| /* |
| * Locale-dependent format handling |
| * |
| * Copyright 1995 Martin von Loewis |
| * Copyright 1998 David Lee Lambert |
| * Copyright 2000 Julio César Gázquez |
| * Copyright 2003 Jon Griffiths |
| * Copyright 2005 Dmitry Timoshkov |
| * |
| * This library is free software; you can redistribute it and/or |
| * modify it under the terms of the GNU Lesser General Public |
| * License as published by the Free Software Foundation; either |
| * version 2.1 of the License, or (at your option) any later version. |
| * |
| * This library is distributed in the hope that it will be useful, |
| * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| * Lesser General Public License for more details. |
| * |
| * You should have received a copy of the GNU Lesser General Public |
| * License along with this library; if not, write to the Free Software |
| * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA |
| */ |
| |
| #include "config.h" |
| #include "wine/port.h" |
| |
| #include <string.h> |
| #include <stdarg.h> |
| #include <stdio.h> |
| #include <stdlib.h> |
| |
| #include "windef.h" |
| #include "winbase.h" |
| #include "wine/unicode.h" |
| #include "wine/debug.h" |
| #include "winternl.h" |
| |
| #include "kernel_private.h" |
| |
| WINE_DEFAULT_DEBUG_CHANNEL(nls); |
| |
| #define DATE_DATEVARSONLY 0x0100 /* only date stuff: yMdg */ |
| #define TIME_TIMEVARSONLY 0x0200 /* only time stuff: hHmst */ |
| |
| /* Since calculating the formatting data for each locale is time-consuming, |
| * we get the format data for each locale only once and cache it in memory. |
| * We cache both the system default and user overridden data, after converting |
| * them into the formats that the functions here expect. Since these functions |
| * will typically be called with only a small number of the total locales |
| * installed, the memory overhead is minimal while the speedup is significant. |
| * |
| * Our cache takes the form of a singly linked list, whose node is below: |
| */ |
| #define NLS_NUM_CACHED_STRINGS 45 |
| |
| typedef struct _NLS_FORMAT_NODE |
| { |
| LCID lcid; /* Locale Id */ |
| DWORD dwFlags; /* 0 or LOCALE_NOUSEROVERRIDE */ |
| DWORD dwCodePage; /* Default code page (if LOCALE_USE_ANSI_CP not given) */ |
| NUMBERFMTW fmt; /* Default format for numbers */ |
| CURRENCYFMTW cyfmt; /* Default format for currencies */ |
| LPWSTR lppszStrings[NLS_NUM_CACHED_STRINGS]; /* Default formats,day/month names */ |
| WCHAR szShortAM[2]; /* Short 'AM' marker */ |
| WCHAR szShortPM[2]; /* Short 'PM' marker */ |
| struct _NLS_FORMAT_NODE *next; |
| } NLS_FORMAT_NODE; |
| |
| /* Macros to get particular data strings from a format node */ |
| #define GetNegative(fmt) fmt->lppszStrings[0] |
| #define GetLongDate(fmt) fmt->lppszStrings[1] |
| #define GetShortDate(fmt) fmt->lppszStrings[2] |
| #define GetTime(fmt) fmt->lppszStrings[3] |
| #define GetAM(fmt) fmt->lppszStrings[42] |
| #define GetPM(fmt) fmt->lppszStrings[43] |
| #define GetYearMonth(fmt) fmt->lppszStrings[44] |
| |
| #define GetLongDay(fmt,day) fmt->lppszStrings[4 + day] |
| #define GetShortDay(fmt,day) fmt->lppszStrings[11 + day] |
| #define GetLongMonth(fmt,mth) fmt->lppszStrings[18 + mth] |
| #define GetShortMonth(fmt,mth) fmt->lppszStrings[30 + mth] |
| |
| /* Write access to the cache is protected by this critical section */ |
| static CRITICAL_SECTION NLS_FormatsCS; |
| static CRITICAL_SECTION_DEBUG NLS_FormatsCS_debug = |
| { |
| 0, 0, &NLS_FormatsCS, |
| { &NLS_FormatsCS_debug.ProcessLocksList, |
| &NLS_FormatsCS_debug.ProcessLocksList }, |
| 0, 0, { (DWORD_PTR)(__FILE__ ": NLS_Formats") } |
| }; |
| static CRITICAL_SECTION NLS_FormatsCS = { &NLS_FormatsCS_debug, -1, 0, 0, 0, 0 }; |
| |
| /************************************************************************** |
| * NLS_GetLocaleNumber <internal> |
| * |
| * Get a numeric locale format value. |
| */ |
| static DWORD NLS_GetLocaleNumber(LCID lcid, DWORD dwFlags) |
| { |
| WCHAR szBuff[80]; |
| DWORD dwVal = 0; |
| |
| szBuff[0] = '\0'; |
| GetLocaleInfoW(lcid, dwFlags, szBuff, sizeof(szBuff) / sizeof(WCHAR)); |
| |
| if (szBuff[0] && szBuff[1] == ';' && szBuff[2] != '0') |
| dwVal = (szBuff[0] - '0') * 10 + (szBuff[2] - '0'); |
| else |
| { |
| const WCHAR* iter = szBuff; |
| dwVal = 0; |
| while(*iter >= '0' && *iter <= '9') |
| dwVal = dwVal * 10 + (*iter++ - '0'); |
| } |
| return dwVal; |
| } |
| |
| /************************************************************************** |
| * NLS_GetLocaleString <internal> |
| * |
| * Get a string locale format value. |
| */ |
| static WCHAR* NLS_GetLocaleString(LCID lcid, DWORD dwFlags) |
| { |
| WCHAR szBuff[80], *str; |
| DWORD dwLen; |
| |
| szBuff[0] = '\0'; |
| GetLocaleInfoW(lcid, dwFlags, szBuff, sizeof(szBuff) / sizeof(WCHAR)); |
| dwLen = strlenW(szBuff) + 1; |
| str = HeapAlloc(GetProcessHeap(), 0, dwLen * sizeof(WCHAR)); |
| if (str) |
| memcpy(str, szBuff, dwLen * sizeof(WCHAR)); |
| return str; |
| } |
| |
| #define GET_LOCALE_NUMBER(num, type) num = NLS_GetLocaleNumber(lcid, type|dwFlags); \ |
| TRACE( #type ": %ld (%08lx)\n", (DWORD)num, (DWORD)num) |
| |
| #define GET_LOCALE_STRING(str, type) str = NLS_GetLocaleString(lcid, type|dwFlags); \ |
| TRACE( #type ": '%s'\n", debugstr_w(str)) |
| |
| /************************************************************************** |
| * NLS_GetFormats <internal> |
| * |
| * Calculate (and cache) the number formats for a locale. |
| */ |
| static const NLS_FORMAT_NODE *NLS_GetFormats(LCID lcid, DWORD dwFlags) |
| { |
| /* GetLocaleInfo() identifiers for cached formatting strings */ |
| static const USHORT NLS_LocaleIndices[] = { |
| LOCALE_SNEGATIVESIGN, |
| LOCALE_SLONGDATE, LOCALE_SSHORTDATE, |
| LOCALE_STIMEFORMAT, |
| LOCALE_SDAYNAME1, LOCALE_SDAYNAME2, LOCALE_SDAYNAME3, |
| LOCALE_SDAYNAME4, LOCALE_SDAYNAME5, LOCALE_SDAYNAME6, LOCALE_SDAYNAME7, |
| LOCALE_SABBREVDAYNAME1, LOCALE_SABBREVDAYNAME2, LOCALE_SABBREVDAYNAME3, |
| LOCALE_SABBREVDAYNAME4, LOCALE_SABBREVDAYNAME5, LOCALE_SABBREVDAYNAME6, |
| LOCALE_SABBREVDAYNAME7, |
| LOCALE_SMONTHNAME1, LOCALE_SMONTHNAME2, LOCALE_SMONTHNAME3, |
| LOCALE_SMONTHNAME4, LOCALE_SMONTHNAME5, LOCALE_SMONTHNAME6, |
| LOCALE_SMONTHNAME7, LOCALE_SMONTHNAME8, LOCALE_SMONTHNAME9, |
| LOCALE_SMONTHNAME10, LOCALE_SMONTHNAME11, LOCALE_SMONTHNAME12, |
| LOCALE_SABBREVMONTHNAME1, LOCALE_SABBREVMONTHNAME2, LOCALE_SABBREVMONTHNAME3, |
| LOCALE_SABBREVMONTHNAME4, LOCALE_SABBREVMONTHNAME5, LOCALE_SABBREVMONTHNAME6, |
| LOCALE_SABBREVMONTHNAME7, LOCALE_SABBREVMONTHNAME8, LOCALE_SABBREVMONTHNAME9, |
| LOCALE_SABBREVMONTHNAME10, LOCALE_SABBREVMONTHNAME11, LOCALE_SABBREVMONTHNAME12, |
| LOCALE_S1159, LOCALE_S2359, |
| LOCALE_SYEARMONTH |
| }; |
| static NLS_FORMAT_NODE *NLS_CachedFormats = NULL; |
| NLS_FORMAT_NODE *node = NLS_CachedFormats; |
| |
| dwFlags &= LOCALE_NOUSEROVERRIDE; |
| |
| TRACE("(0x%04lx,0x%08lx)\n", lcid, dwFlags); |
| |
| /* See if we have already cached the locales number format */ |
| while (node && (node->lcid != lcid || node->dwFlags != dwFlags) && node->next) |
| node = node->next; |
| |
| if (!node || node->lcid != lcid || node->dwFlags != dwFlags) |
| { |
| NLS_FORMAT_NODE *new_node; |
| DWORD i; |
| |
| TRACE("Creating new cache entry\n"); |
| |
| if (!(new_node = HeapAlloc(GetProcessHeap(), 0, sizeof(NLS_FORMAT_NODE)))) |
| return NULL; |
| |
| GET_LOCALE_NUMBER(new_node->dwCodePage, LOCALE_IDEFAULTANSICODEPAGE); |
| |
| /* Number Format */ |
| new_node->lcid = lcid; |
| new_node->dwFlags = dwFlags; |
| new_node->next = NULL; |
| |
| GET_LOCALE_NUMBER(new_node->fmt.NumDigits, LOCALE_IDIGITS); |
