summaryrefslogtreecommitdiffstats
path: root/svl/source/numbers/zformat.cxx
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-07 09:06:44 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-07 09:06:44 +0000
commited5640d8b587fbcfed7dd7967f3de04b37a76f26 (patch)
tree7a5f7c6c9d02226d7471cb3cc8fbbf631b415303 /svl/source/numbers/zformat.cxx
parentInitial commit. (diff)
downloadlibreoffice-upstream.tar.xz
libreoffice-upstream.zip
Adding upstream version 4:7.4.7.upstream/4%7.4.7upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'svl/source/numbers/zformat.cxx')
-rw-r--r--svl/source/numbers/zformat.cxx6035
1 files changed, 6035 insertions, 0 deletions
diff --git a/svl/source/numbers/zformat.cxx b/svl/source/numbers/zformat.cxx
new file mode 100644
index 000000000..0ddd6e084
--- /dev/null
+++ b/svl/source/numbers/zformat.cxx
@@ -0,0 +1,6035 @@
+/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+/*
+ * This file is part of the LibreOffice project.
+ *
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/.
+ *
+ * This file incorporates work covered by the following license notice:
+ *
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed
+ * with this work for additional information regarding copyright
+ * ownership. The ASF licenses this file to you under the Apache
+ * License, Version 2.0 (the "License"); you may not use this file
+ * except in compliance with the License. You may obtain a copy of
+ * the License at http://www.apache.org/licenses/LICENSE-2.0 .
+ */
+
+#include <stdio.h>
+#include <string_view>
+
+#include <o3tl/string_view.hxx>
+#include <comphelper/string.hxx>
+#include <sal/log.hxx>
+#include <tools/debug.hxx>
+#include <tools/long.hxx>
+#include <i18nlangtag/mslangid.hxx>
+#include <rtl/math.hxx>
+#include <unotools/charclass.hxx>
+#include <unotools/calendarwrapper.hxx>
+#include <unotools/nativenumberwrapper.hxx>
+#include <com/sun/star/i18n/CalendarFieldIndex.hpp>
+#include <com/sun/star/i18n/CalendarDisplayIndex.hpp>
+#include <com/sun/star/i18n/CalendarDisplayCode.hpp>
+#include <com/sun/star/i18n/AmPmValue.hpp>
+#include <com/sun/star/i18n/NativeNumberMode.hpp>
+#include <com/sun/star/i18n/NativeNumberXmlAttributes2.hpp>
+
+#include <svl/zformat.hxx>
+#include "zforscan.hxx"
+
+#include "zforfind.hxx"
+#include <svl/zforlist.hxx>
+#include <unotools/digitgroupingiterator.hxx>
+#include <svl/nfsymbol.hxx>
+
+#include <cmath>
+#include <array>
+
+using namespace svt;
+
+namespace {
+
+constexpr OUStringLiteral GREGORIAN = u"gregorian";
+
+const sal_uInt16 UPPER_PRECISION = 300; // entirely arbitrary...
+const double EXP_LOWER_BOUND = 1.0E-4; // prefer scientific notation below this value.
+const double EXP_ABS_UPPER_BOUND = 1.0E15; // use exponential notation above that absolute value.
+ // Back in time was E16 that lead
+ // to display rounding errors, see
+ // also sal/rtl/math.cxx
+ // doubleToString()
+
+} // namespace
+
+const double D_MAX_U_INT32 = double(0xffffffff); // 4294967295.0
+constexpr double D_MAX_INTEGER = (sal_uInt64(1) << 53) - 1;
+
+const double D_MAX_D_BY_100 = 1.7E306;
+const double D_MIN_M_BY_1000 = 2.3E-305;
+
+const sal_uInt8 cCharWidths[ 128-32 ] = {
+ 1,1,1,2,2,3,2,1,1,1,1,2,1,1,1,1,
+ 2,2,2,2,2,2,2,2,2,2,1,1,2,2,2,2,
+ 3,2,2,2,2,2,2,3,2,1,2,2,2,3,3,3,
+ 2,3,2,2,2,2,2,3,2,2,2,1,1,1,2,2,
+ 1,2,2,2,2,2,1,2,2,1,1,2,1,3,2,2,
+ 2,2,1,2,1,2,2,2,2,2,2,1,1,1,2,1
+};
+
+// static
+sal_Int32 SvNumberformat::InsertBlanks( OUStringBuffer& r, sal_Int32 nPos, sal_Unicode c )
+{
+ if( c >= 32 )
+ {
+ int n = 2; // Default for chars > 128 (HACK!)
+ if( c <= 127 )
+ {
+ n = static_cast<int>(cCharWidths[ c - 32 ]);
+ }
+ while( n-- )
+ {
+ r.insert( nPos++, ' ');
+ }
+ }
+ return nPos;
+}
+
+static tools::Long GetPrecExp( double fAbsVal )
+{
+ DBG_ASSERT( fAbsVal > 0.0, "GetPrecExp: fAbsVal <= 0.0" );
+ if ( fAbsVal < 1e-7 || fAbsVal > 1e7 )
+ {
+ // Shear: whether it's faster or not, falls in between 1e6 and 1e7
+ return static_cast<tools::Long>(floor( log10( fAbsVal ) )) + 1;
+ }
+ else
+ {
+ tools::Long nPrecExp = 1;
+ while( fAbsVal < 1 )
+ {
+ fAbsVal *= 10;
+ nPrecExp--;
+ }
+ while( fAbsVal >= 10 )
+ {
+ fAbsVal /= 10;
+ nPrecExp++;
+ }
+ return nPrecExp;
+ }
+}
+
+/**
+ * SvNumberformatInfo
+ * */
+
+void ImpSvNumberformatInfo::Copy( const ImpSvNumberformatInfo& rNumFor, sal_uInt16 nCnt )
+{
+ for (sal_uInt16 i = 0; i < nCnt; ++i)
+ {
+ sStrArray[i] = rNumFor.sStrArray[i];
+ nTypeArray[i] = rNumFor.nTypeArray[i];
+ }
+ eScannedType = rNumFor.eScannedType;
+ bThousand = rNumFor.bThousand;
+ nThousand = rNumFor.nThousand;
+ nCntPre = rNumFor.nCntPre;
+ nCntPost = rNumFor.nCntPost;
+ nCntExp = rNumFor.nCntExp;
+}
+
+const std::map<LanguageType, std::array<sal_uInt8, 4>> tblDBNumToNatNum
+ = { { primary(LANGUAGE_CHINESE), { 4, 5, 3, 0 } },
+ { primary(LANGUAGE_JAPANESE), { 4, 5, 3, 0 } },
+ { primary(LANGUAGE_KOREAN), { 4, 5, 6, 10 } } };
+
+// static
+sal_uInt8 SvNumberNatNum::MapDBNumToNatNum( sal_uInt8 nDBNum, LanguageType eLang, bool bDate )
+{
+ sal_uInt8 nNatNum = 0;
+ eLang = MsLangId::getRealLanguage( eLang ); // resolve SYSTEM etc.
+ eLang = primary(eLang); // 10 bit primary language
+ if ( bDate )
+ {
+ if ( nDBNum == 4 && eLang == primary(LANGUAGE_KOREAN) )
+ {
+ nNatNum = 10;
+ }
+ else if ( nDBNum <= 3 )
+ {
+ nNatNum = nDBNum; // known to be good for: zh,ja,ko / 1,2,3
+ }
+ }
+ else
+ {
+ if (1 <= nDBNum && nDBNum <= 4)
+ {
+ auto const it = tblDBNumToNatNum.find(eLang);
+ if (it != tblDBNumToNatNum.end())
+ nNatNum = it->second[nDBNum - 1];
+
+ }
+ }
+ return nNatNum;
+}
+
+const std::map<LanguageType, std::array<sal_uInt8, 9>> tblNatNumToDBNum
+ = { { primary(LANGUAGE_CHINESE), { 1, 0, 0, 1, 2, 3, 0, 0, 0 } },
+ { primary(LANGUAGE_JAPANESE), { 1, 2, 3, 1, 2, 3, 1, 2, 0 } },
+ { primary(LANGUAGE_KOREAN), { 1, 2, 3, 1, 2, 3, 1, 2, 4 } } };
+
+// static
+sal_uInt8 SvNumberNatNum::MapNatNumToDBNum( sal_uInt8 nNatNum, LanguageType eLang, bool bDate )
+{
+ sal_uInt8 nDBNum = 0;
+ eLang = MsLangId::getRealLanguage( eLang ); // resolve SYSTEM etc.
+ eLang = primary(eLang); // 10 bit primary language
+ if ( bDate )
+ {
+ if ( nNatNum == 10 && eLang == primary(LANGUAGE_KOREAN) )
+ {
+ nDBNum = 4;
+ }
+ else if ( nNatNum <= 3 )
+ {
+ nDBNum = nNatNum; // known to be good for: zh,ja,ko / 1,2,3
+ }
+ }
+ else
+ {
+ if (1 <= nNatNum && nNatNum <= 9)
+ {
+ auto const it = tblNatNumToDBNum.find(eLang);
+ if (it != tblNatNumToDBNum.end())
+ nDBNum = it->second[nNatNum - 1];
+ }
+ }
+ return nDBNum;
+}
+
+/**
+ * SvNumFor
+ */
+
+ImpSvNumFor::ImpSvNumFor()
+{
+ nStringsCnt = 0;
+ aI.eScannedType = SvNumFormatType::UNDEFINED;
+ aI.bThousand = false;
+ aI.nThousand = 0;
+ aI.nCntPre = 0;
+ aI.nCntPost = 0;
+ aI.nCntExp = 0;
+ pColor = nullptr;
+}
+
+ImpSvNumFor::~ImpSvNumFor()
+{
+}
+
+void ImpSvNumFor::Enlarge(sal_uInt16 nCnt)
+{
+ if ( nStringsCnt != nCnt )
+ {
+ nStringsCnt = nCnt;
+ aI.nTypeArray.resize(nCnt);
+ aI.sStrArray.resize(nCnt);
+ }
+}
+
+void ImpSvNumFor::Copy( const ImpSvNumFor& rNumFor, const ImpSvNumberformatScan* pSc )
+{
+ Enlarge( rNumFor.nStringsCnt );
+ aI.Copy( rNumFor.aI, nStringsCnt );
+ sColorName = rNumFor.sColorName;
+ if ( pSc )
+ {
+ pColor = pSc->GetColor( sColorName ); // #121103# don't copy pointer between documents
+ }
+ else
+ {
+ pColor = rNumFor.pColor;
+ }
+ aNatNum = rNumFor.aNatNum;
+}
+
+bool ImpSvNumFor::HasNewCurrency() const
+{
+ for ( sal_uInt16 j=0; j<nStringsCnt; j++ )
+ {
+ if ( aI.nTypeArray[j] == NF_SYMBOLTYPE_CURRENCY )
+ {
+ return true;
+ }
+ }
+ return false;
+}
+
+bool ImpSvNumFor::GetNewCurrencySymbol( OUString& rSymbol,
+ OUString& rExtension ) const
+{
+ for ( sal_uInt16 j=0; j<nStringsCnt; j++ )
+ {
+ if ( aI.nTypeArray[j] == NF_SYMBOLTYPE_CURRENCY )
+ {
+ rSymbol = aI.sStrArray[j];
+ if ( j < nStringsCnt-1 && aI.nTypeArray[j+1] == NF_SYMBOLTYPE_CURREXT )
+ {
+ rExtension = aI.sStrArray[j+1];
+ }
+ else
+ {
+ rExtension.clear();
+ }
+ return true;
+ }
+ }
+ //! No Erase at rSymbol, rExtension
+ return false;
+}
+
+/**
+ * SvNumberformat
+ */
+
+namespace {
+
+enum BracketFormatSymbolType
+{
+ BRACKET_SYMBOLTYPE_FORMAT = -1, // subformat string
+ BRACKET_SYMBOLTYPE_COLOR = -2, // color
+ BRACKET_SYMBOLTYPE_ERROR = -3, // error
+ BRACKET_SYMBOLTYPE_DBNUM1 = -4, // DoubleByteNumber, represent numbers
+ BRACKET_SYMBOLTYPE_DBNUM2 = -5, // using CJK characters, Excel compatible
+ BRACKET_SYMBOLTYPE_DBNUM3 = -6,
+ BRACKET_SYMBOLTYPE_DBNUM4 = -7,
+ BRACKET_SYMBOLTYPE_DBNUM5 = -8,
+ BRACKET_SYMBOLTYPE_DBNUM6 = -9,
+ BRACKET_SYMBOLTYPE_DBNUM7 = -10,
+ BRACKET_SYMBOLTYPE_DBNUM8 = -11,
+ BRACKET_SYMBOLTYPE_DBNUM9 = -12,
+ BRACKET_SYMBOLTYPE_LOCALE = -13,
+ BRACKET_SYMBOLTYPE_NATNUM0 = -14, // Our NativeNumber support, ASCII
+ BRACKET_SYMBOLTYPE_NATNUM1 = -15, // Our NativeNumber support, represent
+ BRACKET_SYMBOLTYPE_NATNUM2 = -16, // numbers using CJK, CTL, ...
