc3 & 0xc0) != 0x80) {
goto fail; // not a continuation
}
codePoint = (uc1 << 12) + (uc2 << 6) + uc3 - 0xE2080;
} else if (uc1 < 0xf5) {
uc2 = *str++;
if ((uc2 & 0xc0) != 0x80) {
goto fail; // not a continuation
}
if (uc1 == 0xf0 && uc2 < 0x90) {
goto fail; // overlong
}
if (uc1 == 0xf4 && uc2 >= 0x90) {
goto fail; // U+10FFFF
}
uc3 = *str++;
if ((uc3 & 0xc0) != 0x80) {
goto fail; // not a continuation
}
uc4 = *str++;
if ((uc4 & 0xc0) != 0x80) {
goto fail; // not a continuation
}
codePoint = (uc1 << 18) + (uc2 << 12) + (uc3 << 6) + uc4 - 0x3C82080;
} else {
// illegal: > U+10FFFF
goto fail;
}
*pStr = (const UNICHAR*) str;
return codePoint;
fail:
(*pStr)++; // advance one char only
return 0xdc00 | uc1;
}
/*
* Converts stringUNI to Mac OS Roman, storing the output in bufMOR
* until it's full. Null termination is guaranteed. If the buffer
* size is zero or bufMOR is NULL, no string data is returned.
*
* Returns the number of bytes required to represent stringUNI in MOR.
*/
size_t Nu_ConvertUNIToMOR(const UNICHAR* stringUNI, char* bufMOR,
size_t bufSize)
{
Assert(stringUNI != 0);
#ifdef _WIN32
/*
* Place-holder if we're not using UTF-16 yet. This doesn't pass
* tests that check for behavior with non-MOR Unicode values.
*/
Assert(sizeof(UNICHAR) == 1);
size_t uniLen = strlen(stringUNI) + 1;
if (bufMOR != NULL && bufSize != 0) {
size_t copyLen = uniLen < bufSize ? uniLen : bufSize;
memcpy(bufMOR, stringUNI, copyLen);
bufMOR[bufSize-1] = '\0';
}
return uniLen;
#else
/*
* Convert UTF-8 to Mac OS Roman. If the code point doesn't have
* a valid conversion (either because it's not in the table, or the
* UTF-8 code is damaged) we just insert an ASCII '?'.
*/
if (gUnicodeToMOR[0] == 0xff) {
Nu_GenerateUnicodeToMOR();
Assert(gUnicodeToMOR[0] != 0xff);
}
uint32_t codePoint;
size_t morLen = 0;
Boolean doOutput = (bufMOR != NULL);
while (*stringUNI != '\0') {
codePoint = Nu_DecodeUTF8(&stringUNI);
char mc;
if (codePoint < 0x80) {
mc = (char) codePoint;
} else if (codePoint < 0xffff) {
// UTF-8 errors come back as 0xDCnn, which has no mapping in table
mc = gUnicodeToMOR[codePoint];
if (mc == 0x00) {
mc = '?';
}
} else {
// non-BMP code point
mc = '?';
}
if (morLen+1 >= bufSize) {
doOutput = false;
}
if (doOutput) {
bufMOR[morLen] = mc;
}
morLen++;
}
// null-terminate
if (doOutput && morLen < bufSize) {
bufMOR[morLen] = '\0';
}
morLen++;
return morLen;
#endif
}
/*
* Utility function that wraps NuConvertMORToUTF8, allocating a new
* buffer to hold the converted string. The caller must free the result.
*
* Returns NULL if stringMOR is NULL or the conversion fails.
*/
UNICHAR* Nu_CopyMORToUNI(const char* stringMOR)
{
size_t uniLen;
UNICHAR* uniBuf;
if (stringMOR == NULL) {
return NULL;
}
uniLen = Nu_ConvertMORToUNI(stringMOR, NULL, 0);
if (uniLen == (size_t) -1) {
return NULL;
}
uniBuf = (UNICHAR*) Nu_Malloc(NULL, uniLen);
Nu_ConvertMORToUNI(stringMOR, uniBuf, uniLen);
return uniBuf;
}