| GET_LOCALE_NUMBER(new_node->fmt.LeadingZero, LOCALE_ILZERO); |
| GET_LOCALE_NUMBER(new_node->fmt.NegativeOrder, LOCALE_INEGNUMBER); |
| |
| GET_LOCALE_NUMBER(new_node->fmt.Grouping, LOCALE_SGROUPING); |
| if (new_node->fmt.Grouping > 9 && new_node->fmt.Grouping != 32) |
| { |
| WARN("LOCALE_SGROUPING (%d) unhandled, please report!\n", |
| new_node->fmt.Grouping); |
| new_node->fmt.Grouping = 0; |
| } |
| |
| GET_LOCALE_STRING(new_node->fmt.lpDecimalSep, LOCALE_SDECIMAL); |
| GET_LOCALE_STRING(new_node->fmt.lpThousandSep, LOCALE_STHOUSAND); |
| |
| /* Currency Format */ |
| new_node->cyfmt.NumDigits = new_node->fmt.NumDigits; |
| new_node->cyfmt.LeadingZero = new_node->fmt.LeadingZero; |
| |
| GET_LOCALE_NUMBER(new_node->cyfmt.Grouping, LOCALE_SGROUPING); |
| |
| if (new_node->cyfmt.Grouping > 9) |
| { |
| WARN("LOCALE_SMONGROUPING (%d) unhandled, please report!\n", |
| new_node->cyfmt.Grouping); |
| new_node->cyfmt.Grouping = 0; |
| } |
| |
| GET_LOCALE_NUMBER(new_node->cyfmt.NegativeOrder, LOCALE_INEGCURR); |
| if (new_node->cyfmt.NegativeOrder > 15) |
| { |
| WARN("LOCALE_INEGCURR (%d) unhandled, please report!\n", |
| new_node->cyfmt.NegativeOrder); |
| new_node->cyfmt.NegativeOrder = 0; |
| } |
| GET_LOCALE_NUMBER(new_node->cyfmt.PositiveOrder, LOCALE_ICURRENCY); |
| if (new_node->cyfmt.PositiveOrder > 3) |
| { |
| WARN("LOCALE_IPOSCURR (%d) unhandled,please report!\n", |
| new_node->cyfmt.PositiveOrder); |
| new_node->cyfmt.PositiveOrder = 0; |
| } |
| GET_LOCALE_STRING(new_node->cyfmt.lpDecimalSep, LOCALE_SMONDECIMALSEP); |
| GET_LOCALE_STRING(new_node->cyfmt.lpThousandSep, LOCALE_SMONTHOUSANDSEP); |
| GET_LOCALE_STRING(new_node->cyfmt.lpCurrencySymbol, LOCALE_SCURRENCY); |
| |
| /* Date/Time Format info, negative character, etc */ |
| for (i = 0; i < sizeof(NLS_LocaleIndices)/sizeof(NLS_LocaleIndices[0]); i++) |
| { |
| GET_LOCALE_STRING(new_node->lppszStrings[i], NLS_LocaleIndices[i]); |
| } |
| new_node->szShortAM[0] = GetAM(new_node)[0]; new_node->szShortAM[1] = '\0'; |
| new_node->szShortPM[0] = GetPM(new_node)[0]; new_node->szShortPM[1] = '\0'; |
| |
| /* Now add the computed format to the cache */ |
| RtlEnterCriticalSection(&NLS_FormatsCS); |
| |
| /* Search again: We may have raced to add the node */ |
| node = NLS_CachedFormats; |
| while (node && (node->lcid != lcid || node->dwFlags != dwFlags) && node->next) |
| node = node->next; |
| |
| if (!node) |
| { |
| node = NLS_CachedFormats = new_node; /* Empty list */ |
| new_node = NULL; |
| } |
| else if (node->lcid != lcid || node->dwFlags != dwFlags) |
| { |
| node->next = new_node; /* Not in the list, add to end */ |
| node = new_node; |
| new_node = NULL; |
| } |
| |
| RtlLeaveCriticalSection(&NLS_FormatsCS); |
| |
| if (new_node) |
| { |
| /* We raced and lost: The node was already added by another thread. |
| * node points to the currently cached node, so free new_node. |
| */ |
| for (i = 0; i < sizeof(NLS_LocaleIndices)/sizeof(NLS_LocaleIndices[0]); i++) |
| HeapFree(GetProcessHeap(), 0, new_node->lppszStrings[i]); |
| HeapFree(GetProcessHeap(), 0, new_node->fmt.lpDecimalSep); |
| HeapFree(GetProcessHeap(), 0, new_node->fmt.lpThousandSep); |
| HeapFree(GetProcessHeap(), 0, new_node->cyfmt.lpDecimalSep); |
| HeapFree(GetProcessHeap(), 0, new_node->cyfmt.lpThousandSep); |
| HeapFree(GetProcessHeap(), 0, new_node->cyfmt.lpCurrencySymbol); |
| HeapFree(GetProcessHeap(), 0, new_node); |
| } |
| } |
| return node; |
| } |
| |
| /************************************************************************** |
| * NLS_IsUnicodeOnlyLcid <internal> |
| * |
| * Determine if a locale is Unicode only, and thus invalid in ASCII calls. |
| */ |
| BOOL NLS_IsUnicodeOnlyLcid(LCID lcid) |
| { |
| switch (PRIMARYLANGID(lcid)) |
| { |
| case LANG_ARMENIAN: |
| case LANG_DIVEHI: |
| case LANG_GEORGIAN: |
| case LANG_GUJARATI: |
| case LANG_HINDI: |
| case LANG_KANNADA: |
| case LANG_KONKANI: |
| case LANG_MARATHI: |
| case LANG_PUNJABI: |
| case LANG_SANSKRIT: |
| TRACE("lcid 0x%08lx: langid 0x%4x is Unicode Only\n", lcid, PRIMARYLANGID(lcid)); |
| return TRUE; |
| default: |
| return FALSE; |
| } |
| } |
| |
| /* |
| * Formatting of dates, times, numbers and currencies. |
| */ |
| |
| #define IsLiteralMarker(p) (p == '\'') |
| #define IsDateFmtChar(p) (p == 'd'||p == 'M'||p == 'y'||p == 'g') |
| #define IsTimeFmtChar(p) (p == 'H'||p == 'h'||p == 'm'||p == 's'||p == 't') |
| |
| /* Only the following flags can be given if a date/time format is specified */ |
| #define DATE_FORMAT_FLAGS (DATE_DATEVARSONLY|LOCALE_NOUSEROVERRIDE) |
| #define TIME_FORMAT_FLAGS (TIME_TIMEVARSONLY|TIME_FORCE24HOURFORMAT| \ |
| TIME_NOMINUTESORSECONDS|TIME_NOSECONDS| \ |
| TIME_NOTIMEMARKER|LOCALE_NOUSEROVERRIDE) |
| |
| /****************************************************************************** |
| * NLS_GetDateTimeFormatW <internal> |
| * |
| * Performs the formatting for GetDateFormatW/GetTimeFormatW. |
| * |
| * FIXME |
| * DATE_USE_ALT_CALENDAR - Requires GetCalendarInfo to work first. |
| * DATE_LTRREADING/DATE_RTLREADING - Not yet implemented. |
| */ |
| static INT NLS_GetDateTimeFormatW(LCID lcid, DWORD dwFlags, |
| const SYSTEMTIME* lpTime, LPCWSTR lpFormat, |
| LPWSTR lpStr, INT cchOut) |
| { |
| const NLS_FORMAT_NODE *node; |
| SYSTEMTIME st; |
| INT cchWritten = 0; |
| INT lastFormatPos = 0; |
| BOOL bSkipping = FALSE; /* Skipping text around marker? */ |
| |
| /* Verify our arguments */ |
| if ((cchOut && !lpStr) || !(node = NLS_GetFormats(lcid, dwFlags))) |
| { |
| NLS_GetDateTimeFormatW_InvalidParameter: |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return 0; |
| } |
| |
| if (dwFlags & ~(DATE_DATEVARSONLY|TIME_TIMEVARSONLY)) |
| { |
| if (lpFormat && |
| ((dwFlags & DATE_DATEVARSONLY && dwFlags & ~DATE_FORMAT_FLAGS) || |
| (dwFlags & TIME_TIMEVARSONLY && dwFlags & ~TIME_FORMAT_FLAGS))) |
| { |
| NLS_GetDateTimeFormatW_InvalidFlags: |
| SetLastError(ERROR_INVALID_FLAGS); |
| return 0; |
| } |
| |
| if (dwFlags & DATE_DATEVARSONLY) |
| { |
| if ((dwFlags & (DATE_LTRREADING|DATE_RTLREADING)) == (DATE_LTRREADING|DATE_RTLREADING)) |
| goto NLS_GetDateTimeFormatW_InvalidFlags; |
| else if (dwFlags & (DATE_LTRREADING|DATE_RTLREADING)) |
| FIXME("Unsupported flags: DATE_LTRREADING/DATE_RTLREADING\n"); |
| |
| switch (dwFlags & (DATE_SHORTDATE|DATE_LONGDATE|DATE_YEARMONTH)) |
| { |
| case 0: |
| break; |
| case DATE_SHORTDATE: |
| case DATE_LONGDATE: |
| case DATE_YEARMONTH: |
| if (lpFormat) |
| goto NLS_GetDateTimeFormatW_InvalidFlags; |
| break; |
| default: |
| goto NLS_GetDateTimeFormatW_InvalidFlags; |
| } |
| } |
| } |
| |
| if (!lpFormat) |
| { |
| /* Use the appropriate default format */ |
| if (dwFlags & DATE_DATEVARSONLY) |
| { |
| if (dwFlags & DATE_YEARMONTH) |
| lpFormat = GetYearMonth(node); |
| else if (dwFlags & DATE_LONGDATE) |
| lpFormat = GetLongDate(node); |
| else |
| lpFormat = GetShortDate(node); |
| } |
| else |
| lpFormat = GetTime(node); |
| } |
| |
| if (!lpTime) |
| { |
| GetLocalTime(&st); /* Default to current time */ |
| lpTime = &st; |
| } |
| else |
| { |
| if (dwFlags & DATE_DATEVARSONLY) |
| { |
| FILETIME ftTmp; |
| |
| /* Verify the date and correct the D.O.W. if needed */ |
| memset(&st, 0, sizeof(st)); |
| st.wYear = lpTime->wYear; |
| st.wMonth = lpTime->wMonth; |
| st.wDay = lpTime->wDay; |
| |
| if (st.wDay > 31 || st.wMonth > 12 || !SystemTimeToFileTime(&st, &ftTmp)) |
| goto NLS_GetDateTimeFormatW_InvalidParameter; |
| |
| FileTimeToSystemTime(&ftTmp, &st); |
| lpTime = &st; |
| } |
| |
| if (dwFlags & TIME_TIMEVARSONLY) |