+ BRACKET_SYMBOLTYPE_NATNUM3 = -17,
+ BRACKET_SYMBOLTYPE_NATNUM4 = -18,
+ BRACKET_SYMBOLTYPE_NATNUM5 = -19,
+ BRACKET_SYMBOLTYPE_NATNUM6 = -20,
+ BRACKET_SYMBOLTYPE_NATNUM7 = -21,
+ BRACKET_SYMBOLTYPE_NATNUM8 = -22,
+ BRACKET_SYMBOLTYPE_NATNUM9 = -23,
+ BRACKET_SYMBOLTYPE_NATNUM10 = -24,
+ BRACKET_SYMBOLTYPE_NATNUM11 = -25,
+ BRACKET_SYMBOLTYPE_NATNUM12 = -26,
+ BRACKET_SYMBOLTYPE_NATNUM13 = -27,
+ BRACKET_SYMBOLTYPE_NATNUM14 = -28,
+ BRACKET_SYMBOLTYPE_NATNUM15 = -29,
+ BRACKET_SYMBOLTYPE_NATNUM16 = -30,
+ BRACKET_SYMBOLTYPE_NATNUM17 = -31,
+ BRACKET_SYMBOLTYPE_NATNUM18 = -32,
+ BRACKET_SYMBOLTYPE_NATNUM19 = -33
+};
+
+}
+
+void SvNumberformat::ImpCopyNumberformat( const SvNumberformat& rFormat )
+{
+ sFormatstring = rFormat.sFormatstring;
+ eType = rFormat.eType;
+ maLocale = rFormat.maLocale;
+ fLimit1 = rFormat.fLimit1;
+ fLimit2 = rFormat.fLimit2;
+ eOp1 = rFormat.eOp1;
+ eOp2 = rFormat.eOp2;
+ bStandard = rFormat.bStandard;
+ bIsUsed = rFormat.bIsUsed;
+ sComment = rFormat.sComment;
+ bAdditionalBuiltin = rFormat.bAdditionalBuiltin;
+
+ // #121103# when copying between documents, get color pointers from own scanner
+ ImpSvNumberformatScan* pColorSc = ( &rScan != &rFormat.rScan ) ? &rScan : nullptr;
+
+ for (sal_uInt16 i = 0; i < 4; i++)
+ {
+ NumFor[i].Copy(rFormat.NumFor[i], pColorSc);
+ }
+}
+
+SvNumberformat::SvNumberformat( SvNumberformat const & rFormat )
+ : rScan(rFormat.rScan), bStarFlag( rFormat.bStarFlag )
+{
+ ImpCopyNumberformat( rFormat );
+}
+
+SvNumberformat::SvNumberformat( SvNumberformat const & rFormat, ImpSvNumberformatScan& rSc )
+ : rScan(rSc)
+ , bStarFlag( rFormat.bStarFlag )
+{
+ ImpCopyNumberformat( rFormat );
+}
+
+static bool lcl_SvNumberformat_IsBracketedPrefix( short nSymbolType )
+{
+ if ( nSymbolType > 0 )
+ {
+ return true; // conditions
+ }
+ switch ( nSymbolType )
+ {
+ case BRACKET_SYMBOLTYPE_COLOR :
+ case BRACKET_SYMBOLTYPE_DBNUM1 :
+ case BRACKET_SYMBOLTYPE_DBNUM2 :
+ case BRACKET_SYMBOLTYPE_DBNUM3 :
+ case BRACKET_SYMBOLTYPE_DBNUM4 :
+ case BRACKET_SYMBOLTYPE_DBNUM5 :
+ case BRACKET_SYMBOLTYPE_DBNUM6 :
+ case BRACKET_SYMBOLTYPE_DBNUM7 :
+ case BRACKET_SYMBOLTYPE_DBNUM8 :
+ case BRACKET_SYMBOLTYPE_DBNUM9 :
+ case BRACKET_SYMBOLTYPE_LOCALE :
+ case BRACKET_SYMBOLTYPE_NATNUM0 :
+ case BRACKET_SYMBOLTYPE_NATNUM1 :
+ case BRACKET_SYMBOLTYPE_NATNUM2 :
+ case BRACKET_SYMBOLTYPE_NATNUM3 :
+ case BRACKET_SYMBOLTYPE_NATNUM4 :
+ case BRACKET_SYMBOLTYPE_NATNUM5 :
+ case BRACKET_SYMBOLTYPE_NATNUM6 :
+ case BRACKET_SYMBOLTYPE_NATNUM7 :
+ case BRACKET_SYMBOLTYPE_NATNUM8 :
+ case BRACKET_SYMBOLTYPE_NATNUM9 :
+ case BRACKET_SYMBOLTYPE_NATNUM10 :
+ case BRACKET_SYMBOLTYPE_NATNUM11 :
+ case BRACKET_SYMBOLTYPE_NATNUM12 :
+ case BRACKET_SYMBOLTYPE_NATNUM13 :
+ case BRACKET_SYMBOLTYPE_NATNUM14 :
+ case BRACKET_SYMBOLTYPE_NATNUM15 :
+ case BRACKET_SYMBOLTYPE_NATNUM16 :
+ case BRACKET_SYMBOLTYPE_NATNUM17 :
+ case BRACKET_SYMBOLTYPE_NATNUM18 :
+ case BRACKET_SYMBOLTYPE_NATNUM19 :
+ return true;
+ }
+ return false;
+}
+
+/** Import extended LCID from Excel
+ */
+OUString SvNumberformat::ImpObtainCalendarAndNumerals( OUStringBuffer& rString, sal_Int32 nPos,
+ LanguageType& nLang, const LocaleType& aTmpLocale )
+{
+ OUString sCalendar;
+ sal_uInt16 nNatNum = 0;
+ LanguageType nLocaleLang = MsLangId::getRealLanguage( maLocale.meLanguage );
+ LanguageType nTmpLocaleLang = MsLangId::getRealLanguage( aTmpLocale.meLanguage );
+ /* NOTE: enhancement to allow other possible locale dependent
+ * calendars and numerals. BUT only if our locale data allows it! For LCID
+ * numerals and calendars see
+ * http://office.microsoft.com/en-us/excel/HA010346351033.aspx
+ * Calendar is inserted after
+ * all prefixes have been consumed as it is actually a format modifier
+ * and not a prefix.
+ * Currently calendars are tied to the locale of the entire number
+ * format, e.g. [~buddhist] in en_US doesn't work.
+ * => Having different locales in sub formats does not work!
+ * */
+ /* TODO: calendars could be tied to a sub format's NatNum info
+ * instead, or even better be available for any locale. Needs a
+ * different implementation of GetCal() and locale data calendars.
+ * */
+ switch ( aTmpLocale.mnCalendarType & 0x7F )
+ {
+ case 0x03 : // Gengou calendar
+ // Only Japanese language support Gengou calendar.
+ // It is an implicit "other" calendar where E, EE, R and RR
+ // automatically switch to and YY and YYYY switch to Gregorian. Do
+ // not add the "[~gengou]" modifier.
+ if ( nLocaleLang != LANGUAGE_JAPANESE )
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_JAPANESE;
+ }
+ break;
+ case 0x05 : // Korean Dangi calendar
+ sCalendar = "[~dangi]";
+ // Only Korean language support dangi calendar
+ if ( nLocaleLang != LANGUAGE_KOREAN )
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_KOREAN;
+ }
+ break;
+ case 0x06 : // Hijri calendar
+ case 0x17 : // same?
