/* $Id$ */ /** @file strings.cpp Handling of translated strings. */ #include "stdafx.h" #include "openttd.h" #include "currency.h" #include "namegen_func.h" #include "station_base.h" #include "town.h" #include "screenshot.h" #include "waypoint.h" #include "industry.h" #include "variables.h" #include "newgrf_text.h" #include "music.h" #include "industry.h" #include "fileio.h" #include "cargotype.h" #include "group.h" #include "debug.h" #include "newgrf_townname.h" #include "signs_base.h" #include "newgrf_engine.h" #include "spritecache.h" #include "fontcache.h" #include "gui.h" #include "strings_func.h" #include "functions.h" #include "rev.h" #include "core/endian_func.hpp" #include "date_func.h" #include "vehicle_base.h" #include "string_func.h" #include "player_func.h" #include "player_base.h" #include "fios.h" #include "settings_type.h" #include "video/video_driver.hpp" #include "engine_func.h" #include "engine_base.h" #include "saveload.h" #include "table/strings.h" #include "table/control_codes.h" DynamicLanguages _dynlang; char _userstring[128]; uint64 _decode_parameters[20]; static char *StationGetSpecialString(char *buff, int x, const char* last); static char *GetSpecialTownNameString(char *buff, int ind, uint32 seed, const char* last); static char *GetSpecialPlayerNameString(char *buff, int ind, const int64 *argv, const char* last); static char *FormatString(char *buff, const char *str, const int64 *argv, uint casei, const char* last); struct LanguagePack { uint32 ident; // 32-bits identifier uint32 version; // 32-bits of auto generated version info which is basically a hash of strings.h char name[32]; // the international name of this language char own_name[32]; // the localized name of this language char isocode[16]; // the ISO code for the language (not country code) uint16 offsets[32]; // the offsets byte plural_form; // how to compute plural forms byte pad[3]; // pad header to be a multiple of 4 char data[VARARRAY_SIZE]; // list of strings }; static char **_langpack_offs; static LanguagePack *_langpack; static uint _langtab_num[32]; // Offset into langpack offs static uint _langtab_start[32]; // Offset into langpack offs /** Read an int64 from the argv array. */ static inline int64 GetInt64(const int64 **argv) { assert(argv); return *(*argv)++; } /** Read an int32 from the argv array. */ static inline int32 GetInt32(const int64 **argv) { return (int32)GetInt64(argv); } /** Read an array from the argv array. */ static inline const int64 *GetArgvPtr(const int64 **argv, int n) { const int64 *result; assert(*argv); result = *argv; (*argv) += n; return result; } #define NUM_BOUND_STRINGS 8 /* Array to hold the bound strings. */ static const char *_bound_strings[NUM_BOUND_STRINGS]; /* This index is used to implement a "round-robin" allocating of * slots for BindCString. NUM_BOUND_STRINGS slots are reserved. * Which means that after NUM_BOUND_STRINGS calls to BindCString, * the indices will be reused. */ static int _bind_index; const char *GetStringPtr(StringID string) { switch (GB(string, 11, 5)) { case 28: return GetGRFStringPtr(GB(string, 0, 11)); case 29: return GetGRFStringPtr(GB(string, 0, 11) + 0x0800); case 30: return GetGRFStringPtr(GB(string, 0, 11) + 0x1000); default: return _langpack_offs[_langtab_start[string >> 11] + (string & 0x7FF)]; } } /** The highest 8 bits of string contain the "case index". * These 8 bits will only be set when FormatString wants to print * the string in a different case. No one else except FormatString * should set those bits, therefore string CANNOT be StringID, but uint32. * @param buffr * @param string * @param argv * @param last * @return a formatted string of char */ static char *GetStringWithArgs(char *buffr, uint string, const int64 *argv, const char* last) { if (GB(string, 0, 16) == 0) return GetStringWithArgs(buffr, STR_UNDEFINED, argv, last); uint index = GB(string, 0, 11); uint tab = GB(string, 11, 5); switch (tab) { case 4: if (index >= 0xC0) return GetSpecialTownNameString(buffr, index - 0xC0, GetInt32(&argv), last); break; case 14: if (index >= 0xE4) return GetSpecialPlayerNameString(buffr, index - 0xE4, argv, last); break; case 15: /* Old table for custom names. This is no longer used */ error("Incorrect conversion of custom name string."); case 26: /* Include string within newgrf text (format code 81) */ if (HasBit(index, 10)) { StringID string = GetGRFStringID(0, 0xD000 + GB(index, 0, 10)); return GetStringWithArgs(buffr, string, argv, last); } break; case 28: return FormatString(buffr, GetGRFStringPtr(index), argv, 0, last); case 29: return FormatString(buffr, GetGRFStringPtr(index + 0x0800), argv, 0, last); case 30: return FormatString(buffr, GetGRFStringPtr(index + 0x1000), argv, 0, last); case 31: /* dynamic strings. These are NOT to be passed through the formatter, * but passed through verbatim. */ if (index < (STR_SPEC_USERSTRING & 0x7FF)) { return strecpy(buffr, _bound_strings[index], last); } return FormatString(buffr, _userstring, NULL, 0, last); } if (index >= _langtab_num[tab]) { error( "String 0x%X is invalid. " "Probably because an old version of the .lng file.