| { |
| /* Verify the time */ |
| if (lpTime->wHour > 24 || lpTime->wMinute > 59 || lpTime->wSecond > 59) |
| goto NLS_GetDateTimeFormatW_InvalidParameter; |
| } |
| } |
| |
| /* Format the output */ |
| while (*lpFormat) |
| { |
| if (IsLiteralMarker(*lpFormat)) |
| { |
| /* Start of a literal string */ |
| lpFormat++; |
| |
| /* Loop until the end of the literal marker or end of the string */ |
| while (*lpFormat) |
| { |
| if (IsLiteralMarker(*lpFormat)) |
| { |
| lpFormat++; |
| if (!IsLiteralMarker(*lpFormat)) |
| break; /* Terminating literal marker */ |
| } |
| |
| if (!cchOut) |
| cchWritten++; /* Count size only */ |
| else if (cchWritten >= cchOut) |
| goto NLS_GetDateTimeFormatW_Overrun; |
| else if (!bSkipping) |
| { |
| lpStr[cchWritten] = *lpFormat; |
| cchWritten++; |
| } |
| lpFormat++; |
| } |
| } |
| else if ((dwFlags & DATE_DATEVARSONLY && IsDateFmtChar(*lpFormat)) || |
| (dwFlags & TIME_TIMEVARSONLY && IsTimeFmtChar(*lpFormat))) |
| { |
| char buffA[32]; |
| WCHAR buff[32], fmtChar; |
| LPCWSTR szAdd = NULL; |
| DWORD dwVal = 0; |
| int count = 0, dwLen; |
| |
| bSkipping = FALSE; |
| |
| fmtChar = *lpFormat; |
| while (*lpFormat == fmtChar) |
| { |
| count++; |
| lpFormat++; |
| } |
| buff[0] = '\0'; |
| |
| switch(fmtChar) |
| { |
| case 'd': |
| if (count >= 4) |
| szAdd = GetLongDay(node, (lpTime->wDayOfWeek + 6) % 7); |
| else if (count == 3) |
| szAdd = GetShortDay(node, (lpTime->wDayOfWeek + 6) % 7); |
| else |
| { |
| dwVal = lpTime->wDay; |
| szAdd = buff; |
| } |
| break; |
| |
| case 'M': |
| if (count >= 4) |
| szAdd = GetLongMonth(node, lpTime->wMonth - 1); |
| else if (count == 3) |
| szAdd = GetShortMonth(node, lpTime->wMonth - 1); |
| else |
| { |
| dwVal = lpTime->wMonth; |
| szAdd = buff; |
| } |
| break; |
| |
| case 'y': |
| if (count >= 4) |
| { |
| count = 4; |
| dwVal = lpTime->wYear; |
| } |
| else |
| { |
| count = count > 2 ? 2 : count; |
| dwVal = lpTime->wYear % 100; |
| } |
| szAdd = buff; |
| break; |
| |
| case 'g': |
| if (count == 2) |
| { |
| /* FIXME: Our GetCalendarInfo() does not yet support CAL_SERASTRING. |
| * When it is fixed, this string should be cached in 'node'. |
| */ |
| FIXME("Should be using GetCalendarInfo(CAL_SERASTRING), defaulting to 'AD'\n"); |
| buff[0] = 'A'; buff[1] = 'D'; buff[2] = '\0'; |
| } |
| else |
| { |
| buff[0] = 'g'; buff[1] = '\0'; /* Add a literal 'g' */ |
| } |
| szAdd = buff; |
| break; |
| |
| case 'h': |
| if (!(dwFlags & TIME_FORCE24HOURFORMAT)) |
| { |
| count = count > 2 ? 2 : count; |
| dwVal = lpTime->wHour == 0 ? 12 : (lpTime->wHour - 1) % 12 + 1; |
| szAdd = buff; |
| break; |
| } |
| /* .. fall through if we are forced to output in 24 hour format */ |
| |
| case 'H': |
| count = count > 2 ? 2 : count; |
| dwVal = lpTime->wHour; |
| szAdd = buff; |
| break; |
| |
| case 'm': |
| if (dwFlags & TIME_NOMINUTESORSECONDS) |
| { |
| cchWritten = lastFormatPos; /* Skip */ |
| bSkipping = TRUE; |
| } |
| else |
| { |
| count = count > 2 ? 2 : count; |
| dwVal = lpTime->wMinute; |
| szAdd = buff; |
| } |
| break; |
| |
| case 's': |
| if (dwFlags & (TIME_NOSECONDS|TIME_NOMINUTESORSECONDS)) |
| { |
| cchWritten = lastFormatPos; /* Skip */ |
| bSkipping = TRUE; |
| } |
| else |
| { |
| count = count > 2 ? 2 : count; |
| dwVal = lpTime->wSecond; |
| szAdd = buff; |
| } |
| break; |
| |
| case 't': |
| if (dwFlags & TIME_NOTIMEMARKER) |
| { |
| cchWritten = lastFormatPos; /* Skip */ |
| bSkipping = TRUE; |
| } |
| else |
| { |
| if (count == 1) |
| szAdd = lpTime->wHour < 12 ? node->szShortAM : node->szShortPM; |
| else |
| szAdd = lpTime->wHour < 12 ? GetAM(node) : GetPM(node); |
| } |
| break; |
| } |
| |
| if (szAdd == buff && buff[0] == '\0') |
| { |
| /* We have a numeric value to add */ |
| sprintf(buffA, "%.*ld", count, dwVal); |
| MultiByteToWideChar(CP_ACP, 0, buffA, -1, buff, sizeof(buff)/sizeof(WCHAR)); |
| } |
| |
| dwLen = szAdd ? strlenW(szAdd) : 0; |
| |
| if (cchOut && dwLen) |
| { |
| if (cchWritten + dwLen < cchOut) |
| memcpy(lpStr + cchWritten, szAdd, dwLen * sizeof(WCHAR)); |
| else |
| { |
| memcpy(lpStr + cchWritten, szAdd, (cchOut - cchWritten) * sizeof(WCHAR)); |
| goto NLS_GetDateTimeFormatW_Overrun; |
| } |
| } |
| cchWritten += dwLen; |
| lastFormatPos = cchWritten; /* Save position of last output format text */ |
| } |
| else |
| { |
| /* Literal character */ |
| if (!cchOut) |
| cchWritten++; /* Count size only */ |
| else if (cchWritten >= cchOut) |
| goto NLS_GetDateTimeFormatW_Overrun; |
| else if (!bSkipping || *lpFormat == ' ') |
| { |
| lpStr[cchWritten] = *lpFormat; |
| cchWritten++; |
| } |
| lpFormat++; |
| } |
| } |
| |
| /* Final string terminator and sanity check */ |
| if (cchOut) |
| { |
| if (cchWritten >= cchOut) |
| goto NLS_GetDateTimeFormatW_Overrun; |
| else |
| lpStr[cchWritten] = '\0'; |
| } |
| cchWritten++; /* Include terminating NUL */ |
| |
| TRACE("returning length=%d, ouput='%s'\n", cchWritten, debugstr_w(lpStr)); |
| return cchWritten; |
| |
| NLS_GetDateTimeFormatW_Overrun: |
| TRACE("returning 0, (ERROR_INSUFFICIENT_BUFFER)\n"); |
| SetLastError(ERROR_INSUFFICIENT_BUFFER); |
| return 0; |
| } |
| |
| /****************************************************************************** |
| * NLS_GetDateTimeFormatA <internal> |
| * |
| * ASCII wrapper for GetDateFormatA/GetTimeFormatA. |
| */ |
| static INT NLS_GetDateTimeFormatA(LCID lcid, DWORD dwFlags, |
| const SYSTEMTIME* lpTime, |
| LPCSTR lpFormat, LPSTR lpStr, INT cchOut) |
| { |
| DWORD cp = CP_ACP; |
| WCHAR szFormat[128], szOut[128]; |
| INT iRet; |
| |
| TRACE("(0x%04lx,0x%08lx,%p,%s,%p,%d)\n", lcid, dwFlags, lpTime, |
| debugstr_a(lpFormat), lpStr, cchOut); |
| |
| if (NLS_IsUnicodeOnlyLcid(lcid)) |
| { |
| GetDateTimeFormatA_InvalidParameter: |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return 0; |
| } |
| |
| if (!(dwFlags & LOCALE_USE_CP_ACP)) |
| { |
| const NLS_FORMAT_NODE *node = NLS_GetFormats(lcid, dwFlags); |
| if (!node) |
| goto GetDateTimeFormatA_InvalidParameter; |
| cp = node->dwCodePage; |
| } |
| |
| if (lpFormat) |
| MultiByteToWideChar(cp, 0, lpFormat, -1, szFormat, sizeof(szFormat)/sizeof(WCHAR)); |
| |
| if (cchOut > (int)(sizeof(szOut)/sizeof(WCHAR))) |
| cchOut = sizeof(szOut)/sizeof(WCHAR); |
| |
| szOut[0] = '\0'; |
| |
| iRet = NLS_GetDateTimeFormatW(lcid, dwFlags, lpTime, lpFormat ? szFormat : NULL, |
| lpStr ? szOut : NULL, cchOut); |
| |
| if (lpStr) |
| { |
| if (szOut[0]) |
| WideCharToMultiByte(cp, 0, szOut, iRet ? -1 : cchOut, lpStr, cchOut, 0, 0); |
| else if (cchOut && iRet) |
| *lpStr = '\0'; |
| } |
| return iRet; |
| } |
| |
| /****************************************************************************** |
| * GetDateFormatA [KERNEL32.@] |
| * |
| * Format a date for a given locale. |
| * |
| * PARAMS |
| * lcid [I] Locale to format for |
| * dwFlags [I] LOCALE_ and DATE_ flags from "winnls.h" |
| * lpTime [I] Date to format |
| * lpFormat [I] Format string, or NULL to use the system defaults |
| * lpDateStr [O] Destination for formatted string |
| * cchOut [I] Size of lpDateStr, or 0 to calculate the resulting size |
| * |
| * NOTES |
| * - If lpFormat is NULL, lpDateStr will be formatted according to the format |
| * details returned by GetLocaleInfoA() and modified by dwFlags. |
| * - lpFormat is a string of characters and formatting tokens. Any characters |
| * in the string are copied verbatim to lpDateStr, with tokens being replaced |
| * by the date values they represent. |
| * - The following tokens have special meanings in a date format string: |