+ sCalendar = "[~hijri]";
+ // Only Arabic or Farsi languages support Hijri calendar
+ if ( ( primary( nLocaleLang ) != LANGUAGE_ARABIC_PRIMARY_ONLY )
+ && nLocaleLang != LANGUAGE_FARSI )
+ {
+ if ( ( primary( nTmpLocaleLang ) == LANGUAGE_ARABIC_PRIMARY_ONLY )
+ || nTmpLocaleLang == LANGUAGE_FARSI )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_ARABIC_SAUDI_ARABIA;
+ }
+ }
+ break;
+ case 0x07 : // Buddhist calendar
+ sCalendar="[~buddhist]";
+ // Only Thai or Lao languages support Buddhist calendar
+ if ( nLocaleLang != LANGUAGE_THAI && nLocaleLang != LANGUAGE_LAO )
+ {
+ if ( nTmpLocaleLang == LANGUAGE_THAI || nTmpLocaleLang == LANGUAGE_LAO )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_THAI;
+ }
+ }
+ break;
+ case 0x08 : // Hebrew calendar
+ sCalendar = "[~jewish]";
+ // Many languages (but not all) support Jewish calendar
+ // Unable to find any logic => keep same language
+ break;
+ case 0x0E : // unknown calendar
+ case 0x0F : // unknown calendar
+ case 0x10 : // Indian calendar (unsupported)
+ case 0x11 : // unknown calendar
+ case 0x12 : // unknown calendar
+ case 0x13 : // unknown calendar
+ default : // other calendars (see tdf#36038) are not handle by LibO
+ break;
+ }
+ /** Reference language for each numeral ID */
+ static const LanguageType aNumeralIDtoLanguage []=
+ {
+ LANGUAGE_DONTKNOW, // 0x00
+ LANGUAGE_ENGLISH_US, // 0x01
+ LANGUAGE_ARABIC_SAUDI_ARABIA, // 0x02 + all Arabic
+ LANGUAGE_FARSI, // 0x03
+ LANGUAGE_HINDI, // 0x04 + Devanagari
+ LANGUAGE_BENGALI, // 0x05
+ LANGUAGE_PUNJABI, // 0x06
+ LANGUAGE_GUJARATI, // 0x07
+ LANGUAGE_ODIA, // 0x08
+ LANGUAGE_TAMIL, // 0x09
+ LANGUAGE_TELUGU, // 0x0A
+ LANGUAGE_KANNADA, // 0x0B
+ LANGUAGE_MALAYALAM, // 0x0C
+ LANGUAGE_THAI, // 0x0D
+ LANGUAGE_LAO, // 0x0E
+ LANGUAGE_TIBETAN, // 0x0F
+ LANGUAGE_BURMESE, // 0x10
+ LANGUAGE_TIGRIGNA_ETHIOPIA, // 0x11
+ LANGUAGE_KHMER, // 0x12
+ LANGUAGE_MONGOLIAN_MONGOLIAN_MONGOLIA, // 0x13
+ LANGUAGE_DONTKNOW, // 0x14
+ LANGUAGE_DONTKNOW, // 0x15
+ LANGUAGE_DONTKNOW, // 0x16
+ LANGUAGE_DONTKNOW, // 0x17
+ LANGUAGE_DONTKNOW, // 0x18
+ LANGUAGE_DONTKNOW, // 0x19
+ LANGUAGE_DONTKNOW, // 0x1A
+ LANGUAGE_JAPANESE, // 0x1B
+ LANGUAGE_JAPANESE, // 0x1C
+ LANGUAGE_JAPANESE, // 0x1D
+ LANGUAGE_CHINESE_SIMPLIFIED, // 0x1E
+ LANGUAGE_CHINESE_SIMPLIFIED, // 0x1F
+ LANGUAGE_CHINESE_SIMPLIFIED, // 0x20
+ LANGUAGE_CHINESE_TRADITIONAL, // 0x21
+ LANGUAGE_CHINESE_TRADITIONAL, // 0x22
+ LANGUAGE_CHINESE_TRADITIONAL, // 0x23
+ LANGUAGE_KOREAN, // 0x24
+ LANGUAGE_KOREAN, // 0x25
+ LANGUAGE_KOREAN, // 0x26
+ LANGUAGE_KOREAN // 0x27
+ };
+
+ sal_uInt16 nNumeralID = aTmpLocale.mnNumeralShape & 0x7F;
+ LanguageType nReferenceLanguage = nNumeralID <= 0x27 ? aNumeralIDtoLanguage[nNumeralID] : LANGUAGE_DONTKNOW;
+
+ switch ( nNumeralID )
+ {
+ // Regular cases: all languages with same primary mask use same numerals
+ case 0x03 : // Perso-Arabic (Farsi) numerals
+ case 0x05 : // Bengali numerals
+ case 0x06 : // Punjabi numerals
+ case 0x07 : // Gujarati numerals
+ case 0x08 : // Odia (Orya) numerals
+ case 0x09 : // Tamil numerals
+ case 0x0A : // Telugu numerals
+ case 0x0B : // Kannada numerals
+ case 0x0C : // Malayalam numerals
+ case 0x0D : // Thai numerals
+ case 0x0E : // Lao numerals
+ case 0x0F : // Tibetan numerals
+ case 0x10 : // Burmese (Myanmar) numerals
+ case 0x11 : // Tigrigna (Ethiopia) numerals
+ case 0x12 : // Khmer numerals
+ if ( primary( nLocaleLang ) != primary( nReferenceLanguage ) )
+ {
+ if ( primary( nTmpLocaleLang ) == primary( nReferenceLanguage ) )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = nReferenceLanguage;
+ }
+ }
+ break;
+ // Special cases
+ case 0x04 : // Devanagari (Hindi) numerals
+ // same numerals (Devanagari) for languages with different primary masks
+ if ( nLocaleLang != LANGUAGE_HINDI && nLocaleLang != LANGUAGE_MARATHI
+ && primary( nLocaleLang ) != primary( LANGUAGE_NEPALI ) )
+ {
+ if ( nTmpLocaleLang == LANGUAGE_HINDI || nTmpLocaleLang == LANGUAGE_MARATHI
+ || primary( nTmpLocaleLang ) == primary( LANGUAGE_NEPALI ) )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_HINDI;
+ }
+ }
+ break;
+ case 0x13 : // Mongolian numerals
+ // not all Mongolian languages use Mongolian numerals
+ if ( nLocaleLang != LANGUAGE_MONGOLIAN_MONGOLIAN_MONGOLIA
+ && nLocaleLang != LANGUAGE_MONGOLIAN_MONGOLIAN_CHINA
+ && nLocaleLang != LANGUAGE_MONGOLIAN_MONGOLIAN_LSO )
+ {
+ if ( nTmpLocaleLang == LANGUAGE_MONGOLIAN_MONGOLIAN_MONGOLIA
+ || nTmpLocaleLang == LANGUAGE_MONGOLIAN_MONGOLIAN_CHINA
+ || nTmpLocaleLang == LANGUAGE_MONGOLIAN_MONGOLIAN_LSO )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_MONGOLIAN_MONGOLIAN_MONGOLIA;
+ }
+ }
+ break;
+ case 0x02 : // Eastern-Arabic numerals
+ // all arabic primary mask + LANGUAGE_PUNJABI_ARABIC_LSO
+ if ( primary( nLocaleLang ) != LANGUAGE_ARABIC_PRIMARY_ONLY
+ && nLocaleLang != LANGUAGE_PUNJABI_ARABIC_LSO )
+ {
+ if ( primary( nTmpLocaleLang ) == LANGUAGE_ARABIC_PRIMARY_ONLY
+ || nTmpLocaleLang != LANGUAGE_PUNJABI_ARABIC_LSO )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = nReferenceLanguage;
+ }
+ }
+ break;
+ // CJK numerals
+ case 0x1B : // simple Asian numerals, Japanese
+ case 0x1C : // financial Asian numerals, Japanese
+ case 0x1D : // Arabic fullwidth numerals, Japanese
+ case 0x24 : // simple Asian numerals, Korean
+ case 0x25 : // financial Asian numerals, Korean
+ case 0x26 : // Arabic fullwidth numerals, Korean
+ case 0x27 : // Korean Hangul numerals
+ // Japanese and Korean are regular
+ if ( primary( nLocaleLang ) != primary( nReferenceLanguage ) )
+ {
+ if ( primary( nTmpLocaleLang ) == primary( nReferenceLanguage ) )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = nReferenceLanguage;
+ }
+ }
+ [[fallthrough]];
+ case 0x1E : // simple Asian numerals, Chinese-PRC
+ case 0x1F : // financial Asian numerals, Chinese-PRC
+ case 0x20 : // Arabic fullwidth numerals, Chinese-PRC
+ case 0x21 : // simple Asian numerals, Chinese-Taiwan
+ case 0x22 : // financial Asian numerals, Chinese-Taiwan
+ case 0x23 : // Arabic fullwidth numerals, Chinese-Taiwan
+ nNatNum = nNumeralID == 0x27 ? 9 : ( ( nNumeralID - 0x1B ) % 3 ) + 1;
+ // [NatNum1] simple numerals
+ // [natNum2] financial numerals
+ // [NatNum3] Arabic fullwidth numerals
+ // Chinese simplified and Chinese traditional have same primary mask
+ // Chinese-PRC
+ if ( nReferenceLanguage == LANGUAGE_CHINESE_SIMPLIFIED
+ && nLocaleLang != LANGUAGE_CHINESE_SIMPLIFIED
+ && nLocaleLang != LANGUAGE_CHINESE_SINGAPORE
+ && nLocaleLang != LANGUAGE_CHINESE_LSO )
+ {
+ if ( nTmpLocaleLang == LANGUAGE_CHINESE_SIMPLIFIED
+ || nTmpLocaleLang == LANGUAGE_CHINESE_SINGAPORE
+ || nTmpLocaleLang == LANGUAGE_CHINESE_LSO )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_CHINESE_SIMPLIFIED;
+ }
+ }
+ // Chinese-Taiwan
+ else if ( nReferenceLanguage == LANGUAGE_CHINESE_TRADITIONAL
+ && nLocaleLang != LANGUAGE_CHINESE_TRADITIONAL
+ && nLocaleLang != LANGUAGE_CHINESE_HONGKONG
+ && nLocaleLang != LANGUAGE_CHINESE_MACAU )
+ {
+ if ( nTmpLocaleLang == LANGUAGE_CHINESE_TRADITIONAL
+ || nTmpLocaleLang == LANGUAGE_CHINESE_HONGKONG
+ || nTmpLocaleLang == LANGUAGE_CHINESE_MACAU )
+ {
+ nLang = maLocale.meLanguage = aTmpLocale.meLanguage;
+ }
+ else
+ {
+ nLang = maLocale.meLanguage = LANGUAGE_CHINESE_TRADITIONAL;
+ }
+ }
+ break;
+ }
+ if ( nNumeralID >= 0x02 && nNumeralID <= 0x13 )
+ nNatNum = 1;
+ if ( nNatNum )
+ rString.insert( nPos, OUStringConcatenation("[NatNum"+OUString::number(nNatNum)+"]"));
+ return sCalendar;
+}
+
+namespace
+{
+bool NatNumTakesParameters(sal_Int16 nNum)
+{
+ return (nNum == css::i18n::NativeNumberMode::NATNUM12);
+}
+}
+
+SvNumberformat::SvNumberformat(OUString& rString,
+ ImpSvNumberformatScan* pSc,
+ ImpSvNumberInputScan* pISc,
+ sal_Int32& nCheckPos,
+ LanguageType& eLan,
+ bool bReplaceBooleanEquivalent)
+ : rScan(*pSc)
+ , bAdditionalBuiltin( false )
+ , bStarFlag( false )
+{
+ if (bReplaceBooleanEquivalent)
+ rScan.ReplaceBooleanEquivalent( rString);
+
+ OUStringBuffer sBuff(rString);
+
+ // If the group (AKA thousand) separator is a No-Break Space (French)
+ // replace all occurrences by a simple space.
+ // The same for Narrow No-Break Space just in case some locale uses it.
+ // The tokens will be changed to the LocaleData separator again later on.