\n", string ); } return FormatString(buffr, GetStringPtr(GB(string, 0, 16)), argv, GB(string, 24, 8), last); } char *GetString(char *buffr, StringID string, const char* last) { return GetStringWithArgs(buffr, string, (int64*)_decode_parameters, last); } char *InlineString(char *buf, StringID string) { buf += Utf8Encode(buf, SCC_STRING_ID); buf += Utf8Encode(buf, string); return buf; } /** * This function takes a C-string and allocates a temporary string ID. * The StringID of the bound string is valid until BindCString is called * another NUM_BOUND_STRINGS times. So be careful when using it. * @param str temp string to add * @return the id of that temp string * @note formatting a DATE_TINY calls BindCString twice, thus reduces the * amount of 'user' bound strings by 2. * @todo rewrite the BindCString system to make the limit flexible and * non-round-robin. For example by using smart pointers that free * the allocated StringID when they go out-of-scope/are freed. */ StringID BindCString(const char *str) { int idx = (++_bind_index) & (NUM_BOUND_STRINGS - 1); _bound_strings[idx] = str; return idx + STR_SPEC_DYNSTRING; } /** This function is used to "bind" a C string to a OpenTTD dparam slot. * @param n slot of the string * @param str string to bind */ void SetDParamStr(uint n, const char *str) { SetDParam(n, BindCString(str)); } void InjectDParam(int amount) { memmove(_decode_parameters + amount, _decode_parameters, sizeof(_decode_parameters) - amount * sizeof(uint64)); } // TODO static char *FormatCommaNumber(char *buff, int64 number, const char *last) { uint64 divisor = 10000000000000000000ULL; uint64 quot; int i; uint64 tot; uint64 num; if (number < 0) { *buff++ = '-'; number = -number; } num = number; tot = 0; for (i = 0; i < 20; i++) { quot = 0; if (num >= divisor) { quot = num / divisor; num = num % divisor; } if (tot |= quot || i == 19) { *buff++ = '0' + quot; if ((i % 3) == 1 && i != 19) *buff++ = ','; } divisor /= 10; } *buff = '\0'; return buff; } // TODO static char *FormatNoCommaNumber(char *buff, int64 number, const char *last) { uint64 divisor = 10000000000000000000ULL; uint64 quot; int i; uint64 tot; uint64 num; if (number < 0) { buff = strecpy(buff, "-", last); number = -number; } num = number; tot = 0; for (i = 0; i < 20; i++) { quot = 0; if (num >= divisor) { quot = num / divisor; num = num % divisor; } if (tot |= quot || i == 19) { *buff++ = '0' + quot; } divisor /= 10; } *buff = '\0'; return buff; } static char *FormatYmdString(char *buff, Date date, const char* last) { YearMonthDay ymd; ConvertDateToYMD(date, &ymd); int64 args[3] = { ymd.day + STR_01AC_1ST - 1, STR_0162_JAN + ymd.month, ymd.year }; return FormatString(buff, GetStringPtr(STR_DATE_LONG), args, 0, last); } static char *FormatMonthAndYear(char *buff, Date date, const char* last) { YearMonthDay ymd; ConvertDateToYMD(date, &ymd); int64 args[2] = { STR_MONTH_JAN + ymd.month, ymd.year }; return FormatString(buff, GetStringPtr(STR_DATE_SHORT), args, 0, last); } static char *FormatTinyDate(char *buff, Date date, const char* last) { YearMonthDay ymd; ConvertDateToYMD(date, &ymd); char day[3]; char month[3]; /* We want to zero-pad the days and months */ snprintf(day, lengthof(day), "%02i", ymd.day); snprintf(month, lengthof(month), "%02i", ymd.month + 1); int64 args[3] = { BindCString(day), BindCString(month), ymd.year }; return FormatString(buff, GetStringPtr(STR_DATE_TINY), args, 0, last); } static char *FormatGenericCurrency(char *buff, const CurrencySpec *spec, Money number, bool compact, const char* last) { /* We are going to make number absolute for printing, so * keep this piece of data as we need it later on */ bool negative = number < 0; const char *multiplier = ""; char buf[40]; char *p; int j; number *= spec->rate; /* convert from negative */ if (number < 0) { if (buff + Utf8CharLen(SCC_RED) > last) return buff; buff += Utf8Encode(buff, SCC_RED); buff = strecpy(buff, "-", last); number = -number; } /* Add prefix part, folowing symbol_pos specification. * Here, it can can be either 0 (prefix) or 2 (both prefix anf suffix). * The only remaining value is 1 (suffix), so everything that is not 1 */ if (spec->symbol_pos != 1) buff = strecpy(buff, spec->prefix, last); /* for huge numbers, compact the number into k or M */ if (compact) { if (number >= 1000000000) { number = (number + 500000) / 1000000; multiplier = "M"; } else if (number >= 1000000) { number = (number + 500) / 1000; multiplier = "k"; } } /* convert to ascii number and add commas */ p = endof(buf); *--p = '\0'; j = 4; do { if (--j == 0) { *--p = spec->separator; j = 3; } *--p = '0' + (char)(number % 10); } while ((number /= 10) != 0); buff = strecpy(buff, p, last); buff = strecpy(buff, multiplier, last); /* Add suffix part, folowing symbol_pos specification. * Here, it can can be either 1 (suffix) or 2 (both prefix anf suffix). * The only remaining value is 1 (prefix), so everything that is not 0 */ if (spec->symbol_pos != 0) buff = strecpy(buff, spec->suffix, last); if (negative) { if (buff + Utf8CharLen(SCC_PREVIOUS_COLOUR) > last) return buff; buff += Utf8Encode(buff, SCC_PREVIOUS_COLOUR); *buff = '\0'; } return buff; } static int DeterminePluralForm(int64 count) { /* The absolute value determines plurality */ uint64 n = abs(count); switch (_langpack->plural_form) { default: NOT_REACHED(); /* Two forms, singular used for one only * Used in: * Danish, Dutch, English, German, Norwegian, Swedish, Estonian, Finnish, * Greek, Hebrew, Italian, Portuguese, Spanish, Esperanto */ case 0: return n != 1; /* Only one form * Used in: * Hungarian, Japanese, Korean, Turkish */ case 1: return 0; /* Two forms, singular used for zero and one * Used in: * French, Brazilian Portuguese */ case 2: return n > 1; /* Three forms, special case for zero * Used in: * Latvian */ case 3: return n % 10 == 1 && n % 100 != 11 ? 