| *| Token Meaning |
| *| ----- ------- |
| *| d Single digit day of the month (no leading 0) |
| *| dd Double digit day of the month |
| *| ddd Short name for the day of the week |
| *| dddd Long name for the day of the week |
| *| M Single digit month of the year (no leading 0) |
| *| MM Double digit month of the year |
| *| MMM Short name for the month of the year |
| *| MMMM Long name for the month of the year |
| *| y Double digit year number (no leading 0) |
| *| yy Double digit year number |
| *| yyyy Four digit year number |
| *| gg Era string, for example 'AD'. |
| * - To output any literal character that could be misidentified as a token, |
| * enclose it in single quotes. |
| * - The Ascii version of this function fails if lcid is Unicode only. |
| * |
| * RETURNS |
| * Success: The number of character written to lpDateStr, or that would |
| * have been written, if cchOut is 0. |
| * Failure: 0. Use GetLastError() to determine the cause. |
| */ |
| INT WINAPI GetDateFormatA( LCID lcid, DWORD dwFlags, const SYSTEMTIME* lpTime, |
| LPCSTR lpFormat, LPSTR lpDateStr, INT cchOut) |
| { |
| TRACE("(0x%04lx,0x%08lx,%p,%s,%p,%d)\n",lcid, dwFlags, lpTime, |
| debugstr_a(lpFormat), lpDateStr, cchOut); |
| |
| return NLS_GetDateTimeFormatA(lcid, dwFlags | DATE_DATEVARSONLY, lpTime, |
| lpFormat, lpDateStr, cchOut); |
| } |
| |
| |
| /****************************************************************************** |
| * GetDateFormatW [KERNEL32.@] |
| * |
| * See GetDateFormatA. |
| */ |
| INT WINAPI GetDateFormatW(LCID lcid, DWORD dwFlags, const SYSTEMTIME* lpTime, |
| LPCWSTR lpFormat, LPWSTR lpDateStr, INT cchOut) |
| { |
| TRACE("(0x%04lx,0x%08lx,%p,%s,%p,%d)\n", lcid, dwFlags, lpTime, |
| debugstr_w(lpFormat), lpDateStr, cchOut); |
| |
| return NLS_GetDateTimeFormatW(lcid, dwFlags|DATE_DATEVARSONLY, lpTime, |
| lpFormat, lpDateStr, cchOut); |
| } |
| |
| /****************************************************************************** |
| * GetTimeFormatA [KERNEL32.@] |
| * |
| * Format a time for a given locale. |
| * |
| * PARAMS |
| * lcid [I] Locale to format for |
| * dwFlags [I] LOCALE_ and TIME_ flags from "winnls.h" |
| * lpTime [I] Time to format |
| * lpFormat [I] Formatting overrides |
| * lpTimeStr [O] Destination for formatted string |
| * cchOut [I] Size of lpTimeStr, or 0 to calculate the resulting size |
| * |
| * NOTES |
| * - If lpFormat is NULL, lpszValue will be formatted according to the format |
| * details returned by GetLocaleInfoA() and modified by dwFlags. |
| * - lpFormat is a string of characters and formatting tokens. Any characters |
| * in the string are copied verbatim to lpTimeStr, with tokens being replaced |
| * by the time values they represent. |
| * - The following tokens have special meanings in a time format string: |
| *| Token Meaning |
| *| ----- ------- |
| *| h Hours with no leading zero (12-hour clock) |
| *| hh Hours with full two digits (12-hour clock) |
| *| H Hours with no leading zero (24-hour clock) |
| *| HH Hours with full two digits (24-hour clock) |
| *| m Minutes with no leading zero |
| *| mm Minutes with full two digits |
| *| s Seconds with no leading zero |
| *| ss Seconds with full two digits |
| *| t Short time marker (e.g. "A" or "P") |
| *| tt Long time marker (e.g. "AM", "PM") |
| * - To output any literal character that could be misidentified as a token, |
| * enclose it in single quotes. |
| * - The Ascii version of this function fails if lcid is Unicode only. |
| * |
| * RETURNS |
| * Success: The number of character written to lpTimeStr, or that would |
| * have been written, if cchOut is 0. |
| * Failure: 0. Use GetLastError() to determine the cause. |
| */ |
| INT WINAPI GetTimeFormatA(LCID lcid, DWORD dwFlags, const SYSTEMTIME* lpTime, |
| LPCSTR lpFormat, LPSTR lpTimeStr, INT cchOut) |
| { |
| TRACE("(0x%04lx,0x%08lx,%p,%s,%p,%d)\n",lcid, dwFlags, lpTime, |
| debugstr_a(lpFormat), lpTimeStr, cchOut); |
| |
| return NLS_GetDateTimeFormatA(lcid, dwFlags|TIME_TIMEVARSONLY, lpTime, |
| lpFormat, lpTimeStr, cchOut); |
| } |
| |
| /****************************************************************************** |
| * GetTimeFormatW [KERNEL32.@] |
| * |
| * See GetTimeFormatA. |
| */ |
| INT WINAPI GetTimeFormatW(LCID lcid, DWORD dwFlags, const SYSTEMTIME* lpTime, |
| LPCWSTR lpFormat, LPWSTR lpTimeStr, INT cchOut) |
| { |
| TRACE("(0x%04lx,0x%08lx,%p,%s,%p,%d)\n",lcid, dwFlags, lpTime, |
| debugstr_w(lpFormat), lpTimeStr, cchOut); |
| |
| return NLS_GetDateTimeFormatW(lcid, dwFlags|TIME_TIMEVARSONLY, lpTime, |
| lpFormat, lpTimeStr, cchOut); |
| } |
| |
| /************************************************************************** |
| * GetNumberFormatA (KERNEL32.@) |
| * |
| * Format a number string for a given locale. |
| * |
| * PARAMS |
| * lcid [I] Locale to format for |
| * dwFlags [I] LOCALE_ flags from "winnls.h" |
| * lpszValue [I] String to format |
| * lpFormat [I] Formatting overrides |
| * lpNumberStr [O] Destination for formatted string |
| * cchOut [I] Size of lpNumberStr, or 0 to calculate the resulting size |
| * |
| * NOTES |
| * - lpszValue can contain only '0' - '9', '-' and '.'. |
| * - If lpFormat is non-NULL, dwFlags must be 0. In this case lpszValue will |
| * be formatted according to the format details returned by GetLocaleInfoA(). |
| * - This function rounds the number string if the number of decimals exceeds the |
| * locales normal number of decimal places. |
| * - If cchOut is 0, this function does not write to lpNumberStr. |
| * - The Ascii version of this function fails if lcid is Unicode only. |
| * |
| * RETURNS |
| * Success: The number of character written to lpNumberStr, or that would |
| * have been written, if cchOut is 0. |
| * Failure: 0. Use GetLastError() to determine the cause. |
| */ |
| INT WINAPI GetNumberFormatA(LCID lcid, DWORD dwFlags, |
| LPCSTR lpszValue, const NUMBERFMTA *lpFormat, |
| LPSTR lpNumberStr, int cchOut) |
| { |
| DWORD cp = CP_ACP; |
| WCHAR szDec[8], szGrp[8], szIn[128], szOut[128]; |
| NUMBERFMTW fmt; |
| const NUMBERFMTW *pfmt = NULL; |
| INT iRet; |
| |
| TRACE("(0x%04lx,0x%08lx,%s,%p,%p,%d)\n", lcid, dwFlags, debugstr_a(lpszValue), |
| lpFormat, lpNumberStr, cchOut); |
| |
| if (NLS_IsUnicodeOnlyLcid(lcid)) |
| { |
| GetNumberFormatA_InvalidParameter: |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return 0; |
| } |
| |
| if (!(dwFlags & LOCALE_USE_CP_ACP)) |
| { |
| const NLS_FORMAT_NODE *node = NLS_GetFormats(lcid, dwFlags); |
| if (!node) |
| goto GetNumberFormatA_InvalidParameter; |
| cp = node->dwCodePage; |
| } |
| |
| if (lpFormat) |
| { |
| memcpy(&fmt, lpFormat, sizeof(fmt)); |
| pfmt = &fmt; |
| if (lpFormat->lpDecimalSep) |
| { |
| MultiByteToWideChar(cp, 0, lpFormat->lpDecimalSep, -1, szDec, sizeof(szDec)/sizeof(WCHAR)); |
| fmt.lpDecimalSep = szDec; |
| } |
| if (lpFormat->lpThousandSep) |
| { |
| MultiByteToWideChar(cp, 0, lpFormat->lpThousandSep, -1, szGrp, sizeof(szGrp)/sizeof(WCHAR)); |
| fmt.lpThousandSep = szGrp; |
| } |
| } |
| |
| if (lpszValue) |
| MultiByteToWideChar(cp, 0, lpszValue, -1, szIn, sizeof(szIn)/sizeof(WCHAR)); |
| |
| if (cchOut > (int)(sizeof(szOut)/sizeof(WCHAR))) |
| cchOut = sizeof(szOut)/sizeof(WCHAR); |
| |
| szOut[0] = '\0'; |
| |
| iRet = GetNumberFormatW(lcid, dwFlags, lpszValue ? szIn : NULL, pfmt, |
| lpNumberStr ? szOut : NULL, cchOut); |
| |
| if (szOut[0] && lpNumberStr) |
| WideCharToMultiByte(cp, 0, szOut, -1, lpNumberStr, cchOut, 0, 0); |
| return iRet; |
| } |
| |
| /* Number parsing state flags */ |
| #define NF_ISNEGATIVE 0x1 /* '-' found */ |