+ const OUString& rThSep = GetFormatter().GetNumThousandSep();
+ if ( rThSep.getLength() == 1)
+ {
+ const sal_Unicode cNBSp = 0xA0;
+ const sal_Unicode cNNBSp = 0x202F;
+ if (rThSep[0] == cNBSp )
+ sBuff.replace( cNBSp, ' ');
+ else if (rThSep[0] == cNNBSp )
+ sBuff.replace( cNNBSp, ' ');
+ }
+
+ OUString aConvertFromDecSep;
+ OUString aConvertToDecSep;
+ if (rScan.GetConvertMode())
+ {
+ aConvertFromDecSep = GetFormatter().GetNumDecimalSep();
+ maLocale.meLanguage = rScan.GetNewLnge();
+ eLan = maLocale.meLanguage; // Make sure to return switch
+ }
+ else
+ {
+ maLocale.meLanguage = eLan;
+ }
+ bStandard = false;
+ bIsUsed = false;
+ fLimit1 = 0.0;
+ fLimit2 = 0.0;
+ eOp1 = NUMBERFORMAT_OP_NO;
+ eOp2 = NUMBERFORMAT_OP_NO;
+ eType = SvNumFormatType::DEFINED;
+
+ bool bCancel = false;
+ bool bCondition = false;
+ short eSymbolType;
+ sal_Int32 nPos = 0;
+ sal_Int32 nPosOld;
+ nCheckPos = 0;
+
+ // Split into 4 sub formats
+ sal_uInt16 nIndex;
+ for ( nIndex = 0; nIndex < 4 && !bCancel; nIndex++ )
+ {
+ // Original language/country may have to be reestablished
+ if (rScan.GetConvertMode())
+ {
+ rScan.GetNumberformatter()->ChangeIntl(rScan.GetTmpLnge());
+ }
+ OUString sInsertCalendar; // a calendar resulting from parsing LCID
+ OUString sStr;
+ nPosOld = nPos; // Start position of substring
+ // first get bracketed prefixes; e.g. conditions, color
+ do
+ {
+ eSymbolType = ImpNextSymbol(sBuff, nPos, sStr);
+ if (eSymbolType > 0) // condition
+ {
+ if ( nIndex == 0 && !bCondition )
+ {
+ bCondition = true;
+ eOp1 = static_cast<SvNumberformatLimitOps>(eSymbolType);
+ }
+ else if ( nIndex == 1 && bCondition )
+ {
+ eOp2 = static_cast<SvNumberformatLimitOps>(eSymbolType);
+ }
+ else // error
+ {
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ }
+ if (!bCancel)
+ {
+ double fNumber;
+ sal_Int32 nCntChars = ImpGetNumber(sBuff, nPos, sStr);
+ if (nCntChars > 0)
+ {
+ sal_Int32 nDecPos;
+ SvNumFormatType F_Type = SvNumFormatType::UNDEFINED;
+ if (!pISc->IsNumberFormat(sStr, F_Type, fNumber, nullptr, SvNumInputOptions::NONE) ||
+ ( F_Type != SvNumFormatType::NUMBER &&
+ F_Type != SvNumFormatType::SCIENTIFIC) )
+ {
+ fNumber = 0.0;
+ nPos = nPos - nCntChars;
+ sBuff.remove(nPos, nCntChars);
+ sBuff.insert(nPos, '0');
+ nPos++;
+ }
+ else if (rScan.GetConvertMode() && ((nDecPos = sStr.indexOf( aConvertFromDecSep)) >= 0))
+ {
+ if (aConvertToDecSep.isEmpty())
+ aConvertToDecSep = GetFormatter().GetLangDecimalSep( rScan.GetNewLnge());
+ if (aConvertToDecSep != aConvertFromDecSep)
+ {
+ const OUString aStr( sStr.replaceAt( nDecPos,
+ aConvertFromDecSep.getLength(), aConvertToDecSep));
+ nPos = nPos - nCntChars;
+ sBuff.remove(nPos, nCntChars);
+ sBuff.insert(nPos, aStr);
+ nPos += aStr.getLength();
+ }
+ }
+ }
+ else
+ {
+ fNumber = 0.0;
+ sBuff.insert(nPos++, '0');
+ }
+ if (nIndex == 0)
+ {
+ fLimit1 = fNumber;
+ }
+ else
+ {
+ fLimit2 = fNumber;
+ }
+ if ( nPos < sBuff.getLength() && sBuff[nPos] == ']' )
+ {
+ nPos++;
+ }
+ else
+ {
+ bCancel = true; // break for
+ nCheckPos = nPos;
+ }
+ }
+ nPosOld = nPos; // position before string
+ }
+ else if ( lcl_SvNumberformat_IsBracketedPrefix( eSymbolType ) )
+ {
+ OUString sSymbol( sStr);
+ switch ( eSymbolType )
+ {
+ case BRACKET_SYMBOLTYPE_COLOR :
+ if ( NumFor[nIndex].GetColor() != nullptr )
+ { // error, more than one color
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ }
+ else
+ {
+ const Color* pColor = pSc->GetColor( sStr);
+ NumFor[nIndex].SetColor( pColor, sStr);
+ if (pColor == nullptr)
+ { // error
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ }
+ }
+ break;
+ case BRACKET_SYMBOLTYPE_NATNUM0 :
+ case BRACKET_SYMBOLTYPE_NATNUM1 :
+ case BRACKET_SYMBOLTYPE_NATNUM2 :
+ case BRACKET_SYMBOLTYPE_NATNUM3 :
+ case BRACKET_SYMBOLTYPE_NATNUM4 :
+ case BRACKET_SYMBOLTYPE_NATNUM5 :
+ case BRACKET_SYMBOLTYPE_NATNUM6 :
+ case BRACKET_SYMBOLTYPE_NATNUM7 :
+ case BRACKET_SYMBOLTYPE_NATNUM8 :
+ case BRACKET_SYMBOLTYPE_NATNUM9 :
+ case BRACKET_SYMBOLTYPE_NATNUM10 :
+ case BRACKET_SYMBOLTYPE_NATNUM11 :
+ case BRACKET_SYMBOLTYPE_NATNUM12 :
+ case BRACKET_SYMBOLTYPE_NATNUM13 :
+ case BRACKET_SYMBOLTYPE_NATNUM14 :
+ case BRACKET_SYMBOLTYPE_NATNUM15 :
+ case BRACKET_SYMBOLTYPE_NATNUM16 :
+ case BRACKET_SYMBOLTYPE_NATNUM17 :
+ case BRACKET_SYMBOLTYPE_NATNUM18 :
+ case BRACKET_SYMBOLTYPE_NATNUM19 :
+ if ( NumFor[nIndex].GetNatNum().IsSet() )
+ {
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ }
+ else
+ {
+ OUString sParams;
+ sal_Int32 nSpacePos = sStr.indexOf(' ');
+ if (nSpacePos >= 0)
+ {
+ sParams = o3tl::trim(sStr.subView(nSpacePos+1));
+ }
+ //! eSymbolType is negative
+ sal_uInt8 nNum = static_cast<sal_uInt8>(0 - (eSymbolType - BRACKET_SYMBOLTYPE_NATNUM0));
+ if (!sParams.isEmpty() && !NatNumTakesParameters(nNum))
+ {
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ break;
+ }
+ sStr = "NatNum" + OUString::number(nNum);
+ NumFor[nIndex].SetNatNumNum( nNum, false );
+ // NatNum12 supports arguments
+ if (nNum == 12)
+ {
+ if (sParams.isEmpty())
+ sParams = "cardinal"; // default NatNum12 format is "cardinal"
+ NumFor[nIndex].SetNatNumParams(sParams);
+ sStr += " " + sParams;
+ }
+ }
+ break;
+ case BRACKET_SYMBOLTYPE_DBNUM1 :
+ case BRACKET_SYMBOLTYPE_DBNUM2 :
+ case BRACKET_SYMBOLTYPE_DBNUM3 :
+ case BRACKET_SYMBOLTYPE_DBNUM4 :
+ case BRACKET_SYMBOLTYPE_DBNUM5 :
+ case BRACKET_SYMBOLTYPE_DBNUM6 :
+ case BRACKET_SYMBOLTYPE_DBNUM7 :
+ case BRACKET_SYMBOLTYPE_DBNUM8 :
+ case BRACKET_SYMBOLTYPE_DBNUM9 :
+ if ( NumFor[nIndex].GetNatNum().IsSet() )
+ {
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ }
+ else
+ {
+ //! eSymbolType is negative
+ sal_uInt8 nNum = static_cast<sal_uInt8>(1 - (eSymbolType - BRACKET_SYMBOLTYPE_DBNUM1));
+ sStr = "DBNum" + OUStringChar(sal_Unicode('0' + nNum));
+ NumFor[nIndex].SetNatNumNum( nNum, true );
+ }
+ break;
+ case BRACKET_SYMBOLTYPE_LOCALE :
+ if ( NumFor[nIndex].GetNatNum().GetLang() != LANGUAGE_DONTKNOW ||
+ sBuff[nPos-1] != ']' )
+ // Check also for ']' to avoid pulling in
+ // locale data for the preview string for not
+ // yet completed LCIDs in the dialog.
+ {
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ }
+ else
+ {
+ sal_Int32 nTmp = 2;
+ LocaleType aTmpLocale( ImpGetLocaleType( sStr, nTmp));
+ if (aTmpLocale.meLanguage == LANGUAGE_DONTKNOW)
+ {
+ bCancel = true; // break for
+ nCheckPos = nPosOld;
+ }
+ else
+ {
+ // Only the first sub format's locale will be
+ // used as the format's overall locale.
+ // Sorts this also under the corresponding
+ // locale for the dialog.
+ // If we don't support the locale this would
+ // result in an unknown (empty) language
+ // listbox entry and the user would never see
+ // this format.
+ if (nIndex == 0 && (aTmpLocale.meLanguage == LANGUAGE_SYSTEM ||
+ SvNumberFormatter::IsLocaleInstalled( aTmpLocale.meLanguage)))
+ {
+ maLocale = aTmpLocale;
+ eLan = aTmpLocale.meLanguage; // return to caller
+
+ // Set new target locale also at scanner.
+ // We have to do this because switching locale
+ // may make replacing keywords and separators
+ // necessary.
+ // We can do this because it's the first
+ // subformat and we're still at parsing the
+ // modifiers, not keywords.
+ rScan.SetNewLnge( eLan);
+ // We can not force conversion though because
+ // the caller may have explicitly not set it.
+ // In the usual case the target locale is the
+ // originating locale the conversion is not
+ // necessary, when reading alien documents
+ // conversion is enabled anyway.
+
+ /* TODO: fiddle with scanner to make this
+ * known? A change in the locale may affect
+ * separators and keywords. On the other
+ * hand they may have been entered as used
+ * in the originating locale, there's no
+ * way to predict other than analyzing the
+ * format code, we assume here the current
+ * context is used, which is most likely
+ * the case.
+ * */
+
+ // Strip a plain locale identifier if locale
+ // data is available to avoid duplicated
+ // formats with and without LCID for the same
+ // locale. Besides it looks ugly and confusing
+ // and is unnecessary as the format will be
+ // listed for the resulting locale.
+ if (aTmpLocale.isPlainLocale())
+ sStr.clear();
+ else
+ sStr = "$-" + aTmpLocale.generateCode();
+ }
+ else
+ {
+ if (nIndex == 0)
+ // Locale data not available, remember.
+ maLocale.meLanguageWithoutLocaleData = aTmpLocale.meLanguage;
+
+ sStr = "$-" + aTmpLocale.generateCode();
+ }
+ NumFor[nIndex].SetNatNumLang( MsLangId::getRealLanguage( aTmpLocale.meLanguage));
+
+ // "$-NNCCLLLL" Numerals and Calendar
+ if (sSymbol.getLength() > 6)
+ {
+ sInsertCalendar = ImpObtainCalendarAndNumerals( sBuff, nPos, eLan, aTmpLocale);
+ }
+ /* NOTE: there can be only one calendar
+ * inserted so the last one wins, though
+ * our own calendar modifiers support
+ * multiple calendars within one sub format
+ * code if at different positions. */
+ }
+ }
+ break;
+ }
+ if ( !bCancel )
+ {
+ if (sStr == sSymbol)
+ {
+ nPosOld = nPos;
+ }
+ else
+ {
+ sBuff.remove(nPosOld, nPos - nPosOld);
+ if (!sStr.isEmpty())
+ {
+ sBuff.insert(nPosOld, sStr);
+ nPos = nPosOld + sStr.getLength();
+ sBuff.insert(nPos, "]");
+ sBuff.insert(nPosOld, "[");
+ nPos += 2;
+ nPosOld = nPos; // position before string
+ }
+ else
+ {
+ nPos = nPosOld; // prefix removed for whatever reason
+ }
+ }
+ }
+ }
+ }
+ while ( !bCancel && lcl_SvNumberformat_IsBracketedPrefix( eSymbolType ) );
+
+ // The remaining format code string
+ if ( !bCancel )
+ {
+ if (eSymbolType == BRACKET_SYMBOLTYPE_FORMAT)
+ {
+ if (nIndex == 1 && eOp1 == NUMBERFORMAT_OP_NO)
+ {
+ eOp1 = NUMBERFORMAT_OP_GT; // undefined condition, default: > 0
+ }
+ else if (nIndex == 2 && eOp2 == NUMBERFORMAT_OP_NO)
+ {
+ eOp2 = NUMBERFORMAT_OP_LT; // undefined condition, default: < 0
+ }
+ if (sStr.isEmpty())
+ {
+ // Empty sub format.