0 : n != 0 ? 1 : 2; /* Three forms, special case for one and two * Used in: * Gaelige (Irish) */ case 4: return n == 1 ? 0 : n == 2 ? 1 : 2; /* Three forms, special case for numbers ending in 1[2-9] * Used in: * Lithuanian */ case 5: return n % 10 == 1 && n % 100 != 11 ? 0 : n % 10 >= 2 && (n % 100 < 10 || n % 100 >= 20) ? 1 : 2; /* Three forms, special cases for numbers ending in 1 and 2, 3, 4, except those ending in 1[1-4] * Used in: * Croatian, Czech, Russian, Slovak, Ukrainian */ case 6: return n % 10 == 1 && n % 100 != 11 ? 0 : n % 10 >= 2 && n % 10 <= 4 && (n % 100 < 10 || n % 100 >= 20) ? 1 : 2; /* Three forms, special case for one and some numbers ending in 2, 3, or 4 * Used in: * Polish */ case 7: return n == 1 ? 0 : n % 10 >= 2 && n % 10 <= 4 && (n % 100 < 10 || n % 100 >= 20) ? 1 : 2; /* Four forms, special case for one and all numbers ending in 02, 03, or 04 * Used in: * Slovenian */ case 8: return n % 100 == 1 ? 0 : n % 100 == 2 ? 1 : n % 100 == 3 || n % 100 == 4 ? 2 : 3; /* Two forms; singular used for everything ending in 1 but not in 11. * Used in: * Icelandic */ case 9: return n % 10 == 1 && n % 100 != 11 ? 0 : 1; } } static const char *ParseStringChoice(const char *b, uint form, char *dst, int *dstlen) { //<NUM> {Length of each string} {each string} uint n = (byte)*b++; uint pos, i, mylen = 0, mypos = 0; for (i = pos = 0; i != n; i++) { uint len = (byte)*b++; if (i == form) { mypos = pos; mylen = len; } pos += len; } *dstlen = mylen; memcpy(dst, b + mypos, mylen); return b + pos; } struct Units { int s_m; ///< Multiplier for velocity int s_s; ///< Shift for velocity StringID velocity; ///< String for velocity int p_m; ///< Multiplier for power int p_s; ///< Shift for power StringID power; ///< String for velocity int w_m; ///< Multiplier for weight int w_s; ///< Shift for weight StringID s_weight; ///< Short string for weight StringID l_weight; ///< Long string for weight int v_m; ///< Multiplier for volume int v_s; ///< Shift for volume StringID s_volume; ///< Short string for volume StringID l_volume; ///< Long string for volume int f_m; ///< Multiplier for force int f_s; ///< Shift for force StringID force; ///< String for force }; /* Unit conversions */ static const Units units[] = { { // Imperial (Original, mph, hp, metric ton, litre, kN) 1, 0, STR_UNITS_VELOCITY_IMPERIAL, 1, 0, STR_UNITS_POWER_IMPERIAL, 1, 0, STR_UNITS_WEIGHT_SHORT_METRIC, STR_UNITS_WEIGHT_LONG_METRIC, 1000, 0, STR_UNITS_VOLUME_SHORT_METRIC, STR_UNITS_VOLUME_LONG_METRIC, 1, 0, STR_UNITS_FORCE_SI, }, { // Metric (km/h, hp, metric ton, litre, kN) 103, 6, STR_UNITS_VELOCITY_METRIC, 1, 0, STR_UNITS_POWER_METRIC, 1, 0, STR_UNITS_WEIGHT_SHORT_METRIC, STR_UNITS_WEIGHT_LONG_METRIC, 1000, 0, STR_UNITS_VOLUME_SHORT_METRIC, STR_UNITS_VOLUME_LONG_METRIC, 1, 0, STR_UNITS_FORCE_SI, }, { // SI (m/s, kilowatt, kilogram, cubic metres, kilonewton) 1831, 12, STR_UNITS_VELOCITY_SI, 764, 10, STR_UNITS_POWER_SI, 1000, 0, STR_UNITS_WEIGHT_SHORT_SI, STR_UNITS_WEIGHT_LONG_SI, 1, 0, STR_UNITS_VOLUME_SHORT_SI, STR_UNITS_VOLUME_LONG_SI, 1, 0, STR_UNITS_FORCE_SI, }, }; /** * Convert the given (internal) speed to the display speed. * @param speed the speed to convert * @return the converted speed. */ uint ConvertSpeedToDisplaySpeed(uint speed) { return (speed * units[_settings_game.locale.units].s_m) >> units[_settings_game.locale.units].s_s; } /** * Convert the given display speed to the (internal) speed. * @param speed the speed to convert * @return the converted speed. */ uint ConvertDisplaySpeedToSpeed(uint speed) { return ((speed << units[_settings_game.locale.units].s_s) + units[_settings_game.locale.units].s_m / 2) / units[_settings_game.locale.units].s_m; } static char* FormatString(char* buff, const char* str, const int64* argv, uint casei, const char* last) { WChar b; const int64 *argv_orig = argv; uint modifier = 0; while ((b = Utf8Consume(&str)) != '\0') { if (SCC_NEWGRF_FIRST <= b && b <= SCC_NEWGRF_LAST) { /* We need to pass some stuff as it might be modified; oh boy. */ b = RemapNewGRFStringControlCode(b, &buff, &str, (int64*)argv); if (b == 0) continue; } switch (b) { case SCC_SETX: // {SETX} if (buff + Utf8CharLen(SCC_SETX) + 1 < last) { buff += Utf8Encode(buff, SCC_SETX); *buff++ = *str++; } break; case SCC_SETXY: // {SETXY} if (buff + Utf8CharLen(SCC_SETXY) + 2 < last) { buff += Utf8Encode(buff, SCC_SETXY); *buff++ = *str++; *buff++ = *str++; } break; case SCC_STRING_ID: // {STRINL} buff = GetStringWithArgs(buff, Utf8Consume(&str), argv, last); break; case SCC_DATE_LONG: // {DATE_LONG} buff = FormatYmdString(buff, GetInt32(&argv), last); break; case SCC_DATE_SHORT: // {DATE_SHORT} buff = FormatMonthAndYear(buff, GetInt32(&argv), last); break; case SCC_VELOCITY: {// {VELOCITY} int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = ConvertSpeedToDisplaySpeed(GetInt32(&argv)); buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].velocity), args, modifier >> 24, last); modifier = 0; break; } case SCC_CURRENCY_COMPACT: /* {CURRCOMPACT} */ buff = FormatGenericCurrency(buff, _currency, GetInt64(&argv), true, last); break; case SCC_REVISION: /* {REV} */ buff = strecpy(buff, _openttd_revision, last); break; case SCC_CARGO_SHORT: { /* {SHORTCARGO} */ /* Short description of cargotypes. Layout: * 8-bit = cargo type * 16-bit = cargo count */ StringID cargo_str = GetCargo(GetInt32(&argv))->units_volume; switch (cargo_str) { case STR_TONS: { int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].w_m >> units[_settings_game.locale.units].w_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].l_weight), args, modifier >> 24, last); modifier = 0; break; } case STR_LITERS: { int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].v_m >> units[_settings_game.locale.units].v_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].l_volume), args, modifier >> 24, last); modifier = 0; break; } default: if (cargo_str >= 0xE000 && cargo_str < 0xF800) { /* NewGRF strings from Action 4 use a different format here, * of e.g. "x tonnes of coal", so process accordingly. */ buff = GetStringWithArgs(buff, cargo_str, argv++, last); } else { buff = FormatCommaNumber(buff, GetInt32(&argv), last); buff = strecpy(buff, " ", last); buff = strecpy(buff, GetStringPtr(cargo_str), last); } break; } } break; case SCC_STRING1: { /* {STRING1} */ /* String that consumes ONE argument */ uint str = modifier + GetInt32(&argv); buff = GetStringWithArgs(buff, str, GetArgvPtr(&argv, 1), last); modifier = 0; break; } case SCC_STRING2: { /* {STRING2} */ /* String that consumes TWO arguments */ uint str = modifier + GetInt32(&argv); buff = GetStringWithArgs(buff, str, GetArgvPtr(&argv, 2), last); modifier = 0; break; } case SCC_STRING3: { /* {STRING3} */ /* String that consumes THREE arguments */ uint str = modifier + GetInt32(&argv); buff = GetStringWithArgs(buff, str, GetArgvPtr(&argv, 3), last); modifier = 0; break; } case SCC_STRING4: { /* {STRING4} */ /* String that consumes FOUR arguments */ uint str = modifier + GetInt32(&argv); buff = GetStringWithArgs(buff, str, GetArgvPtr(&argv, 4), last); modifier = 0; break; } case SCC_STRING5: { /* {STRING5} */ /* String that consumes FIVE arguments */ uint str = modifier + GetInt32(&argv); buff = GetStringWithArgs(buff, str, GetArgvPtr(&argv, 5), last); modifier = 0; break; } case SCC_STATION_FEATURES: { /* {STATIONFEATURES} */ buff = StationGetSpecialString(buff, GetInt32(&argv), last); break; } case SCC_INDUSTRY_NAME: { /* {INDUSTRY} */ const Industry* i = GetIndustry(GetInt32(&argv)); int64 args[2]; /* industry not valid anymore? */ if (!i->IsValid()) break; /* First print the town name and the industry type name * The string STR_INDUSTRY_PATTERN controls the formatting */ args[0] = i->town->index; args[1] = GetIndustrySpec(i->type)->name; buff = FormatString(buff, GetStringPtr(STR_INDUSTRY_FORMAT), args, modifier >> 24, last); modifier = 0; break; } case SCC_VOLUME: { // {VOLUME} int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].v_m >> units[_settings_game.locale.units].v_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].l_volume), args, modifier >> 24, last); modifier = 0; break; } case SCC_GENDER_LIST: { // {G 0 Der Die Das} const char *s = GetStringPtr(argv_orig[(byte)*str++]); // contains the string that determines gender. int len; int gender = 0; if (s != NULL) { wchar_t c = Utf8Consume(&s); /* Switch case is always put before genders, so remove those bits */ if (c == SCC_SWITCH_CASE) { /* Skip to the last (i.e. default) case */ for (uint num = (byte)*s++; num != 0; num--) s += 3 + (s[1] << 8) + s[2]; c = Utf8Consume(&s); } /* Does this string have a gender, if so, set it */ if (c == SCC_GENDER_INDEX) gender = (byte)s[0]; } str = ParseStringChoice(str, gender, buff, &len); buff += len; break; } case SCC_DATE_TINY: { // {DATE_TINY} buff = FormatTinyDate(buff, GetInt32(&argv), last); break; } case SCC_CARGO: { // {CARGO} /* Layout now is: * 8bit - cargo type * 16-bit - cargo count */ CargoID cargo = GetInt32(&argv); StringID cargo_str = (cargo == CT_INVALID) ? STR_8838_N_A : GetCargo(cargo)->quantifier; buff = GetStringWithArgs(buff, cargo_str, argv++, last); break; } case SCC_POWER: { // {POWER} int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].p_m >> units[_settings_game.locale.units].p_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].power), args, modifier >> 24, last); modifier = 0; break; } case SCC_VOLUME_SHORT: { // {VOLUME_S} int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].v_m >> units[_settings_game.locale.units].v_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].s_volume), args, modifier >> 24, last); modifier = 0; break; } case SCC_WEIGHT: { // {WEIGHT} int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].w_m >> units[_settings_game.locale.units].w_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].l_weight), args, modifier >> 24, last); modifier = 0; break; } case SCC_WEIGHT_SHORT: { // {WEIGHT_S} int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].w_m >> units[_settings_game.locale.units].w_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].s_weight), args, modifier >> 24, last); modifier = 0; break; } case SCC_FORCE: { // {FORCE} int64 args[1]; assert(_settings_game.locale.units < lengthof(units)); args[0] = GetInt32(&argv) * units[_settings_game.locale.units].f_m >> units[_settings_game.locale.units].f_s; buff = FormatString(buff, GetStringPtr(units[_settings_game.locale.units].force), args, modifier >> 24, last); modifier = 0; break; } case SCC_SKIP: // {SKIP} argv++; break; /* This sets up the gender for the string. * We just ignore this one. It's used in {G 0 Der Die Das} to determine the case. */ case SCC_GENDER_INDEX: // {GENDER 0} str++; break; case SCC_STRING: {// {STRING} uint str = modifier + GetInt32(&argv); /* WARNING. It's prohibited for the included string to consume any arguments. * For included strings that consume argument, you should use STRING1, STRING2 etc. * To debug stuff you can set argv to NULL and it will tell you */ buff = GetStringWithArgs(buff, str, argv, last); modifier = 0; break; } case SCC_COMMA: // {COMMA} buff = FormatCommaNumber(buff, GetInt64(&argv), last); break; case SCC_ARG_INDEX: // Move argument pointer argv = argv_orig + (byte)*str++; break; case SCC_PLURAL_LIST: { // {P} int64 v = argv_orig[(byte)*str++]; // contains the number that determines plural int len; str = ParseStringChoice(str, DeterminePluralForm(v), buff, &len); buff += len; break; } case SCC_NUM: // {NUM} buff = FormatNoCommaNumber(buff, GetInt64(&argv), last); break; case SCC_CURRENCY: // {CURRENCY} buff = FormatGenericCurrency(buff, _currency, GetInt64(&argv), false, last); break; case SCC_WAYPOINT_NAME: { // {WAYPOINT} Waypoint *wp = GetWaypoint(GetInt32(&argv)); if (!wp->IsValid()) { // waypoint doesn't exist anymore buff = GetStringWithArgs(buff, STR_UNKNOWN_DESTINATION, NULL, last); } else if (wp->name != NULL) { buff = strecpy(buff, wp->name, last); } else { int64 temp[2]; temp[0] = wp->town_index; temp[1] = wp->town_cn + 1; StringID str = wp->town_cn == 0 ? STR_WAYPOINTNAME_CITY : STR_WAYPOINTNAME_CITY_SERIAL; buff = GetStringWithArgs(buff, str, temp, last); } break; } case SCC_STATION_NAME: { // {STATION} const Station* st = GetStation(GetInt32(&argv)); if (!st->IsValid()) { // station doesn't exist anymore buff = GetStringWithArgs(buff, STR_UNKNOWN_DESTINATION, NULL, last); } else if (st->name != NULL) { buff = strecpy(buff, st->name, last); } else { int64 temp[3]; temp[0] = STR_TOWN; temp[1] = st->town->index; temp[2] = st->index; buff = GetStringWithArgs(buff, st->string_id, temp, last); } break; } case SCC_TOWN_NAME: { // {TOWN} const Town* t = GetTown(GetInt32(&argv)); int64 temp[1]; assert(t->IsValid()); temp[0] = t->townnameparts; uint32 grfid = t->townnamegrfid; if (t->name != NULL) { buff = strecpy(buff, t->name, last); } else if (grfid == 0) { /* Original town name */ buff = GetStringWithArgs(buff, t->townnametype, temp, last); } else { /* Newgrf town name */ if (GetGRFTownName(grfid) != NULL) { /* The grf is loaded */ buff = GRFTownNameGenerate(buff, t->townnamegrfid, t->townnametype, t->townnameparts, last); } else { /* Fallback to english original */ buff = GetStringWithArgs(buff, SPECSTR_TOWNNAME_ENGLISH, temp, last); } } break; } case SCC_GROUP_NAME: { // {GROUP} const Group *g = GetGroup(GetInt32(&argv)); assert(g->IsValid()); if (g->name != NULL) { buff = strecpy(buff, g->name, last); } else { int64 args[1]; args[0] = g->index; buff = GetStringWithArgs(buff, STR_GROUP_NAME_FORMAT, args, last); } break; } case SCC_ENGINE_NAME: { // {ENGINE} EngineID engine = (EngineID)GetInt32(&argv); const Engine *e = GetEngine(engine); if (e->name != NULL) { buff = strecpy(buff, e->name, last); } else { buff = GetStringWithArgs(buff, e->info.string_id, NULL, last); } break; } case SCC_VEHICLE_NAME: { // {VEHICLE} const Vehicle *v = GetVehicle(GetInt32(&argv)); if (v->name != NULL) { buff = strecpy(buff, v->name, last); } else { int64 args[1]; args[0] = v->unitnumber; StringID str; switch (v->type) { default: NOT_REACHED(); case VEH_TRAIN: str = STR_SV_TRAIN_NAME; break; case VEH_ROAD: str = STR_SV_ROADVEH_NAME; break; case VEH_SHIP: str = STR_SV_SHIP_NAME; break; case VEH_AIRCRAFT: str = STR_SV_AIRCRAFT_NAME; break; } buff = GetStringWithArgs(buff, str, args, last); } break; } case SCC_SIGN_NAME: { // {SIGN} const Sign *si = GetSign(GetInt32(&argv)); if (si->name != NULL) { buff = strecpy(buff, si->name, last); } else { buff = GetStringWithArgs(buff, STR_280A_SIGN, NULL, last); } break; } case SCC_COMPANY_NAME: { // {COMPANY} const Player *p = GetPlayer((PlayerID)GetInt32(&argv)); if (p->name != NULL) { buff = strecpy(buff, p->name, last); } else { int64 args[1]; args[0] = p->name_2; buff = GetStringWithArgs(buff, p->name_1, args, last); } break; } case SCC_COMPANY_NUM: { // {COMPANYNUM} PlayerID player = (PlayerID)GetInt32(&argv); /* Nothing is added for AI or inactive players */ if (IsHumanPlayer(player) && IsValidPlayer(player)) { int64 args[1]; args[0] = player + 1; buff = GetStringWithArgs(buff, STR_7002_PLAYER, args, last); } break; } case SCC_PLAYER_NAME: { // {PLAYERNAME} const Player *p = GetPlayer((PlayerID)GetInt32(&argv)); if (p->president_name != NULL) { buff = strecpy(buff, p->president_name, last); } else { int64 args[1]; args[0] = p->president_name_2; buff = GetStringWithArgs(buff, p->president_name_1, args, last); } break; } case SCC_SETCASE: { // {SETCASE} /* This is a pseudo command, it's outputted when someone does {STRING.ack} * The modifier is added to all subsequent GetStringWithArgs that accept the modifier. */ modifier = (byte)*str++ << 24; break; } case SCC_SWITCH_CASE: { // {Used to implement case switching} /* <0x9E> <NUM CASES> <CASE1> <LEN1> <STRING1> <CASE2> <LEN2> <STRING2> <CASE3> <LEN3> <STRING3> <STRINGDEFAULT> * Each LEN is printed using 2 bytes in big endian order. */ uint num = (byte)*str++; while (num) { if ((byte)str[0] == casei) { /* Found the case, adjust str pointer and continue */ str += 3; break; } /* Otherwise skip to the next case */ str += 3 + (str[1] << 8) + str[2]; num--; } break; } default: if (buff + Utf8CharLen(b) < last) buff += Utf8Encode(buff, b); break; } } *buff = '\0'; return buff; } static char *StationGetSpecialString(char *buff, int x, const char* last) { if ((x & FACIL_TRAIN) && (buff + Utf8CharLen(SCC_TRAIN) < last)) buff += Utf8Encode(buff, SCC_TRAIN); if ((x & FACIL_TRUCK_STOP) && (buff + Utf8CharLen(SCC_LORRY) < last)) buff += Utf8Encode(buff, SCC_LORRY); if ((x & FACIL_BUS_STOP) && (buff + Utf8CharLen(SCC_BUS) < last)) buff += Utf8Encode(buff, SCC_BUS); if ((x & FACIL_AIRPORT) && (buff + Utf8CharLen(SCC_PLANE) < last)) buff += Utf8Encode(buff, SCC_PLANE); if ((x & FACIL_DOCK) && (buff + Utf8CharLen(SCC_SHIP) < last)) buff += Utf8Encode(buff, SCC_SHIP); *buff = '\0'; return buff; } static char *GetSpecialTownNameString(char *buff, int ind, uint32 seed, const char* last) { char name[512]; _town_name_generators[ind](name, seed, lastof(name)); return strecpy(buff, name, last); } static const char* const _silly_company_names[] = { "Bloggs Brothers", "Tiny Transport Ltd.", "Express Travel", "Comfy-Coach & Co.", "Crush & Bump Ltd.", "Broken & Late Ltd.", "Sam Speedy & Son", "Supersonic Travel", "Mike's Motors", "Lightning International", "Pannik & Loozit Ltd.", "Inter-City Transport", "Getout & Pushit Ltd." }; static const char* const _surname_list[] = { "Adams", "Allan", "Baker", "Bigwig", "Black", "Bloggs", "Brown", "Campbell", "Gordon", "Hamilton", "Hawthorn", "Higgins", "Green", "Gribble", "Jones", "McAlpine", "MacDonald", "McIntosh", "Muir", "Murphy", "Nelson", "O'Donnell", "Parker", "Phillips", "Pilkington", "Quigley", "Sharkey", "Thomson", "Watkins" }; static const char* const _silly_surname_list[] = { "Grumpy", "Dozy", "Speedy", "Nosey", "Dribble", "Mushroom", "Cabbage", "Sniffle", "Fishy", "Swindle", "Sneaky", "Nutkins" }; static const char _initial_name_letters[] = { 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'P', 'R', 'S', 'T', 'W', }; static char *GenAndCoName(char *buff, uint32 arg, const char* last) { const char* const* base; uint num; if (_settings_game.game_creation.landscape == LT_TOYLAND) { base = _silly_surname_list; num = lengthof(_silly_surname_list); } else { base = _surname_list; num = lengthof(_surname_list); } buff = strecpy(buff, base[num * GB(arg, 16, 8) >> 8], last); buff = strecpy(buff, " & Co.", last); return buff; } static char *GenPresidentName(char *buff, uint32 x, const char* last) { char initial[] = "?. "; const char* const* base; uint num; uint i; initial[0] = _initial_name_letters[sizeof(_initial_name_letters) * GB(x, 0, 8) >> 8]; buff = strecpy(buff, initial, last); i = (sizeof(_initial_name_letters) + 35) * GB(x, 8, 8) >> 8; if (i < sizeof(_initial_name_letters)) { initial[0] = _initial_name_letters[i]; buff = strecpy(buff, initial, last); } if (_settings_game.game_creation.landscape == LT_TOYLAND) { base = _silly_surname_list; num = lengthof(_silly_surname_list); } else { base = _surname_list; num = lengthof(_surname_list); } buff = strecpy(buff, base[num * GB(x, 16, 8) >> 8], last); return buff; } static char *GetSpecialPlayerNameString(char *buff, int ind, const int64 *argv, const char* last) { switch (ind) { case 1: // not used return strecpy(buff, _silly_company_names[GetInt32(&argv) & 0xFFFF], last); case 2: // used for Foobar & Co company names return GenAndCoName(buff, GetInt32(&argv), last); case 3: // President name return GenPresidentName(buff, GetInt32(&argv), last); case 4: // song names return strecpy(buff, origin_songs_specs[GetInt32(&argv) - 1].song_name, last); } /* town name? */ if (IsInsideMM(ind - 6, 0, SPECSTR_TOWNNAME_LAST - SPECSTR_TOWNNAME_START + 1)) { buff = GetSpecialTownNameString(buff, ind - 6, GetInt32(&argv), last); return strecpy(buff, " Transport", last); } /* language name? */ if (IsInsideMM(ind, (SPECSTR_LANGUAGE_START - 0x70E4), (SPECSTR_LANGUAGE_END - 0x70E4) + 1)) { int i = ind - (SPECSTR_LANGUAGE_START - 0x70E4); return strecpy(buff, i == _dynlang.curr ? _langpack->own_name : _dynlang.ent[i].name, last); } /* resolution size? */ if (IsInsideMM(ind, (SPECSTR_RESOLUTION_START - 0x70E4), (SPECSTR_RESOLUTION_END - 0x70E4) + 1)) { int i = ind - (SPECSTR_RESOLUTION_START - 0x70E4); buff += snprintf( buff, last - buff + 1, "%dx%d", _resolutions[i].width, _resolutions[i].height ); return buff; } /* screenshot format name? */ if (IsInsideMM(ind, (SPECSTR_SCREENSHOT_START - 0x70E4), (SPECSTR_SCREENSHOT_END - 0x70E4) + 1)) { int i = ind - (SPECSTR_SCREENSHOT_START - 0x70E4); return strecpy(buff, GetScreenshotFormatDesc(i), last); } assert(0); return NULL; } #ifdef ENABLE_NETWORK extern void SortNetworkLanguages(); #else /* ENABLE_NETWORK */ static inline void SortNetworkLanguages() {} #endif /* ENABLE_NETWORK */ bool ReadLanguagePack(int lang_index) { int tot_count, i; size_t len; char **langpack_offs; char *s; LanguagePack *lang_pack = (LanguagePack*)ReadFileToMem(_dynlang.ent[lang_index].file, &len, 200000); if (lang_pack == NULL) return false; if (len < sizeof(LanguagePack) || lang_pack->ident != TO_LE32(LANGUAGE_PACK_IDENT) || lang_pack->version != TO_LE32(LANGUAGE_PACK_VERSION)) { free(lang_pack); return false; } #if TTD_ENDIAN == TTD_BIG_ENDIAN for (i = 0; i != 32; i++) { lang_pack->offsets[i] = ReadLE16Aligned(&lang_pack->offsets[i]); } #endif /* TTD_ENDIAN == TTD_BIG_ENDIAN */ tot_count = 0; for (i = 0; i != 32; i++) { uint num = lang_pack->offsets[i]; _langtab_start[i] = tot_count; _langtab_num[i] = num; tot_count += num; } /* Allocate offsets */ langpack_offs = MallocT<char*>(tot_count); /* Fill offsets */ s = lang_pack->data; for (i = 0; i != tot_count; i++) { len = (byte)*s; *s++ = '\0'; // zero terminate the string before. if (len >= 0xC0) len = ((len & 0x3F) << 8) + (byte)*s++; langpack_offs[i] = s; s += len; } free(_langpack); _langpack = lang_pack; free(_langpack_offs); _langpack_offs = langpack_offs; const char *c_file = strrchr(_dynlang.ent[lang_index].file, PATHSEPCHAR) + 1; ttd_strlcpy(_dynlang.curr_file, c_file, lengthof(_dynlang.curr_file)); _dynlang.curr = lang_index; SetCurrentGrfLangID(_langpack->isocode); SortNetworkLanguages(); return true; } /* Win32 implementation in win32.cpp. */ /* OS X implementation in os/macosx/macos.mm. */ #if !