| #define NF_ISREAL 0x2 /* '.' found */ |
| #define NF_DIGITS 0x4 /* '0'-'9' found */ |
| #define NF_DIGITS_OUT 0x8 /* Digits before the '.' found */ |
| #define NF_ROUND 0x10 /* Number needs to be rounded */ |
| |
| /* Formatting options for Numbers */ |
| #define NLS_NEG_PARENS 0 /* "(1.1)" */ |
| #define NLS_NEG_LEFT 1 /* "-1.1" */ |
| #define NLS_NEG_LEFT_SPACE 2 /* "- 1.1" */ |
| #define NLS_NEG_RIGHT 3 /* "1.1-" */ |
| #define NLS_NEG_RIGHT_SPACE 4 /* "1.1 -" */ |
| |
| /************************************************************************** |
| * GetNumberFormatW (KERNEL32.@) |
| * |
| * See GetNumberFormatA. |
| */ |
| INT WINAPI GetNumberFormatW(LCID lcid, DWORD dwFlags, |
| LPCWSTR lpszValue, const NUMBERFMTW *lpFormat, |
| LPWSTR lpNumberStr, int cchOut) |
| { |
| WCHAR szBuff[128], *szOut = szBuff + sizeof(szBuff) / sizeof(WCHAR) - 1; |
| WCHAR szNegBuff[8]; |
| const WCHAR *lpszNeg = NULL, *lpszNegStart, *szSrc; |
| DWORD dwState = 0, dwDecimals = 0, dwGroupCount = 0, dwCurrentGroupCount = 0; |
| INT iRet; |
| |
| TRACE("(0x%04lx,0x%08lx,%s,%p,%p,%d)\n", lcid, dwFlags, debugstr_w(lpszValue), |
| lpFormat, lpNumberStr, cchOut); |
| |
| if (!lpszValue || cchOut < 0 || (cchOut > 0 && !lpNumberStr) || |
| !IsValidLocale(lcid, 0) || |
| (lpFormat && (dwFlags || !lpFormat->lpDecimalSep || !lpFormat->lpThousandSep))) |
| { |
| GetNumberFormatW_Error: |
| SetLastError(lpFormat && dwFlags ? ERROR_INVALID_FLAGS : ERROR_INVALID_PARAMETER); |
| return 0; |
| } |
| |
| if (!lpFormat) |
| { |
| const NLS_FORMAT_NODE *node = NLS_GetFormats(lcid, dwFlags); |
| |
| if (!node) |
| goto GetNumberFormatW_Error; |
| lpFormat = &node->fmt; |
| lpszNegStart = lpszNeg = GetNegative(node); |
| } |
| else |
| { |
| GetLocaleInfoW(lcid, LOCALE_SNEGATIVESIGN|(dwFlags & LOCALE_NOUSEROVERRIDE), |
| szNegBuff, sizeof(szNegBuff)/sizeof(WCHAR)); |
| lpszNegStart = lpszNeg = szNegBuff; |
| } |
| lpszNeg = lpszNeg + strlenW(lpszNeg) - 1; |
| |
| dwFlags &= (LOCALE_NOUSEROVERRIDE|LOCALE_USE_CP_ACP); |
| |
| /* Format the number backwards into a temporary buffer */ |
| |
| szSrc = lpszValue; |
| *szOut-- = '\0'; |
| |
| /* Check the number for validity */ |
| while (*szSrc) |
| { |
| if (*szSrc >= '0' && *szSrc <= '9') |
| { |
| dwState |= NF_DIGITS; |
| if (dwState & NF_ISREAL) |
| dwDecimals++; |
| } |
| else if (*szSrc == '-') |
| { |
| if (dwState) |
| goto GetNumberFormatW_Error; /* '-' not first character */ |
| dwState |= NF_ISNEGATIVE; |
| } |
| else if (*szSrc == '.') |
| { |
| if (dwState & NF_ISREAL) |
| goto GetNumberFormatW_Error; /* More than one '.' */ |
| dwState |= NF_ISREAL; |
| } |
| else |
| goto GetNumberFormatW_Error; /* Invalid char */ |
| szSrc++; |
| } |
| szSrc--; /* Point to last character */ |
| |
| if (!(dwState & NF_DIGITS)) |
| goto GetNumberFormatW_Error; /* No digits */ |
| |
| /* Add any trailing negative sign */ |
| if (dwState & NF_ISNEGATIVE) |
| { |
| switch (lpFormat->NegativeOrder) |
| { |
| case NLS_NEG_PARENS: |
| *szOut-- = ')'; |
| break; |
| case NLS_NEG_RIGHT: |
| case NLS_NEG_RIGHT_SPACE: |
| while (lpszNeg >= lpszNegStart) |
| *szOut-- = *lpszNeg--; |
| if (lpFormat->NegativeOrder == NLS_NEG_RIGHT_SPACE) |
| *szOut-- = ' '; |
| break; |
| } |
| } |
| |
| /* Copy all digits up to the decimal point */ |
| if (!lpFormat->NumDigits) |
| { |
| if (dwState & NF_ISREAL) |
| { |
| while (*szSrc != '.') /* Don't write any decimals or a separator */ |
| { |
| if (*szSrc >= '5' || (*szSrc == '4' && (dwState & NF_ROUND))) |
| dwState |= NF_ROUND; |
| else |
| dwState &= ~NF_ROUND; |
| szSrc--; |
| } |
| szSrc--; |
| } |
| } |
| else |
| { |
| LPWSTR lpszDec = lpFormat->lpDecimalSep + strlenW(lpFormat->lpDecimalSep) - 1; |
| |
| if (dwDecimals <= lpFormat->NumDigits) |
| { |
| dwDecimals = lpFormat->NumDigits - dwDecimals; |
| while (dwDecimals--) |
| *szOut-- = '0'; /* Pad to correct number of dp */ |
| } |
| else |
| { |
| dwDecimals -= lpFormat->NumDigits; |
| /* Skip excess decimals, and determine if we have to round the number */ |
| while (dwDecimals--) |
| { |
| if (*szSrc >= '5' || (*szSrc == '4' && (dwState & NF_ROUND))) |
| dwState |= NF_ROUND; |
| else |
| dwState &= ~NF_ROUND; |
| szSrc--; |
| } |
| } |
| |
| if (dwState & NF_ISREAL) |
| { |
| while (*szSrc != '.') |
| { |
| if (dwState & NF_ROUND) |
| { |
| if (*szSrc == '9') |
| *szOut-- = '0'; /* continue rounding */ |
| else |
| { |
| dwState &= ~NF_ROUND; |
| *szOut-- = (*szSrc)+1; |
| } |
| szSrc--; |
| } |
| else |
| *szOut-- = *szSrc--; /* Write existing decimals */ |
| } |
| szSrc--; /* Skip '.' */ |
| } |
| |
| while (lpszDec >= lpFormat->lpDecimalSep) |
| *szOut-- = *lpszDec--; /* Write decimal separator */ |
| } |
| |
| dwGroupCount = lpFormat->Grouping == 32 ? 3 : lpFormat->Grouping; |
| |
| /* Write the remaining whole number digits, including grouping chars */ |
| while (szSrc >= lpszValue && *szSrc >= '0' && *szSrc <= '9') |
| { |
| if (dwState & NF_ROUND) |
| { |
| if (*szSrc == '9') |
| *szOut-- = '0'; /* continue rounding */ |
| else |
| { |
| dwState &= ~NF_ROUND; |
| *szOut-- = (*szSrc)+1; |
| } |
| szSrc--; |
| } |
| else |
| *szOut-- = *szSrc--; |
| |
| dwState |= NF_DIGITS_OUT; |
| dwCurrentGroupCount++; |
| if (szSrc >= lpszValue && dwCurrentGroupCount == dwGroupCount && *szSrc != '-') |
| { |
| LPWSTR lpszGrp = lpFormat->lpThousandSep + strlenW(lpFormat->lpThousandSep) - 1; |
| |
| while (lpszGrp >= lpFormat->lpThousandSep) |
| *szOut-- = *lpszGrp--; /* Write grouping char */ |
| |
| dwCurrentGroupCount = 0; |
| if (lpFormat->Grouping == 32) |
| dwGroupCount = 2; /* Indic grouping: 3 then 2 */ |
| } |
| } |
| if (dwState & NF_ROUND) |
| { |
| *szOut-- = '1'; /* e.g. .6 > 1.0 */ |
| } |
| else if (!(dwState & NF_DIGITS_OUT) && lpFormat->LeadingZero) |
| *szOut-- = '0'; /* Add leading 0 if we have no digits before the decimal point */ |
| |
| /* Add any leading negative sign */ |
| if (dwState & NF_ISNEGATIVE) |
| { |
| switch (lpFormat->NegativeOrder) |
| { |
| case NLS_NEG_PARENS: |
| *szOut-- = '('; |
| break; |
| case NLS_NEG_LEFT_SPACE: |
| *szOut-- = ' '; |
| /* Fall through */ |
| case NLS_NEG_LEFT: |
| while (lpszNeg >= lpszNegStart) |
| *szOut-- = *lpszNeg--; |
| break; |
| } |
| } |
| szOut++; |
| |
| iRet = strlenW(szOut) + 1; |
| if (cchOut) |
| { |
| if (iRet <= cchOut) |
| memcpy(lpNumberStr, szOut, iRet * sizeof(WCHAR)); |
| else |
| { |
| memcpy(lpNumberStr, szOut, cchOut * sizeof(WCHAR)); |
| lpNumberStr[cchOut - 1] = '\0'; |
| SetLastError(ERROR_INSUFFICIENT_BUFFER); |
| iRet = 0; |
| } |
| } |
| return iRet; |
| } |
| |
| /************************************************************************** |
| * GetCurrencyFormatA (KERNEL32.@) |
| * |
| * Format a currency string for a given locale. |
| * |
| * PARAMS |
| * lcid [I] Locale to format for |
| * dwFlags [I] LOCALE_ flags from "winnls.h" |
| * lpszValue [I] String to format |
| * lpFormat [I] Formatting overrides |
| * lpCurrencyStr [O] Destination for formatted string |
| * cchOut [I] Size of lpCurrencyStr, or 0 to calculate the resulting size |
| * |
| * NOTES |
| * - lpszValue can contain only '0' - '9', '-' and '.'. |
| * - If lpFormat is non-NULL, dwFlags must be 0. In this case lpszValue will |
| * be formatted according to the format details returned by GetLocaleInfoA(). |
| * - This function rounds the currency if the number of decimals exceeds the |
| * locales number of currency decimal places. |
| * - If cchOut is 0, this function does not write to lpCurrencyStr. |
| * - The Ascii version of this function fails if lcid is Unicode only. |