+ NumFor[nIndex].Info().eScannedType = SvNumFormatType::EMPTY;
+ }
+ else
+ {
+ if (!sInsertCalendar.isEmpty())
+ {
+ sStr = sInsertCalendar + sStr;
+ }
+ sal_Int32 nStrPos = pSc->ScanFormat( sStr);
+ sal_uInt16 nCnt = pSc->GetResultStringsCnt();
+ if (nCnt == 0 && nStrPos == 0) // error
+ {
+ nStrPos = 1;
+ }
+ if (nStrPos == 0) // ok
+ {
+ // e.g. Thai T speciality
+ if (pSc->GetNatNumModifier() && !NumFor[nIndex].GetNatNum().IsSet())
+ {
+ sStr = "[NatNum" + OUString::number( pSc->GetNatNumModifier()) + "]" + sStr;
+ NumFor[nIndex].SetNatNumNum( pSc->GetNatNumModifier(), false );
+ }
+ // #i53826# #i42727# For the Thai T speciality we need
+ // to freeze the locale and immunize it against
+ // conversions during exports, just in case we want to
+ // save to Xcl. This disables the feature of being able
+ // to convert a NatNum to another locale. You can't
+ // have both.
+ // FIXME: implement a specialized export conversion
+ // that works on tokens (have to tokenize all first)
+ // and doesn't use the format string and
+ // PutandConvertEntry() to LANGUAGE_ENGLISH_US in
+ // sc/source/filter/excel/xestyle.cxx
+ // XclExpNumFmtBuffer::WriteFormatRecord().
+ LanguageType eLanguage;
+ if (NumFor[nIndex].GetNatNum().GetNatNum() == 1 &&
+ ((eLanguage = MsLangId::getRealLanguage( eLan)) == LANGUAGE_THAI) &&
+ NumFor[nIndex].GetNatNum().GetLang() == LANGUAGE_DONTKNOW)
+ {
+ sStr = "[$-" + OUString::number( sal_uInt16(eLanguage), 16 ).toAsciiUpperCase() + "]" + sStr;
+ NumFor[nIndex].SetNatNumLang( eLanguage);
+ }
+ sBuff.remove(nPosOld, nPos - nPosOld);
+ sBuff.insert(nPosOld, sStr);
+ nPos = nPosOld + sStr.getLength();
+ if (nPos < sBuff.getLength())
+ {
+ sBuff.insert(nPos, ";");
+ nPos++;
+ }
+ else if (nIndex > 0)
+ {
+ // The last subformat. If it is a trailing text
+ // format the omitted subformats act like they were
+ // not specified and "inherited" the first format,
+ // e.g. 0;@ behaves like 0;-0;0;@
+ if (pSc->GetScannedType() == SvNumFormatType::TEXT)
+ {
+ // Reset conditions, reverting any set above.
+ if (nIndex == 1)
+ eOp1 = NUMBERFORMAT_OP_NO;
+ else if (nIndex == 2)
+ eOp2 = NUMBERFORMAT_OP_NO;
+ nIndex = 3;
+ }
+ }
+ NumFor[nIndex].Enlarge(nCnt);
+ pSc->CopyInfo(&(NumFor[nIndex].Info()), nCnt);
+ // type check
+ if (nIndex == 0)
+ {
+ eType = NumFor[nIndex].Info().eScannedType;
+ }
+ else if (nIndex == 3)
+ { // #77026# Everything recognized IS text
+ NumFor[nIndex].Info().eScannedType = SvNumFormatType::TEXT;
+ }
+ else if ( NumFor[nIndex].Info().eScannedType != eType)
+ {
+ eType = SvNumFormatType::DEFINED;
+ }
+ }
+ else
+ {
+ nCheckPos = nPosOld + nStrPos; // error in string
+ bCancel = true; // break for
+ }
+ }
+ }
+ else if (eSymbolType == BRACKET_SYMBOLTYPE_ERROR) // error
+ {
+ nCheckPos = nPosOld;
+ bCancel = true;
+ }
+ else if ( lcl_SvNumberformat_IsBracketedPrefix( eSymbolType ) )
+ {
+ nCheckPos = nPosOld + 1; // error, prefix in string
+ bCancel = true; // break for
+ }
+ }
+ if ( bCancel && !nCheckPos )
+ {
+ nCheckPos = 1; // nCheckPos is used as an error condition
+ }
+ if ( !bCancel )
+ {
+ if ( NumFor[nIndex].GetNatNum().IsSet() &&
+ NumFor[nIndex].GetNatNum().GetLang() == LANGUAGE_DONTKNOW )
+ {
+ NumFor[nIndex].SetNatNumLang( eLan );
+ }
+ }
+ if (sBuff.getLength() == nPos)
+ {
+ if (nIndex < 3 && rString[rString.getLength()-1] == ';')
+ {
+ // A trailing ';' is significant and specifies the following
+ // subformat to be empty. We don't enter the scanning loop
+ // above again though.
+ // Note that the operators apply to the current last scanned
+ // subformat.
+ if (nIndex == 0 && eOp1 == NUMBERFORMAT_OP_NO)
+ {
+ eOp1 = NUMBERFORMAT_OP_GT; // undefined condition, default: > 0
+ }
+ else if (nIndex == 1 && eOp2 == NUMBERFORMAT_OP_NO)
+ {
+ eOp2 = NUMBERFORMAT_OP_LT; // undefined condition, default: < 0
+ }
+ NumFor[nIndex+1].Info().eScannedType = SvNumFormatType::EMPTY;
+ if (sBuff[nPos-1] != ';')
+ sBuff.insert( nPos++, ';');
+ }
+ if (nIndex == 2 && eSymbolType == BRACKET_SYMBOLTYPE_FORMAT && sBuff[nPos-1] == ';')
+ {
+ // #83510# A 4th subformat explicitly specified to be empty
+ // hides any text. Need the type here for HasTextFormat()
+ NumFor[3].Info().eScannedType = SvNumFormatType::TEXT;
+ }
+ bCancel = true;
+ }
+ if ( NumFor[nIndex].GetNatNum().IsSet() )
+ {
+ NumFor[nIndex].SetNatNumDate( bool(NumFor[nIndex].Info().eScannedType & SvNumFormatType::DATE) );
+ }
+ }
+
+ if (!nCheckPos && IsSubstituted())
+ {
+ // For to be substituted formats the scanned type must match the
+ // substitute type.
+ if (IsSystemTimeFormat())
+ {
+ if ((eType & ~SvNumFormatType::DEFINED) != SvNumFormatType::TIME)
+ nCheckPos = std::max<sal_Int32>( sBuff.indexOf(']') + 1, 1);
+ }
+ else if (IsSystemLongDateFormat())
+ {
+ if ((eType & ~SvNumFormatType::DEFINED) != SvNumFormatType::DATE)
+ nCheckPos = std::max<sal_Int32>( sBuff.indexOf(']') + 1, 1);
+ }
+ else
+ assert(!"unhandled substitute");
+ }
+
+ if ( bCondition && !nCheckPos )
+ {
+ if ( nIndex == 1 && NumFor[0].GetCount() == 0 &&
+ sBuff[sBuff.getLength() - 1] != ';' )
+ {
+ // No format code => GENERAL but not if specified empty
+ OUString aAdd( pSc->GetStandardName() );
+ if ( !pSc->ScanFormat( aAdd ) )
+ {
+ sal_uInt16 nCnt = pSc->GetResultStringsCnt();
+ if ( nCnt )
+ {
+ NumFor[0].Enlarge(nCnt);
+ pSc->CopyInfo( &(NumFor[0].Info()), nCnt );
+ sBuff.append(aAdd);
+ }
+ }
+ }
+ else if ( nIndex == 1 && NumFor[nIndex].GetCount() == 0 &&
+ sBuff[sBuff.getLength() - 1] != ';' &&
+ (NumFor[0].GetCount() > 1 ||
+ (NumFor[0].GetCount() == 1 &&
+ NumFor[0].Info().nTypeArray[0] != NF_KEY_GENERAL)) )
+ {
+ // No trailing second subformat => GENERAL but not if specified empty
+ // and not if first subformat is GENERAL
+ OUString aAdd( pSc->GetStandardName() );
+ if ( !pSc->ScanFormat( aAdd ) )
+ {
+ sal_uInt16 nCnt = pSc->GetResultStringsCnt();
+ if ( nCnt )
+ {
+ NumFor[nIndex].Enlarge(nCnt);
+ pSc->CopyInfo( &(NumFor[nIndex].Info()), nCnt );
+ sBuff.append(";");
+ sBuff.append(aAdd);
+ }
+ }
+ }
+ else if ( nIndex == 2 && NumFor[nIndex].GetCount() == 0 &&
+ sBuff[sBuff.getLength() - 1] != ';' &&
+ eOp2 != NUMBERFORMAT_OP_NO )
+ {
+ // No trailing third subformat => GENERAL but not if specified empty
+ OUString aAdd( pSc->GetStandardName() );
+ if ( !pSc->ScanFormat( aAdd ) )
+ {
+ sal_uInt16 nCnt = pSc->GetResultStringsCnt();
+ if ( nCnt )
+ {
+ NumFor[nIndex].Enlarge(nCnt);
+ pSc->CopyInfo( &(NumFor[nIndex].Info()), nCnt );
+ sBuff.append(";");
+ sBuff.append(aAdd);
+ }
+ }
+ }
+ }
+ rString = sBuff.makeStringAndClear();
+ sFormatstring = rString;
+
+ if (NumFor[2].GetCount() == 0 && // No third partial string
+ eOp1 == NUMBERFORMAT_OP_GT && eOp2 == NUMBERFORMAT_OP_NO &&
+ fLimit1 == 0.0 && fLimit2 == 0.0)
+ {
+ eOp1 = NUMBERFORMAT_OP_GE; // Add 0 to the first format
+ }
+
+}
+
+SvNumberformat::~SvNumberformat()
+{
+}
+
+/**
+ * Next_Symbol
+ *
+ * Splits up the symbols for further processing (by the Turing machine)
+ *
+ * Start state = SsStart, * = Special state
+ * ---------------+-------------------+----------------------------+---------------
+ * Old State | Symbol read | Event | New state
+ * ---------------+-------------------+----------------------------+---------------
+ * SsStart | " | Symbol += Character | SsGetQuoted
+ * | ; | Pos-- | SsGetString
+ * | [ | Symbol += Character | SsGetBracketed
+ * | ] | Error | SsStop
+ * | BLANK | |
+ * | Else | Symbol += Character | SsGetString
+ * ---------------+-------------------+----------------------------+---------------
+ * SsGetString | " | Symbol += Character | SsGetQuoted
+ * | ; | | SsStop
+ * | Else | Symbol += Character |
+ * ---------------+-------------------+----------------------------+---------------
+ * SsGetQuoted | " | Symbol += Character | SsGetString