(defined(WIN32) || defined(__APPLE__)) /** Determine the current charset based on the environment * First check some default values, after this one we passed ourselves * and if none exist return the value for $LANG * @param param environment variable to check conditionally if default ones are not * set. Pass NULL if you don't want additional checks. * @return return string containing current charset, or NULL if not-determinable */ const char *GetCurrentLocale(const char *param) { const char *env; env = getenv("LANGUAGE"); if (env != NULL) return env; env = getenv("LC_ALL"); if (env != NULL) return env; if (param != NULL) { env = getenv(param); if (env != NULL) return env; } return getenv("LANG"); } #else const char *GetCurrentLocale(const char *param); #endif /* !(defined(WIN32) || defined(__APPLE__)) */ int CDECL StringIDSorter(const void *a, const void *b) { const StringID va = *(const StringID*)a; const StringID vb = *(const StringID*)b; char stra[512]; char strb[512]; GetString(stra, va, lastof(stra)); GetString(strb, vb, lastof(strb)); return strcmp(stra, strb); } /** * Checks whether the given language is already found. * @param langs languages we've found so fa * @param max the length of the language list * @param language name of the language to check * @return true if and only if a language file with the same name has not been found */ static bool UniqueLanguageFile(const Language *langs, uint max, const char *language) { for (uint i = 0; i < max; i++) { const char *f_name = strrchr(langs[i].file, PATHSEPCHAR) + 1; if (strcmp(f_name, language) == 0) return false; // duplicates } return true; } /** * Reads the language file header and checks compatability. * @param file the file to read * @param hdr the place to write the header information to * @return true if and only if the language file is of a compatible version */ static bool GetLanguageFileHeader(const char *file, LanguagePack *hdr) { FILE *f = fopen(file, "rb"); if (f == NULL) return false; size_t read = fread(hdr, sizeof(*hdr), 1, f); fclose(f); return read == 1 && hdr->ident == TO_LE32(LANGUAGE_PACK_IDENT) && hdr->version == TO_LE32(LANGUAGE_PACK_VERSION); } /** * Gets a list of languages from the given directory. * @param langs the list to write to * @param start the initial offset in the list * @param max the length of the language list * @param path the base directory to search in * @return the number of added languages */ static int GetLanguageList(Language *langs, int start, int max, const char *path) { int i = start; DIR *dir = ttd_opendir(path); if (dir != NULL) { struct dirent *dirent; while ((dirent = readdir(dir)) != NULL && i < max) { const char *d_name = FS2OTTD(dirent->d_name); const char *extension = strrchr(d_name, '.'); /* Not a language file */ if (extension == NULL || strcmp(extension, ".lng") != 0) continue; /* Filter any duplicate language-files, first-come first-serve */ if (!UniqueLanguageFile(langs, i, d_name)) continue; langs[i].file = str_fmt("%s%s", path, d_name); /* Check whether the file is of the correct version */ LanguagePack hdr; if (!GetLanguageFileHeader(langs[i].file, &hdr)) { free(langs[i].file); continue; } i++; } closedir(dir); } return i - start; } /** * Make a list of the available language packs. put the data in * _dynlang struct. */ void InitializeLanguagePacks() { Searchpath sp; Language files[MAX_LANG]; uint language_count = 0; FOR_ALL_SEARCHPATHS(sp) { char path[MAX_PATH]; FioAppendDirectory(path, lengthof(path), sp, LANG_DIR); language_count += GetLanguageList(files, language_count, lengthof(files), path); } if (language_count == 0) usererror("No available language packs (invalid versions?)"); /* Acquire the locale of the current system */ const char *lang = GetCurrentLocale("LC_MESSAGES"); if (lang == NULL) lang = "en_GB"; int chosen_language = -1; ///< Matching the language in the configuartion file or the current locale int language_fallback = -1; ///< Using pt_PT for pt_BR locale when pt_BR is not available int en_GB_fallback = 0; ///< Fallback when no locale-matching language has been found DynamicLanguages *dl = &_dynlang; dl->num = 0; /* Fill the dynamic languages structures */ for (uint i = 0; i < language_count; i++) { /* File read the language header */ LanguagePack hdr; if (!GetLanguageFileHeader(files[i].file, &hdr)) continue; dl->ent[dl->num].file = files[i].file; dl->ent[dl->num].name = strdup(hdr.name); /* We are trying to find a default language. The priority is by * configuration file, local environment and last, if nothing found, * english. If def equals -1, we have not picked a default language */ const char *lang_file = strrchr(dl->ent[dl->num].file, PATHSEPCHAR) + 1; if (strcmp(lang_file, dl->curr_file) == 