| * |
| * RETURNS |
| * Success: The number of character written to lpNumberStr, or that would |
| * have been written, if cchOut is 0. |
| * Failure: 0. Use GetLastError() to determine the cause. |
| */ |
| INT WINAPI GetCurrencyFormatA(LCID lcid, DWORD dwFlags, |
| LPCSTR lpszValue, const CURRENCYFMTA *lpFormat, |
| LPSTR lpCurrencyStr, int cchOut) |
| { |
| DWORD cp = CP_ACP; |
| WCHAR szDec[8], szGrp[8], szCy[8], szIn[128], szOut[128]; |
| CURRENCYFMTW fmt; |
| const CURRENCYFMTW *pfmt = NULL; |
| INT iRet; |
| |
| TRACE("(0x%04lx,0x%08lx,%s,%p,%p,%d)\n", lcid, dwFlags, debugstr_a(lpszValue), |
| lpFormat, lpCurrencyStr, cchOut); |
| |
| if (NLS_IsUnicodeOnlyLcid(lcid)) |
| { |
| GetCurrencyFormatA_InvalidParameter: |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return 0; |
| } |
| |
| if (!(dwFlags & LOCALE_USE_CP_ACP)) |
| { |
| const NLS_FORMAT_NODE *node = NLS_GetFormats(lcid, dwFlags); |
| if (!node) |
| goto GetCurrencyFormatA_InvalidParameter; |
| cp = node->dwCodePage; |
| } |
| |
| if (lpFormat) |
| { |
| memcpy(&fmt, lpFormat, sizeof(fmt)); |
| pfmt = &fmt; |
| if (lpFormat->lpDecimalSep) |
| { |
| MultiByteToWideChar(cp, 0, lpFormat->lpDecimalSep, -1, szDec, sizeof(szDec)/sizeof(WCHAR)); |
| fmt.lpDecimalSep = szDec; |
| } |
| if (lpFormat->lpThousandSep) |
| { |
| MultiByteToWideChar(cp, 0, lpFormat->lpThousandSep, -1, szGrp, sizeof(szGrp)/sizeof(WCHAR)); |
| fmt.lpThousandSep = szGrp; |
| } |
| if (lpFormat->lpCurrencySymbol) |
| { |
| MultiByteToWideChar(cp, 0, lpFormat->lpCurrencySymbol, -1, szCy, sizeof(szCy)/sizeof(WCHAR)); |
| fmt.lpCurrencySymbol = szCy; |
| } |
| } |
| |
| if (lpszValue) |
| MultiByteToWideChar(cp, 0, lpszValue, -1, szIn, sizeof(szIn)/sizeof(WCHAR)); |
| |
| if (cchOut > (int)(sizeof(szOut)/sizeof(WCHAR))) |
| cchOut = sizeof(szOut)/sizeof(WCHAR); |
| |
| szOut[0] = '\0'; |
| |
| iRet = GetCurrencyFormatW(lcid, dwFlags, lpszValue ? szIn : NULL, pfmt, |
| lpCurrencyStr ? szOut : NULL, cchOut); |
| |
| if (szOut[0] && lpCurrencyStr) |
| WideCharToMultiByte(cp, 0, szOut, -1, lpCurrencyStr, cchOut, 0, 0); |
| return iRet; |
| } |
| |
| /* Formatting states for Currencies. We use flags to avoid code duplication. */ |
| #define CF_PARENS 0x1 /* Parentheses */ |
| #define CF_MINUS_LEFT 0x2 /* '-' to the left */ |
| #define CF_MINUS_RIGHT 0x4 /* '-' to the right */ |
| #define CF_MINUS_BEFORE 0x8 /* '-' before '$' */ |
| #define CF_CY_LEFT 0x10 /* '$' to the left */ |
| #define CF_CY_RIGHT 0x20 /* '$' to the right */ |
| #define CF_CY_SPACE 0x40 /* ' ' by '$' */ |
| |
| /************************************************************************** |
| * GetCurrencyFormatW (KERNEL32.@) |
| * |
| * See GetCurrencyFormatA. |
| */ |
| INT WINAPI GetCurrencyFormatW(LCID lcid, DWORD dwFlags, |
| LPCWSTR lpszValue, const CURRENCYFMTW *lpFormat, |
| LPWSTR lpCurrencyStr, int cchOut) |
| { |
| static const BYTE NLS_NegCyFormats[16] = |
| { |
| CF_PARENS|CF_CY_LEFT, /* ($1.1) */ |
| CF_MINUS_LEFT|CF_MINUS_BEFORE|CF_CY_LEFT, /* -$1.1 */ |
| CF_MINUS_LEFT|CF_CY_LEFT, /* $-1.1 */ |
| CF_MINUS_RIGHT|CF_CY_LEFT, /* $1.1- */ |
| CF_PARENS|CF_CY_RIGHT, /* (1.1$) */ |
| CF_MINUS_LEFT|CF_CY_RIGHT, /* -1.1$ */ |
| CF_MINUS_RIGHT|CF_MINUS_BEFORE|CF_CY_RIGHT, /* 1.1-$ */ |
| CF_MINUS_RIGHT|CF_CY_RIGHT, /* 1.1$- */ |
| CF_MINUS_LEFT|CF_CY_RIGHT|CF_CY_SPACE, /* -1.1 $ */ |
| CF_MINUS_LEFT|CF_MINUS_BEFORE|CF_CY_LEFT|CF_CY_SPACE, /* -$ 1.1 */ |
| CF_MINUS_RIGHT|CF_CY_RIGHT|CF_CY_SPACE, /* 1.1 $- */ |
| CF_MINUS_RIGHT|CF_CY_LEFT|CF_CY_SPACE, /* $ 1.1- */ |
| CF_MINUS_LEFT|CF_CY_LEFT|CF_CY_SPACE, /* $ -1.1 */ |
| CF_MINUS_RIGHT|CF_MINUS_BEFORE|CF_CY_RIGHT|CF_CY_SPACE, /* 1.1- $ */ |
| CF_PARENS|CF_CY_LEFT|CF_CY_SPACE, /* ($ 1.1) */ |
| CF_PARENS|CF_CY_RIGHT|CF_CY_SPACE, /* (1.1 $) */ |
| }; |
| static const BYTE NLS_PosCyFormats[4] = |
| { |
| CF_CY_LEFT, /* $1.1 */ |
| CF_CY_RIGHT, /* 1.1$ */ |
| CF_CY_LEFT|CF_CY_SPACE, /* $ 1.1 */ |
| CF_CY_RIGHT|CF_CY_SPACE, /* 1.1 $ */ |
| }; |
| WCHAR szBuff[128], *szOut = szBuff + sizeof(szBuff) / sizeof(WCHAR) - 1; |
| WCHAR szNegBuff[8]; |
| const WCHAR *lpszNeg = NULL, *lpszNegStart, *szSrc, *lpszCy, *lpszCyStart; |
| DWORD dwState = 0, dwDecimals = 0, dwGroupCount = 0, dwCurrentGroupCount = 0, dwFmt; |
| INT iRet; |
| |
| TRACE("(0x%04lx,0x%08lx,%s,%p,%p,%d)\n", lcid, dwFlags, debugstr_w(lpszValue), |
| lpFormat, lpCurrencyStr, cchOut); |
| |
| if (!lpszValue || cchOut < 0 || (cchOut > 0 && !lpCurrencyStr) || |
| !IsValidLocale(lcid, 0) || |
| (lpFormat && (dwFlags || !lpFormat->lpDecimalSep || !lpFormat->lpThousandSep || |
| !lpFormat->lpCurrencySymbol || lpFormat->NegativeOrder > 15 || |
| lpFormat->PositiveOrder > 3))) |
| { |
| GetCurrencyFormatW_Error: |
| SetLastError(lpFormat && dwFlags ? ERROR_INVALID_FLAGS : ERROR_INVALID_PARAMETER); |
| return 0; |
| } |
| |
| if (!lpFormat) |
| { |
| const NLS_FORMAT_NODE *node = NLS_GetFormats(lcid, dwFlags); |
| |
| if (!node) |
| goto GetCurrencyFormatW_Error; |
| lpFormat = &node->cyfmt; |
| lpszNegStart = lpszNeg = GetNegative(node); |
| } |
| else |
| { |
| GetLocaleInfoW(lcid, LOCALE_SNEGATIVESIGN|(dwFlags & LOCALE_NOUSEROVERRIDE), |
| szNegBuff, sizeof(szNegBuff)/sizeof(WCHAR)); |
| lpszNegStart = lpszNeg = szNegBuff; |
| } |
| dwFlags &= (LOCALE_NOUSEROVERRIDE|LOCALE_USE_CP_ACP); |
| |
| lpszNeg = lpszNeg + strlenW(lpszNeg) - 1; |
| lpszCyStart = lpFormat->lpCurrencySymbol; |
| lpszCy = lpszCyStart + strlenW(lpszCyStart) - 1; |
| |
| /* Format the currency backwards into a temporary buffer */ |
| |
| szSrc = lpszValue; |
| *szOut-- = '\0'; |
| |
| /* Check the number for validity */ |
| while (*szSrc) |
| { |
| if (*szSrc >= '0' && *szSrc <= '9') |
| { |
| dwState |= NF_DIGITS; |
| if (dwState & NF_ISREAL) |
| dwDecimals++; |
| } |
| else if (*szSrc == '-') |
| { |
| if (dwState) |
| goto GetCurrencyFormatW_Error; /* '-' not first character */ |
| dwState |= NF_ISNEGATIVE; |
| } |
| else if (*szSrc == '.') |
| { |
| if (dwState & NF_ISREAL) |
| goto GetCurrencyFormatW_Error; /* More than one '.' */ |
| dwState |= NF_ISREAL; |
| } |
| else |
| goto GetCurrencyFormatW_Error; /* Invalid char */ |
| szSrc++; |
| } |
| szSrc--; /* Point to last character */ |
| |
| if (!(dwState & NF_DIGITS)) |
| goto GetCurrencyFormatW_Error; /* No digits */ |
| |
| if (dwState & NF_ISNEGATIVE) |
| dwFmt = NLS_NegCyFormats[lpFormat->NegativeOrder]; |
| else |
| dwFmt = NLS_PosCyFormats[lpFormat->PositiveOrder]; |
| |
| /* Add any trailing negative or currency signs */ |
| if (dwFmt & CF_PARENS) |
| *szOut-- = ')'; |
| |
| while (dwFmt & (CF_MINUS_RIGHT|CF_CY_RIGHT)) |
| { |
| switch (dwFmt & (CF_MINUS_RIGHT|CF_MINUS_BEFORE|CF_CY_RIGHT)) |
| { |
| case CF_MINUS_RIGHT: |
| case CF_MINUS_RIGHT|CF_CY_RIGHT: |
| while (lpszNeg >= lpszNegStart) |
| *szOut-- = *lpszNeg--; |
| dwFmt &= ~CF_MINUS_RIGHT; |
| break; |
| |
| case CF_CY_RIGHT: |
| case CF_MINUS_BEFORE|CF_CY_RIGHT: |
| case CF_MINUS_RIGHT|CF_MINUS_BEFORE|CF_CY_RIGHT: |
| while (lpszCy >= lpszCyStart) |
| *szOut-- = *lpszCy--; |
| if (dwFmt & CF_CY_SPACE) |
| *szOut-- = ' '; |
| dwFmt &= ~(CF_CY_RIGHT|CF_MINUS_BEFORE); |
| break; |
| } |
| } |
| |
| /* Copy all digits up to the decimal point */ |
| if (!lpFormat->NumDigits) |
| { |
| if (dwState & NF_ISREAL) |
| { |
| while (*szSrc != '.') /* Don't write any decimals or a separator */ |
| { |
| if (*szSrc >= '5' || (*szSrc == '4' && (dwState & NF_ROUND))) |
| dwState |= NF_ROUND; |
| else |
| dwState &= ~NF_ROUND; |
| szSrc--; |
| } |
| szSrc--; |
| } |
| } |
| else |
| { |
| LPWSTR lpszDec = lpFormat->lpDecimalSep + strlenW(lpFormat->lpDecimalSep) - 1; |