+ * | Else | Symbol += Character |
+ * ---------------+-------------------+----------------------------+---------------
+ * SsGetBracketed | <, > = | del [ |
+ * | | Symbol += Character | SsGetCon
+ * | BLANK | |
+ * | h, H, m, M, s, S | Symbol += Character | SsGetTime
+ * | Else | del [ |
+ * | | Symbol += Character | SsGetPrefix
+ * ---------------+-------------------+----------------------------+---------------
+ * SsGetTime | ] | Symbol += Character | SsGetString
+ * | h, H, m, M, s, S | Symbol += Character, * | SsGetString
+ * | Else | del [; Symbol += Character | SsGetPrefix
+ * ---------------+-------------------+----------------------------+---------------
+ * SsGetPrefix | ] | | SsStop
+ * | Else | Symbol += Character |
+ * ---------------+-------------------+----------------------------+---------------
+ * SsGetCon | >, = | Symbol += Character |
+ * | ] | | SsStop
+ * | Else | Error | SsStop
+ * ---------------+-------------------+----------------------------+---------------
+ */
+
+namespace {
+
+enum ScanState
+{
+ SsStop,
+ SsStart,
+ SsGetCon, // condition
+ SsGetString, // format string
+ SsGetPrefix, // color or NatNumN
+ SsGetTime, // [HH] for time
+ SsGetBracketed, // any [...] not decided yet
+ SsGetQuoted // quoted text
+};
+
+}
+
+// read a string until ']' and delete spaces in input
+// static
+sal_Int32 SvNumberformat::ImpGetNumber(OUStringBuffer& rString,
+ sal_Int32& nPos,
+ OUString& sSymbol)
+{
+ sal_Int32 nStartPos = nPos;
+ sal_Unicode cToken;
+ sal_Int32 nLen = rString.getLength();
+ OUStringBuffer sBuffSymbol;
+ while ( nPos < nLen )
+ {
+ cToken = rString[nPos];
+ if (cToken == ']')
+ break;
+ if (cToken == ' ')
+ { // delete spaces
+ rString.remove(nPos,1);
+ nLen--;
+ }
+ else
+ {
+ nPos++;
+ sBuffSymbol.append(cToken);
+ }
+ }
+ sSymbol = sBuffSymbol.makeStringAndClear();
+ return nPos - nStartPos;
+}
+
+namespace {
+
+sal_Unicode toUniChar(sal_uInt8 n)
+{
+ char c;
+ if (n < 10)
+ {
+ c = '0' + n;
+ }
+ else
+ {
+ c = 'A' + n - 10;
+ }
+ return sal_Unicode(c);
+}
+
+bool IsCombiningSymbol( OUStringBuffer& rStringBuffer, sal_Int32 nPos )
+{
+ bool bRet = false;
+ while (nPos >= 0)
+ {
+ switch (rStringBuffer[nPos])
+ {
+ case '*':
+ case '\\':
+ case '_':
+ bRet = !bRet;
+ --nPos;
+ break;
+ default:
+ return bRet;
+ }
+ }
+ return bRet;
+}
+
+} // namespace
+
+OUString SvNumberformat::LocaleType::generateCode() const
+{
+ OUStringBuffer aBuf;
+#if 0
+ // TODO: We may re-enable this later. Don't remove it! --Kohei
+ if (mnNumeralShape)
+ {
+ sal_uInt8 nVal = mnNumeralShape;
+ for (sal_uInt8 i = 0; i < 2; ++i)
+ {
+ sal_uInt8 n = (nVal & 0xF0) >> 4;
+ if (n || aBuf.getLength())
+ {
+ aBuf.append(toUniChar(n));
+ }
+ nVal = nVal << 4;
+ }
+ }
+
+ if (mnNumeralShape || mnCalendarType)
+ {
+ sal_uInt8 nVal = mnCalendarType;
+ for (sal_uInt8 i = 0; i < 2; ++i)
+ {
+ sal_uInt8 n = (nVal & 0xF0) >> 4;
+ if (n || aBuf.getLength())
+ {
+ aBuf.append(toUniChar(n));
+ }
+ nVal = nVal << 4;
+ }
+ }
+#endif
+
+ sal_uInt16 n16 = static_cast<sal_uInt16>(
+ (meLanguageWithoutLocaleData == LANGUAGE_DONTKNOW) ? meLanguage :
+ meLanguageWithoutLocaleData);
+ if (meLanguage == LANGUAGE_SYSTEM)
+ {
+ switch (meSubstitute)
+ {
+ case Substitute::NONE:
+ ; // nothing
+ break;
+ case Substitute::TIME:
+ n16 = static_cast<sal_uInt16>(LANGUAGE_NF_SYSTEM_TIME);
+ break;
+ case Substitute::LONGDATE:
+ n16 = static_cast<sal_uInt16>(LANGUAGE_NF_SYSTEM_DATE);
+ break;
+ }
+ }
+ for (sal_uInt8 i = 0; i < 4; ++i)
+ {
+ sal_uInt8 n = static_cast<sal_uInt8>((n16 & 0xF000) >> 12);
+ // Omit leading zeros for consistency.
+ if (n || !aBuf.isEmpty() || i == 3)
+ {
+ aBuf.append(toUniChar(n));
+ }
+ n16 = n16 << 4;
+ }
+
+ return aBuf.makeStringAndClear();
+}
+
+SvNumberformat::LocaleType::LocaleType()
+ : meLanguage(LANGUAGE_DONTKNOW)
+ , meLanguageWithoutLocaleData(LANGUAGE_DONTKNOW)
+ , meSubstitute(Substitute::NONE)
+ , mnNumeralShape(0)
+ , mnCalendarType(0)
+{
+}
+
+SvNumberformat::LocaleType::LocaleType(sal_uInt32 nRawNum)
+ : meLanguage(LANGUAGE_DONTKNOW)
+ , meLanguageWithoutLocaleData(LANGUAGE_DONTKNOW)
+ , meSubstitute(Substitute::NONE)
+ , mnNumeralShape(0)
+ , mnCalendarType(0)
+{
+ meLanguage = static_cast<LanguageType>(nRawNum & 0x0000FFFF);
+ if (meLanguage == LANGUAGE_NF_SYSTEM_TIME)
+ {
+ meSubstitute = Substitute::TIME;
+ meLanguage = LANGUAGE_SYSTEM;
+ }
+ else if (meLanguage == LANGUAGE_NF_SYSTEM_DATE)
+ {
+ meSubstitute = Substitute::LONGDATE;
+ meLanguage = LANGUAGE_SYSTEM;
+ }
+ nRawNum = (nRawNum >> 16);
+ mnCalendarType = static_cast<sal_uInt8>(nRawNum & 0xFF);
+ nRawNum = (nRawNum >> 8);
+ mnNumeralShape = static_cast<sal_uInt8>(nRawNum & 0xFF);
+}
+
+bool SvNumberformat::LocaleType::isPlainLocale() const
+{
+ return meSubstitute == Substitute::NONE && !mnCalendarType && !mnNumeralShape;
+}
+
+// static
+SvNumberformat::LocaleType SvNumberformat::ImpGetLocaleType(const OUString& rString, sal_Int32& nPos )
+{
+ sal_uInt32 nNum = 0;
+ sal_Unicode cToken = 0;
+ sal_Int32 nStart = nPos;
+ sal_Int32 nLen = rString.getLength();
+ while ( nPos < nLen && (nPos - nStart < 8) )
+ {
+ cToken = rString[nPos];
+ if (cToken == ']')
+ break;
+ if ( '0' <= cToken && cToken <= '9' )
+ {
+ nNum *= 16;
+ nNum += cToken - '0';
+ }
+ else if ( 'a' <= cToken && cToken <= 'f' )
+ {
+ nNum *= 16;
+ nNum += cToken - 'a' + 10;
+ }
+ else if ( 'A' <= cToken && cToken <= 'F' )
+ {
+ nNum *= 16;
+ nNum += cToken - 'A' + 10;
+ }
+ else
+ {
+ return LocaleType(); // LANGUAGE_DONTKNOW;
+ }
+ ++nPos;
+ }
+
+ return (cToken == ']' || nPos == nLen) ? LocaleType(nNum) : LocaleType();
+}
+
+static bool lcl_matchKeywordAndGetNumber( const OUString & rString, const sal_Int32 nPos,
+ const OUString & rKeyword, sal_Int32 & nNumber )
+{
+ if (0 <= nPos && nPos + rKeyword.getLength() < rString.getLength() && rString.matchIgnoreAsciiCase( rKeyword, nPos))
+ {
+ nNumber = o3tl::toInt32(rString.subView( nPos + rKeyword.getLength()));
+ return true;
+ }
+ else
+ {
+ nNumber = 0;
+ return false;
+ }
+}
+
+short SvNumberformat::ImpNextSymbol(OUStringBuffer& rString,
+ sal_Int32& nPos,
+ OUString& sSymbol) const
+{
+ short eSymbolType = BRACKET_SYMBOLTYPE_FORMAT;
+ sal_Unicode cToken;
+ sal_Unicode cLetter = ' '; // Preliminary result
+ sal_Int32 nLen = rString.getLength();
+ ScanState eState = SsStart;
+ OUStringBuffer sBuffSymbol(128);
+
+ const NfKeywordTable & rKeywords = rScan.GetKeywords();
+ while (nPos < nLen && eState != SsStop)
+ {
+ cToken = rString[nPos];
+ nPos++;
+ switch (eState)
+ {
+ case SsStart:
+ if (cToken == '\"')
+ {
+ eState = SsGetQuoted;
+ sBuffSymbol.append(cToken);
+ }
+ else if (cToken == '[')
+ {
+ eState = SsGetBracketed;
+ sBuffSymbol.append(cToken);
+ }
+ else if (cToken == ';')
+ {
+ eState = SsGetString;
+ nPos--;
+ eSymbolType = BRACKET_SYMBOLTYPE_FORMAT;
+ }
+ else if (cToken == ']')
+ {
+ eState = SsStop;
+ eSymbolType = BRACKET_SYMBOLTYPE_ERROR;
+ }
+ else if (cToken == ' ') // Skip Blanks
+ {
+ nPos--;
+ rString.remove(nPos, 1);
+ nLen--;
+ }
+ else
+ {
+ sBuffSymbol.append(cToken);
+ eState = SsGetString;
+ eSymbolType = BRACKET_SYMBOLTYPE_FORMAT;
+ }
+ break;
+ case SsGetBracketed:
+ switch (cToken)
+ {
+ case '<':
+ case '>':
+ case '=':
+ sBuffSymbol.stripStart('[');
+ sBuffSymbol.append(cToken);
+ cLetter = cToken;
+ eState = SsGetCon;
+ switch (cToken)
+ {
+ case '<':
+ eSymbolType = NUMBERFORMAT_OP_LT;
+ break;
+ case '>':
+ eSymbolType = NUMBERFORMAT_OP_GT;
+ break;
+ case '=':
+ eSymbolType = NUMBERFORMAT_OP_EQ;
+ break;
+ }
+ break;
+ case ' ':
+ nPos--;
+ rString.remove(nPos, 1);
+ nLen--;
+ break;
+ case '$' :
+ if ( nPos < nLen && rString[nPos] == '-' )
+ {
+ // [$-xxx] locale
+ sBuffSymbol.stripStart('[');
+ eSymbolType = BRACKET_SYMBOLTYPE_LOCALE;