0) chosen_language = dl->num; if (chosen_language == -1) { if (strcmp (hdr.isocode, "en_GB") == 0) en_GB_fallback = dl->num; if (strncmp(hdr.isocode, lang, 5) == 0) chosen_language = dl->num; if (strncmp(hdr.isocode, lang, 2) == 0) language_fallback = dl->num; } dl->num++; } if (dl->num == 0) usererror("Invalid version of language packs"); /* We haven't found the language in the config nor the one in the locale. * Now we set it to one of the fallback languages */ if (chosen_language == -1) { chosen_language = (language_fallback != -1) ? language_fallback : en_GB_fallback; } if (!ReadLanguagePack(chosen_language)) usererror("Can't read language pack '%s'", dl->ent[chosen_language].file); } /** * Check whether the currently loaded language pack * uses characters that the currently loaded font * does not support. If this is the case an error * message will be shown in English. The error * message will not be localized because that would * mean it might use characters that are not in the * font, which is the whole reason this check has * been added. */ void CheckForMissingGlyphsInLoadedLanguagePack() { const Sprite *question_mark = GetGlyph(FS_NORMAL, '?'); for (uint i = 0; i != 32; i++) { for (uint j = 0; j < _langtab_num[i]; j++) { const char *string = _langpack_offs[_langtab_start[i] + j]; WChar c; while ((c = Utf8Consume(&string)) != '\0') { if (c == SCC_SETX) { /* * SetX is, together with SetXY as special character that * uses the next (two) characters as data points. We have * to skip those, otherwise the UTF8 reading will go * haywire. */ string++; } else if (c == SCC_SETXY) { string += 2; } else if (IsPrintable(c) && c != '?' && GetGlyph(FS_NORMAL, c) == question_mark) { /* * The character is printable, but not in the normal font. * This is the case we were testing for. In this case we * have to show the error. As we do not want the string to * be translated by the translators, we 'force' it into the * binary and 'load' it via a BindCString. To do this * properly we have to set the color of the string, * otherwise we end up with a lot of artefacts. The color * 'character' might change in the future, so for safety * we just Utf8 Encode it into the string, which takes * exactly three characters, so it replaces the "XXX" with * the color marker. */ static char *err_str = strdup("XXXThe current font is missing some of the characters used in the texts for this language. Read the readme to see how to solve this."); Utf8Encode(err_str, SCC_YELLOW); StringID err_msg = BindCString(err_str); ShowErrorMessage(INVALID_STRING_ID, err_msg, 0, 0); return; } } } } } /* --- Handling of saving/loading string IDs from old savegames --- */ /** * Remap a string ID from the old format to the new format * @param s StringID that requires remapping * @return translated ID */ StringID RemapOldStringID(StringID s) { switch (s) { case 0x0006: return STR_SV_EMPTY; case 0x7000: return STR_SV_UNNAMED; case 0x70E4: return SPECSTR_PLAYERNAME_ENGLISH; case 0x70E9: return SPECSTR_PLAYERNAME_ENGLISH; case 0x8864: return STR_SV_TRAIN_NAME; case 0x902B: return STR_SV_ROADVEH_NAME; case 0x9830: return STR_SV_SHIP_NAME; case 0xA02F: return STR_SV_AIRCRAFT_NAME; default: if (IsInsideMM(s, 0x300F, 0x3030)) { return s - 0x300F + STR_SV_STNAME; } else { return s; } } } /** Location to load the old names to. */ char *_old_name_array = NULL; /** * Copy and convert old custom names to UTF-8. * They were all stored in a 512 by 32 long string array and are * now stored with stations, waypoints and other places with names. * @param id the StringID of the custom name to clone. * @return the clones custom name. */ char *CopyFromOldName(StringID id) { /* Is this name an (old) custom name? */ if (GB(id, 11, 5) != 15) return NULL; if (CheckSavegameVersion(37)) { /* Old names were 32 characters long, so 128 characters should be * plenty to allow for expansion when converted to UTF-8. */ char tmp[128]; const char *strfrom = &_old_name_array[32 * GB(id, 0, 9)]; char *strto = tmp; for (; *strfrom != '\0'; strfrom++) { WChar c = (byte)*strfrom; /* Map from non-ISO8859-15 characters to UTF-8. */ switch (c) { case 0xA4: c = 0x20AC; break; // Euro case 0xA6: c = 0x0160; break; // S with caron case 0xA8: c = 0x0161; break; // s with caron case 0xB4: c = 0x017D; break; // Z with caron case 0xB8: c = 0x017E; break; // z with caron case 0xBC: c = 0x0152; break; // OE ligature case 0xBD: c = 0x0153; break; // oe ligature case 0xBE: c = 0x0178; break; // Y with diaresis default: break; } /* Check character will fit into our buffer. */ if (strto + Utf8CharLen(c) > lastof(tmp)) break; strto += Utf8Encode(strto, c); } /* Terminate the new string and copy it back to the name array */ *strto = '\0'; return strdup(tmp); } else { /* Name will already be in UTF-8. */ return strdup(&_old_name_array[32 * GB(id, 0, 9)]); } } /** * Free the memory of the old names array. * Should be called once the old names have all been converted. */ void ResetOldNames() { free(_old_name_array); _old_name_array = NULL; } /** * Initialize the old names table memory. */ void InitializeOldNames() { free(_old_name_array); _old_name_array = CallocT<char>(512 * 32); } static void Load_NAME() { int index; while ((index = SlIterateArray()) != -1) { SlArray(&_old_name_array[32 * index], SlGetFieldLength(), SLE_UINT8); } } extern const ChunkHandler _name_chunk_handlers[] = { { 'NAME', NULL, Load_NAME, CH_ARRAY | CH_LAST}, };