| |
| if (dwDecimals <= lpFormat->NumDigits) |
| { |
| dwDecimals = lpFormat->NumDigits - dwDecimals; |
| while (dwDecimals--) |
| *szOut-- = '0'; /* Pad to correct number of dp */ |
| } |
| else |
| { |
| dwDecimals -= lpFormat->NumDigits; |
| /* Skip excess decimals, and determine if we have to round the number */ |
| while (dwDecimals--) |
| { |
| if (*szSrc >= '5' || (*szSrc == '4' && (dwState & NF_ROUND))) |
| dwState |= NF_ROUND; |
| else |
| dwState &= ~NF_ROUND; |
| szSrc--; |
| } |
| } |
| |
| if (dwState & NF_ISREAL) |
| { |
| while (*szSrc != '.') |
| { |
| if (dwState & NF_ROUND) |
| { |
| if (*szSrc == '9') |
| *szOut-- = '0'; /* continue rounding */ |
| else |
| { |
| dwState &= ~NF_ROUND; |
| *szOut-- = (*szSrc)+1; |
| } |
| szSrc--; |
| } |
| else |
| *szOut-- = *szSrc--; /* Write existing decimals */ |
| } |
| szSrc--; /* Skip '.' */ |
| } |
| while (lpszDec >= lpFormat->lpDecimalSep) |
| *szOut-- = *lpszDec--; /* Write decimal separator */ |
| } |
| |
| dwGroupCount = lpFormat->Grouping; |
| |
| /* Write the remaining whole number digits, including grouping chars */ |
| while (szSrc >= lpszValue && *szSrc >= '0' && *szSrc <= '9') |
| { |
| if (dwState & NF_ROUND) |
| { |
| if (*szSrc == '9') |
| *szOut-- = '0'; /* continue rounding */ |
| else |
| { |
| dwState &= ~NF_ROUND; |
| *szOut-- = (*szSrc)+1; |
| } |
| szSrc--; |
| } |
| else |
| *szOut-- = *szSrc--; |
| |
| dwState |= NF_DIGITS_OUT; |
| dwCurrentGroupCount++; |
| if (szSrc >= lpszValue && dwCurrentGroupCount == dwGroupCount) |
| { |
| LPWSTR lpszGrp = lpFormat->lpThousandSep + strlenW(lpFormat->lpThousandSep) - 1; |
| |
| while (lpszGrp >= lpFormat->lpThousandSep) |
| *szOut-- = *lpszGrp--; /* Write grouping char */ |
| |
| dwCurrentGroupCount = 0; |
| } |
| } |
| if (dwState & NF_ROUND) |
| *szOut-- = '1'; /* e.g. .6 > 1.0 */ |
| else if (!(dwState & NF_DIGITS_OUT) && lpFormat->LeadingZero) |
| *szOut-- = '0'; /* Add leading 0 if we have no digits before the decimal point */ |
| |
| /* Add any leading negative or currency sign */ |
| while (dwFmt & (CF_MINUS_LEFT|CF_CY_LEFT)) |
| { |
| switch (dwFmt & (CF_MINUS_LEFT|CF_MINUS_BEFORE|CF_CY_LEFT)) |
| { |
| case CF_MINUS_LEFT: |
| case CF_MINUS_LEFT|CF_CY_LEFT: |
| while (lpszNeg >= lpszNegStart) |
| *szOut-- = *lpszNeg--; |
| dwFmt &= ~CF_MINUS_LEFT; |
| break; |
| |
| case CF_CY_LEFT: |
| case CF_CY_LEFT|CF_MINUS_BEFORE: |
| case CF_MINUS_LEFT|CF_MINUS_BEFORE|CF_CY_LEFT: |
| if (dwFmt & CF_CY_SPACE) |
| *szOut-- = ' '; |
| while (lpszCy >= lpszCyStart) |
| *szOut-- = *lpszCy--; |
| dwFmt &= ~(CF_CY_LEFT|CF_MINUS_BEFORE); |
| break; |
| } |
| } |
| if (dwFmt & CF_PARENS) |
| *szOut-- = '('; |
| szOut++; |
| |
| iRet = strlenW(szOut) + 1; |
| if (cchOut) |
| { |
| if (iRet <= cchOut) |
| memcpy(lpCurrencyStr, szOut, iRet * sizeof(WCHAR)); |
| else |
| { |
| memcpy(lpCurrencyStr, szOut, cchOut * sizeof(WCHAR)); |
| lpCurrencyStr[cchOut - 1] = '\0'; |
| SetLastError(ERROR_INSUFFICIENT_BUFFER); |
| iRet = 0; |
| } |
| } |
| return iRet; |
| } |
| |
| /* FIXME: Everything below here needs to move somewhere else along with the |
| * other EnumXXX functions, when a method for storing resources for |
| * alternate calendars is determined. |
| */ |
| |
| /************************************************************************** |
| * EnumDateFormatsExA (KERNEL32.@) |
| * |
| * FIXME: MSDN mentions only LOCALE_USE_CP_ACP, should we handle |
| * LOCALE_NOUSEROVERRIDE here as well? |
| */ |
| BOOL WINAPI EnumDateFormatsExA(DATEFMT_ENUMPROCEXA proc, LCID lcid, DWORD flags) |
| { |
| CALID cal_id; |
| char buf[256]; |
| |
| if (!proc) |
| { |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| |
| if (!GetLocaleInfoW(lcid, LOCALE_ICALENDARTYPE|LOCALE_RETURN_NUMBER, (LPWSTR)&cal_id, sizeof(cal_id)/sizeof(WCHAR))) |
| return FALSE; |
| |
| switch (flags & ~LOCALE_USE_CP_ACP) |
| { |
| case 0: |
| case DATE_SHORTDATE: |
| if (GetLocaleInfoA(lcid, LOCALE_SSHORTDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf, cal_id); |
| break; |
| |
| case DATE_LONGDATE: |
| if (GetLocaleInfoA(lcid, LOCALE_SLONGDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf, cal_id); |
| break; |
| |
| case DATE_YEARMONTH: |
| if (GetLocaleInfoA(lcid, LOCALE_SYEARMONTH | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf, cal_id); |
| break; |
| |
| default: |
| FIXME("Unknown date format (%ld)\n", flags); |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| return TRUE; |
| } |
| |
| /************************************************************************** |
| * EnumDateFormatsExW (KERNEL32.@) |
| */ |
| BOOL WINAPI EnumDateFormatsExW(DATEFMT_ENUMPROCEXW proc, LCID lcid, DWORD flags) |
| { |
| CALID cal_id; |
| WCHAR buf[256]; |
| |
| if (!proc) |
| { |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| |
| if (!GetLocaleInfoW(lcid, LOCALE_ICALENDARTYPE|LOCALE_RETURN_NUMBER, (LPWSTR)&cal_id, sizeof(cal_id)/sizeof(WCHAR))) |
| return FALSE; |
| |
| switch (flags & ~LOCALE_USE_CP_ACP) |
| { |
| case 0: |
| case DATE_SHORTDATE: |
| if (GetLocaleInfoW(lcid, LOCALE_SSHORTDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf, cal_id); |
| break; |
| |
| case DATE_LONGDATE: |
| if (GetLocaleInfoW(lcid, LOCALE_SLONGDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf, cal_id); |
| break; |
| |
| case DATE_YEARMONTH: |
| if (GetLocaleInfoW(lcid, LOCALE_SYEARMONTH | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf, cal_id); |
| break; |
| |
| default: |
| FIXME("Unknown date format (%ld)\n", flags); |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| return TRUE; |
| } |
| |
| /************************************************************************** |
| * EnumDateFormatsA (KERNEL32.@) |
| * |
| * FIXME: MSDN mentions only LOCALE_USE_CP_ACP, should we handle |
| * LOCALE_NOUSEROVERRIDE here as well? |
| */ |
| BOOL WINAPI EnumDateFormatsA(DATEFMT_ENUMPROCA proc, LCID lcid, DWORD flags) |
| { |
| char buf[256]; |
| |
| if (!proc) |
| { |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| |
| switch (flags & ~LOCALE_USE_CP_ACP) |
| { |
| case 0: |
| case DATE_SHORTDATE: |
| if (GetLocaleInfoA(lcid, LOCALE_SSHORTDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| case DATE_LONGDATE: |
| if (GetLocaleInfoA(lcid, LOCALE_SLONGDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| case DATE_YEARMONTH: |
| if (GetLocaleInfoA(lcid, LOCALE_SYEARMONTH | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| default: |
| FIXME("Unknown date format (%ld)\n", flags); |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| return TRUE; |
| } |
| |
| /************************************************************************** |
| * EnumDateFormatsW (KERNEL32.@) |
| */ |
| BOOL WINAPI EnumDateFormatsW(DATEFMT_ENUMPROCW proc, LCID lcid, DWORD flags) |
| { |
| WCHAR buf[256]; |
| |
| if (!proc) |
| { |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| |
| switch (flags & ~LOCALE_USE_CP_ACP) |
| { |
| case 0: |
| case DATE_SHORTDATE: |
| if (GetLocaleInfoW(lcid, LOCALE_SSHORTDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| case DATE_LONGDATE: |
| if (GetLocaleInfoW(lcid, LOCALE_SLONGDATE | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| case DATE_YEARMONTH: |
| if (GetLocaleInfoW(lcid, LOCALE_SYEARMONTH | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| default: |
| FIXME("Unknown date format (%ld)\n", flags); |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| return TRUE; |
| } |
| |
| /************************************************************************** |