+ eState = SsGetPrefix;
+ }
+ else
+ { // currency
+ eSymbolType = BRACKET_SYMBOLTYPE_FORMAT;
+ eState = SsGetString;
+ }
+ sBuffSymbol.append(cToken);
+ break;
+ case '~' :
+ // calendarID
+ eSymbolType = BRACKET_SYMBOLTYPE_FORMAT;
+ sBuffSymbol.append(cToken);
+ eState = SsGetString;
+ break;
+ default:
+ {
+ static const OUStringLiteral aNatNum(u"NATNUM");
+ static const OUStringLiteral aDBNum(u"DBNUM");
+ const OUString aBufStr( rString.toString());
+ sal_Int32 nNatNumNum;
+ sal_Int32 nDBNum;
+ if ( lcl_matchKeywordAndGetNumber( aBufStr, nPos-1, aNatNum, nNatNumNum) &&
+ 0 <= nNatNumNum && nNatNumNum <= 19 )
+ {
+ sBuffSymbol.stripStart('[');
+ sBuffSymbol.append( aBufStr.subView(--nPos, aNatNum.getLength()+1) );
+ nPos += aNatNum.getLength()+1;
+ //! SymbolType is negative
+ eSymbolType = static_cast<short>(BRACKET_SYMBOLTYPE_NATNUM0 - nNatNumNum);
+ eState = SsGetPrefix;
+ }
+ else if ( lcl_matchKeywordAndGetNumber( aBufStr, nPos-1, aDBNum, nDBNum) &&
+ 1 <= nDBNum && nDBNum <= 9 )
+ {
+ sBuffSymbol.stripStart('[');
+ sBuffSymbol.append( aBufStr.subView(--nPos, aDBNum.getLength()+1) );
+ nPos += aDBNum.getLength()+1;
+ //! SymbolType is negative
+ eSymbolType = sal::static_int_cast< short >( BRACKET_SYMBOLTYPE_DBNUM1 - (nDBNum - 1) );
+ eState = SsGetPrefix;
+ }
+ else
+ {
+ sal_Unicode cUpper = rChrCls().uppercase( aBufStr, nPos-1, 1)[0];
+ if ( cUpper == rKeywords[NF_KEY_H][0] || // H
+ cUpper == rKeywords[NF_KEY_MI][0] || // M
+ cUpper == rKeywords[NF_KEY_S][0] ) // S
+ {
+ sBuffSymbol.append(cToken);
+ eState = SsGetTime;
+ cLetter = cToken;
+ }
+ else
+ {
+ sBuffSymbol.stripStart('[');
+ sBuffSymbol.append(cToken);
+ eSymbolType = BRACKET_SYMBOLTYPE_COLOR;
+ eState = SsGetPrefix;
+ }
+ }
+ }
+ }
+ break;
+ case SsGetString:
+ if (cToken == '\"')
+ {
+ eState = SsGetQuoted;
+ sBuffSymbol.append(cToken);
+ }
+ else if (cToken == ';' && (nPos < 2 || !IsCombiningSymbol( rString, nPos-2)))
+ {
+ eState = SsStop;
+ }
+ else
+ {
+ sBuffSymbol.append(cToken);
+ }
+ break;
+ case SsGetQuoted:
+ if (cToken == '\"')
+ {
+ eState = SsGetString;
+ sBuffSymbol.append(cToken);
+ }
+ else
+ {
+ sBuffSymbol.append(cToken);
+ }
+ break;
+ case SsGetTime:
+ if (cToken == ']')
+ {
+ sBuffSymbol.append(cToken);
+ eState = SsGetString;
+ eSymbolType = BRACKET_SYMBOLTYPE_FORMAT;
+ }
+ else
+ {
+ sal_Unicode cUpper = rChrCls().uppercase(rString.toString(), nPos-1, 1)[0];
+ if (cUpper == rKeywords[NF_KEY_H][0] || // H
+ cUpper == rKeywords[NF_KEY_MI][0] || // M
+ cUpper == rKeywords[NF_KEY_S][0] ) // S
+ {
+ if (cLetter == cToken)
+ {
+ sBuffSymbol.append(cToken);
+ cLetter = ' ';
+ }
+ else
+ {
+ sBuffSymbol.stripStart('[');
+ sBuffSymbol.append(cToken);
+ eState = SsGetPrefix;
+ }
+ }
+ else
+ {
+ sBuffSymbol.stripStart('[');
+ sBuffSymbol.append(cToken);
+ eSymbolType = BRACKET_SYMBOLTYPE_COLOR;
+ eState = SsGetPrefix;
+ }
+ }
+ break;
+ case SsGetCon:
+ switch (cToken)
+ {
+ case '<':
+ eState = SsStop;
+ eSymbolType = BRACKET_SYMBOLTYPE_ERROR;
+ break;
+ case '>':
+ if (cLetter == '<')
+ {
+ sBuffSymbol.append(cToken);
+ cLetter = ' ';
+ eState = SsStop;
+ eSymbolType = NUMBERFORMAT_OP_NE;
+ }
+ else
+ {
+ eState = SsStop;
+ eSymbolType = BRACKET_SYMBOLTYPE_ERROR;
+ }
+ break;
+ case '=':
+ if (cLetter == '<')
+ {
+ sBuffSymbol.append(cToken);
+ cLetter = ' ';
+ eSymbolType = NUMBERFORMAT_OP_LE;
+ }
+ else if (cLetter == '>')
+ {
+ sBuffSymbol.append(cToken);
+ cLetter = ' ';
+ eSymbolType = NUMBERFORMAT_OP_GE;
+ }
+ else
+ {
+ eState = SsStop;
+ eSymbolType = BRACKET_SYMBOLTYPE_ERROR;
+ }
+ break;
+ case ' ':
+ nPos--;
+ rString.remove(nPos,1);
+ nLen--;
+ break;
+ default:
+ eState = SsStop;
+ nPos--;
+ break;
+ }
+ break;
+ case SsGetPrefix:
+ if (cToken == ']')
+ {
+ eState = SsStop;
+ }
+ else
+ {
+ sBuffSymbol.append(cToken);
+ }
+ break;
+ default:
+ break;
+ } // of switch
+ } // of while
+ sSymbol = sBuffSymbol.makeStringAndClear();
+ return eSymbolType;
+}
+
+void SvNumberformat::ConvertLanguage( SvNumberFormatter& rConverter,
+ LanguageType eConvertFrom,
+ LanguageType eConvertTo )
+{
+ sal_Int32 nCheckPos;
+ sal_uInt32 nKey;
+ SvNumFormatType nType = eType;
+ OUString aFormatString( sFormatstring );
+ rConverter.PutandConvertEntry( aFormatString, nCheckPos, nType,
+ nKey, eConvertFrom, eConvertTo, false);
+ const SvNumberformat* pFormat = rConverter.GetEntry( nKey );
+ DBG_ASSERT( pFormat, "SvNumberformat::ConvertLanguage: Conversion without format" );
+ if ( pFormat )
+ {
+ ImpCopyNumberformat( *pFormat );
+ // Reset values taken over from Formatter/Scanner
+ // pColor still points to table in temporary Formatter/Scanner
+ for (ImpSvNumFor & rFormatter : NumFor)
+ {
+ OUString aColorName = rFormatter.GetColorName();
+ const Color* pColor = rScan.GetColor( aColorName );
+ rFormatter.SetColor( pColor, aColorName );
+ }
+ }
+}
+
+bool SvNumberformat::HasNewCurrency() const
+{
+ for (const auto & j : NumFor)
+ {
+ if ( j.HasNewCurrency() )
+ {
+ return true;
+ }
+ }
+ return false;
+}
+
+bool SvNumberformat::GetNewCurrencySymbol( OUString& rSymbol,
+ OUString& rExtension ) const
+{
+ for (const auto & j : NumFor)
+ {
+ if ( j.GetNewCurrencySymbol( rSymbol, rExtension ) )
+ {
+ return true;
+ }
+ }
+ rSymbol.clear();
+ rExtension.clear();
+ return false;
+}
+
+// static
+OUString SvNumberformat::StripNewCurrencyDelimiters( const OUString& rStr )
+{
+ OUStringBuffer aTmp(rStr.getLength());
+ sal_Int32 nStartPos, nPos, nLen;
+ nLen = rStr.getLength();
+ nStartPos = 0;
+ while ( (nPos = rStr.indexOf( "[$", nStartPos )) >= 0 )
+ {
+ sal_Int32 nEnd;
+ if ( (nEnd = GetQuoteEnd( rStr, nPos )) >= 0 )
+ {
+ aTmp.append(rStr.subView( nStartPos, ++nEnd - nStartPos ));
+ nStartPos = nEnd;
+ }
+ else
+ {
+ aTmp.append(rStr.subView(nStartPos, nPos - nStartPos) );
+ nStartPos = nPos + 2;
+ sal_Int32 nDash;
+ nEnd = nStartPos - 1;
+ do
+ {
+ nDash = rStr.indexOf( '-', ++nEnd );
+ nEnd = GetQuoteEnd( rStr, nDash );
+ }
+ while ( nEnd >= 0 );
+ sal_Int32 nClose;
+ nEnd = nStartPos - 1;
+ do
+ {
+ nClose = rStr.indexOf( ']', ++nEnd );
+ nEnd = GetQuoteEnd( rStr, nClose );
+ }
+ while ( nEnd >= 0 );
+
+ if(nClose < 0)
+ {
+ /* there should always be a closing ]
+ * but the old String class would have hidden
+ * that. so be conservative too
+ */
+ nClose = nLen;
+ }
+
+ nPos = nClose;
+ if(nDash >= 0 && nDash < nClose)
+ {
+ nPos = nDash;
+ }
+ aTmp.append(rStr.subView(nStartPos, nPos - nStartPos) );
+ nStartPos = nClose + 1;
+ }
+ }
+ if ( nLen > nStartPos )
+ {
+ aTmp.append(rStr.subView(nStartPos, nLen - nStartPos) );
+ }
+ return aTmp.makeStringAndClear();
+}
+
+void SvNumberformat::ImpGetOutputStandard(double& fNumber, OUStringBuffer& rOutString) const
+{
+ OUString sTemp;
+ ImpGetOutputStandard(fNumber, sTemp);
+ rOutString = sTemp;
+}
+
+void SvNumberformat::ImpGetOutputStandard(double& fNumber, OUString& rOutString) const
+{
+ sal_uInt16 nStandardPrec = rScan.GetStandardPrec();
+
+ if ( fabs(fNumber) > EXP_ABS_UPPER_BOUND )
+ {
+ nStandardPrec = ::std::min(nStandardPrec, static_cast<sal_uInt16>(14)); // limits to 14 decimals
+ rOutString = ::rtl::math::doubleToUString( fNumber,
+ rtl_math_StringFormat_E2, nStandardPrec /*2*/,
+ GetFormatter().GetNumDecimalSep()[0]);
+ }
+ else
+ {
+ ImpGetOutputStdToPrecision(fNumber, rOutString, nStandardPrec);
+ }
+}
+
+namespace
+{
+
+template<typename T>
+bool checkForAll0s(const T& rString, sal_Int32 nIdx=0)
+{
+ if (nIdx>=rString.getLength())
+ return false;
+
+ do
+ {
+ if (rString[nIdx]!='0')
+ return false;
+ }
+ while (++nIdx<rString.getLength());
+
+ return true;
+}
+
+}
+
+void SvNumberformat::ImpGetOutputStdToPrecision(double& rNumber, OUString& rOutString, sal_uInt16 nPrecision) const
+{
+ // Make sure the precision doesn't go over the maximum allowable precision.