| * EnumTimeFormatsA (KERNEL32.@) |
| * |
| * FIXME: MSDN mentions only LOCALE_USE_CP_ACP, should we handle |
| * LOCALE_NOUSEROVERRIDE here as well? |
| */ |
| BOOL WINAPI EnumTimeFormatsA(TIMEFMT_ENUMPROCA proc, LCID lcid, DWORD flags) |
| { |
| char buf[256]; |
| |
| if (!proc) |
| { |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| |
| switch (flags & ~LOCALE_USE_CP_ACP) |
| { |
| case 0: |
| if (GetLocaleInfoA(lcid, LOCALE_STIMEFORMAT | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| default: |
| FIXME("Unknown time format (%ld)\n", flags); |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| return TRUE; |
| } |
| |
| /************************************************************************** |
| * EnumTimeFormatsW (KERNEL32.@) |
| */ |
| BOOL WINAPI EnumTimeFormatsW(TIMEFMT_ENUMPROCW proc, LCID lcid, DWORD flags) |
| { |
| WCHAR buf[256]; |
| |
| if (!proc) |
| { |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| |
| switch (flags & ~LOCALE_USE_CP_ACP) |
| { |
| case 0: |
| if (GetLocaleInfoW(lcid, LOCALE_STIMEFORMAT | (flags & LOCALE_USE_CP_ACP), buf, 256)) |
| proc(buf); |
| break; |
| |
| default: |
| FIXME("Unknown time format (%ld)\n", flags); |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| return TRUE; |
| } |
| |
| /****************************************************************************** |
| * NLS_EnumCalendarInfoAW <internal> |
| * Enumerates calendar information for a specified locale. |
| * |
| * PARAMS |
| * calinfoproc [I] Pointer to the callback |
| * locale [I] The locale for which to retrieve calendar information. |
| * This parameter can be a locale identifier created by the |
| * MAKELCID macro, or one of the following values: |
| * LOCALE_SYSTEM_DEFAULT |
| * Use the default system locale. |
| * LOCALE_USER_DEFAULT |
| * Use the default user locale. |
| * calendar [I] The calendar for which information is requested, or |
| * ENUM_ALL_CALENDARS. |
| * caltype [I] The type of calendar information to be returned. Note |
| * that only one CALTYPE value can be specified per call |
| * of this function, except where noted. |
| * unicode [I] Specifies if the callback expects a unicode string. |
| * ex [I] Specifies if the callback needs the calendar identifier. |
| * |
| * RETURNS |
| * Success: TRUE. |
| * Failure: FALSE. Use GetLastError() to determine the cause. |
| * |
| * NOTES |
| * When the ANSI version of this function is used with a Unicode-only LCID, |
| * the call can succeed because the system uses the system code page. |
| * However, characters that are undefined in the system code page appear |
| * in the string as a question mark (?). |
| * |
| * TODO |
| * The above note should be respected by GetCalendarInfoA. |
| */ |
| static BOOL NLS_EnumCalendarInfoAW(void *calinfoproc, LCID locale, |
| CALID calendar, CALTYPE caltype, BOOL unicode, BOOL ex ) |
| { |
| WCHAR *buf, *opt = NULL, *iter = NULL; |
| BOOL ret = FALSE; |
| int bufSz = 200; /* the size of the buffer */ |
| |
| if (calinfoproc == NULL) |
| { |
| SetLastError(ERROR_INVALID_PARAMETER); |
| return FALSE; |
| } |
| |
| buf = HeapAlloc(GetProcessHeap(), 0, bufSz); |
| if (buf == NULL) |
| { |
| SetLastError(ERROR_NOT_ENOUGH_MEMORY); |
| return FALSE; |
| } |
| |
| if (calendar == ENUM_ALL_CALENDARS) |
| { |
| int optSz = GetLocaleInfoW(locale, LOCALE_IOPTIONALCALENDAR, NULL, 0); |
| if (optSz > 1) |
| { |
| opt = HeapAlloc(GetProcessHeap(), 0, optSz * sizeof(WCHAR)); |
| if (opt == NULL) |
| { |
| SetLastError(ERROR_NOT_ENOUGH_MEMORY); |
| goto NLS_EnumCalendarInfoAW_Cleanup; |
| } |
| if (GetLocaleInfoW(locale, LOCALE_IOPTIONALCALENDAR, opt, optSz)) |
| iter = opt; |
| } |
| calendar = NLS_GetLocaleNumber(locale, LOCALE_ICALENDARTYPE); |
| } |
| |
| while (TRUE) /* loop through calendars */ |
| { |
| do /* loop until there's no error */ |
| { |
| if (unicode) |
| ret = GetCalendarInfoW(locale, calendar, caltype, buf, bufSz / sizeof(WCHAR), NULL); |
| else ret = GetCalendarInfoA(locale, calendar, caltype, (CHAR*)buf, bufSz / sizeof(CHAR), NULL); |
| |
| if (!ret) |
| { |
| if (GetLastError() == ERROR_INSUFFICIENT_BUFFER) |
| { /* so resize it */ |
| int newSz; |
| if (unicode) |
| newSz = GetCalendarInfoW(locale, calendar, caltype, NULL, 0, NULL) * sizeof(WCHAR); |
| else newSz = GetCalendarInfoA(locale, calendar, caltype, NULL, 0, NULL) * sizeof(CHAR); |
| if (bufSz >= newSz) |
| { |
| ERR("Buffer resizing disorder: was %d, requested %d.\n", bufSz, newSz); |
| goto NLS_EnumCalendarInfoAW_Cleanup; |
| } |
| bufSz = newSz; |
| WARN("Buffer too small; resizing to %d bytes.\n", bufSz); |
| buf = HeapReAlloc(GetProcessHeap(), 0, buf, bufSz); |
| if (buf == NULL) |
| goto NLS_EnumCalendarInfoAW_Cleanup; |
| } else goto NLS_EnumCalendarInfoAW_Cleanup; |
| } |
| } while (!ret); |
| |
| /* Here we are. We pass the buffer to the correct version of |
| * the callback. Because it's not the same number of params, |
| * we must check for Ex, but we don't care about Unicode |
| * because the buffer is already in the correct format. |
| */ |
| if (ex) { |
| ret = ((CALINFO_ENUMPROCEXW)calinfoproc)(buf, calendar); |
| } else |
| ret = ((CALINFO_ENUMPROCW)calinfoproc)(buf); |
| |
| if (!ret) { /* the callback told to stop */ |
| ret = TRUE; |
| break; |
| } |
| |
| if ((iter == NULL) || (*iter == 0)) /* no more calendars */ |
| break; |
| |
| calendar = 0; |
| while ((*iter >= '0') && (*iter <= '9')) |
| calendar = calendar * 10 + *iter++ - '0'; |
| |
| if (*iter++ != 0) |
| { |
| SetLastError(ERROR_BADDB); |
| ret = FALSE; |
| break; |
| } |
| } |
| |
| NLS_EnumCalendarInfoAW_Cleanup: |
| HeapFree(GetProcessHeap(), 0, opt); |
| HeapFree(GetProcessHeap(), 0, buf); |
| return ret; |
| } |
| |
| /****************************************************************************** |
| * EnumCalendarInfoA [KERNEL32.@] |
| * |
| * See EnumCalendarInfoAW. |
| */ |
| BOOL WINAPI EnumCalendarInfoA( CALINFO_ENUMPROCA calinfoproc,LCID locale, |
| CALID calendar,CALTYPE caltype ) |
| { |
| TRACE("(%p,0x%08lx,0x%08lx,0x%08lx)\n", calinfoproc, locale, calendar, caltype); |
| return NLS_EnumCalendarInfoAW(calinfoproc, locale, calendar, caltype, FALSE, FALSE); |
| } |
| |
| /****************************************************************************** |
| * EnumCalendarInfoW [KERNEL32.@] |
| * |
| * See EnumCalendarInfoAW. |
| */ |
| BOOL WINAPI EnumCalendarInfoW( CALINFO_ENUMPROCW calinfoproc,LCID locale, |
| CALID calendar,CALTYPE caltype ) |
| { |
| TRACE("(%p,0x%08lx,0x%08lx,0x%08lx)\n", calinfoproc, locale, calendar, caltype); |
| return NLS_EnumCalendarInfoAW(calinfoproc, locale, calendar, caltype, TRUE, FALSE); |
| } |
| |
| /****************************************************************************** |
| * EnumCalendarInfoExA [KERNEL32.@] |
| * |
| * See EnumCalendarInfoAW. |
| */ |
| BOOL WINAPI EnumCalendarInfoExA( CALINFO_ENUMPROCEXA calinfoproc,LCID locale, |
| CALID calendar,CALTYPE caltype ) |
| { |
| TRACE("(%p,0x%08lx,0x%08lx,0x%08lx)\n", calinfoproc, locale, calendar, caltype); |
| return NLS_EnumCalendarInfoAW(calinfoproc, locale, calendar, caltype, FALSE, TRUE); |
| } |
| |
| /****************************************************************************** |
| * EnumCalendarInfoExW [KERNEL32.@] |
| * |
| * See EnumCalendarInfoAW. |
| */ |
| BOOL WINAPI EnumCalendarInfoExW( CALINFO_ENUMPROCEXW calinfoproc,LCID locale, |
| CALID calendar,CALTYPE caltype ) |
| { |
| TRACE("(%p,0x%08lx,0x%08lx,0x%08lx)\n", calinfoproc, locale, calendar, caltype); |
| return NLS_EnumCalendarInfoAW(calinfoproc, locale, calendar, caltype, TRUE, TRUE); |
| } |