+ nPrecision = ::std::min(UPPER_PRECISION, nPrecision);
+
+ // We decided to strip trailing zeros unconditionally, since binary
+ // double-precision rounding error makes it impossible to determine e.g.
+ // whether 844.10000000000002273737 is what the user has typed, or the
+ // user has typed 844.1 but IEEE 754 represents it that way internally.
+
+ rOutString = ::rtl::math::doubleToUString( rNumber,
+ rtl_math_StringFormat_F, nPrecision /*2*/,
+ GetFormatter().GetNumDecimalSep()[0], true );
+ if (rOutString[0] == '-' && checkForAll0s(rOutString, 1))
+ {
+ rOutString = comphelper::string::stripStart(rOutString, '-'); // not -0
+ }
+ rOutString = impTransliterate(rOutString, NumFor[0].GetNatNum());
+}
+
+void SvNumberformat::ImpGetOutputInputLine(double fNumber, OUString& OutString) const
+{
+ bool bModified = false;
+ if ( (eType & SvNumFormatType::PERCENT) && (fabs(fNumber) < D_MAX_D_BY_100))
+ {
+ if (fNumber == 0.0)
+ {
+ OutString = "0%";
+ return;
+ }
+ fNumber *= 100;
+ bModified = true;
+ }
+
+ if (fNumber == 0.0)
+ {
+ OutString = "0";
+ return;
+ }
+
+ OutString = ::rtl::math::doubleToUString( fNumber,
+ rtl_math_StringFormat_Automatic,
+ rtl_math_DecimalPlaces_Max,
+ GetFormatter().GetNumDecimalSep()[0], true );
+
+ if ( eType & SvNumFormatType::PERCENT && bModified)
+ {
+ OutString += "%";
+ }
+}
+
+short SvNumberformat::ImpCheckCondition(double fNumber,
+ double fLimit,
+ SvNumberformatLimitOps eOp)
+{
+ switch(eOp)
+ {
+ case NUMBERFORMAT_OP_NO:
+ return -1;
+ case NUMBERFORMAT_OP_EQ:
+ return static_cast<short>(fNumber == fLimit);
+ case NUMBERFORMAT_OP_NE:
+ return static_cast<short>(fNumber != fLimit);
+ case NUMBERFORMAT_OP_LT:
+ return static_cast<short>(fNumber < fLimit);
+ case NUMBERFORMAT_OP_LE:
+ return static_cast<short>(fNumber <= fLimit);
+ case NUMBERFORMAT_OP_GT:
+ return static_cast<short>(fNumber > fLimit);
+ case NUMBERFORMAT_OP_GE:
+ return static_cast<short>(fNumber >= fLimit);
+ default:
+ return -1;
+ }
+}
+
+static bool lcl_appendStarFillChar( OUStringBuffer& rBuf, const OUString& rStr )
+{
+ // Right during user input the star symbol is the very
+ // last character before the user enters another one.
+ if (rStr.getLength() > 1)
+ {
+ rBuf.append(u'\x001B');
+ rBuf.append(rStr[1]);
+ return true;
+ }
+ return false;
+}
+
+static bool lcl_insertStarFillChar( OUStringBuffer& rBuf, sal_Int32 nPos, const OUString& rStr )
+{
+ if (rStr.getLength() > 1)
+ {
+ rBuf.insert( nPos, rStr[1]);
+ rBuf.insert( nPos, u'\x001B');
+ return true;
+ }
+ return false;
+}
+
+void SvNumberformat::GetOutputString(std::u16string_view sString,
+ OUString& OutString,
+ const Color** ppColor)
+{
+ OUStringBuffer sOutBuff;
+ sal_uInt16 nIx;
+ if (eType & SvNumFormatType::TEXT)
+ {
+ nIx = 0;
+ }
+ else if (NumFor[3].GetCount() > 0)
+ {
+ nIx = 3;
+ }
+ else
+ {
+ *ppColor = nullptr; // no change of color
+ return;
+ }
+ *ppColor = NumFor[nIx].GetColor();
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ if (rInfo.eScannedType == SvNumFormatType::TEXT)
+ {
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+ for (sal_uInt16 i = 0; i < nCnt; i++)
+ {
+ switch (rInfo.nTypeArray[i])
+ {
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ lcl_appendStarFillChar( sOutBuff, rInfo.sStrArray[i]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[i].getLength() >= 2)
+ InsertBlanks( sOutBuff, sOutBuff.getLength(), rInfo.sStrArray[i][1] );
+ break;
+ case NF_KEY_GENERAL : // #77026# "General" is the same as "@"
+ case NF_SYMBOLTYPE_DEL :
+ sOutBuff.append(sString);
+ break;
+ default:
+ sOutBuff.append(rInfo.sStrArray[i]);
+ }
+ }
+ }
+ OutString = sOutBuff.makeStringAndClear();
+}
+
+namespace {
+
+void lcl_GetOutputStringScientific(double fNumber, sal_uInt16 nCharCount,
+ const SvNumberFormatter& rFormatter, OUString& rOutString)
+{
+ bool bSign = std::signbit(fNumber);
+
+ // 1.000E+015 (one digit and the decimal point, and the two chars +
+ // nExpDigit for the exponential part, totalling 6 or 7).
+ double fExp = log10( fabs(fNumber) );
+ if( fExp < 0.0 )
+ fExp = 1.0 - fExp;
+ sal_uInt16 nCharFormat = 6 + (fExp >= 100.0 ? 1 : 0);
+ sal_uInt16 nPrec = nCharCount > nCharFormat ? nCharCount - nCharFormat : 0;
+ if (nPrec && bSign)
+ {
+ // Make room for the negative sign.
+ --nPrec;
+ }
+ nPrec = ::std::min(nPrec, static_cast<sal_uInt16>(14)); // limit to 14 decimals.
+
+ rOutString = ::rtl::math::doubleToUString(fNumber, rtl_math_StringFormat_E2,
+ nPrec, rFormatter.GetNumDecimalSep()[0], true );
+}
+
+OUString lcl_GetPercentString(const ImpSvNumberformatInfo &rInfo, sal_uInt16 nCnt)
+{
+ sal_Int32 i;
+ OUStringBuffer aPercentString;
+ for( i = 0; i < nCnt; i++ )
+ {
+ if( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_PERCENT )
+ {
+ aPercentString.append( rInfo.sStrArray[i] );
+ bool bStringFound = false;
+ for( i--; i >= 0 && rInfo.nTypeArray[i] == NF_SYMBOLTYPE_STRING ; i-- )
+ {
+ if( !bStringFound )
+ {
+ bStringFound = true;
+ aPercentString.insert( 0, "\"" );
+ }
+ aPercentString.insert( 0, rInfo.sStrArray[i] );
+ }
+ i = nCnt;
+ if( bStringFound )
+ aPercentString.insert( 0, "\"" );
+ }
+ }
+ return aPercentString.makeStringAndClear();
+}
+
+OUString lcl_GetDenominatorString(const ImpSvNumberformatInfo &rInfo, sal_uInt16 nCnt)
+{
+ sal_Int32 i;
+ OUStringBuffer aDenominatorString;
+ for( i = 0; i < nCnt; i++ )
+ {
+ if( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_FRAC )
+ {
+ while ( ( ++i < nCnt ) && rInfo.nTypeArray[i] != NF_SYMBOLTYPE_FRAC_FDIV
+ && rInfo.nTypeArray[i] != NF_SYMBOLTYPE_DIGIT );
+ for( ; i < nCnt; i++ )
+ {
+ if( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_FRAC_FDIV || rInfo.nTypeArray[i] == NF_SYMBOLTYPE_DIGIT )
+ aDenominatorString.append( rInfo.sStrArray[i] );
+ else
+ i = nCnt;
+ }
+ }
+ }
+ return aDenominatorString.makeStringAndClear();
+}
+
+OUString lcl_GetNumeratorString(const ImpSvNumberformatInfo &rInfo, sal_uInt16 nCnt)
+{
+ sal_Int32 i;
+ OUStringBuffer aNumeratorString;
+ for( i = 0; i < nCnt; i++ )
+ {
+ if( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_FRAC )
+ {
+ for( i--; i >= 0 && rInfo.nTypeArray[i] == NF_SYMBOLTYPE_DIGIT ; i-- )
+ {
+ aNumeratorString.insert( 0, rInfo.sStrArray[i] );
+ }
+ i = nCnt;
+ }
+ }
+ return aNumeratorString.makeStringAndClear();
+}
+
+OUString lcl_GetFractionIntegerString(const ImpSvNumberformatInfo &rInfo, sal_uInt16 nCnt)
+{
+ sal_Int32 i;
+ OUStringBuffer aIntegerString;
+ for( i = 0; i < nCnt; i++ )
+ {
+ if( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_FRACBLANK )
+ {
+ for( i--; i >= 0 && ( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_DIGIT
+ || rInfo.nTypeArray[i] == NF_SYMBOLTYPE_THSEP ); i-- )
+ {
+ aIntegerString.insert( 0, rInfo.sStrArray[i] );
+ }
+ i = nCnt;
+ }
+ }
+ return aIntegerString.makeStringAndClear();
+}
+
+OUString lcl_GetIntegerFractionDelimiterString(const ImpSvNumberformatInfo &rInfo, sal_uInt16 nCnt)
+{
+ sal_uInt16 i;
+ for( i = 0; i < nCnt; i++ )
+ {
+ if( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_FRACBLANK )
+ {
+ return rInfo.sStrArray[i];
+ }
+ }
+ return OUString();
+}
+
+}
+
+OUString SvNumberformat::GetPercentString( sal_uInt16 nNumFor ) const
+{
+ const ImpSvNumberformatInfo& rInfo = NumFor[nNumFor].Info();
+ sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ return lcl_GetPercentString( rInfo, nCnt );
+}
+
+OUString SvNumberformat::GetDenominatorString( sal_uInt16 nNumFor ) const
+{
+ const ImpSvNumberformatInfo& rInfo = NumFor[nNumFor].Info();
+ sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ return lcl_GetDenominatorString( rInfo, nCnt );
+}
+
+OUString SvNumberformat::GetNumeratorString( sal_uInt16 nNumFor ) const
+{
+ const ImpSvNumberformatInfo& rInfo = NumFor[nNumFor].Info();
+ sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ return lcl_GetNumeratorString( rInfo, nCnt );
+}
+
+OUString SvNumberformat::GetIntegerFractionDelimiterString( sal_uInt16 nNumFor ) const
+{
+ const ImpSvNumberformatInfo& rInfo = NumFor[nNumFor].Info();
+ sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ return lcl_GetIntegerFractionDelimiterString( rInfo, nCnt );
+}
+
+bool SvNumberformat::GetOutputString(double fNumber, sal_uInt16 nCharCount, OUString& rOutString) const
+{
+ if (eType != SvNumFormatType::NUMBER)
+ {
+ return false;
+ }
+ double fTestNum = fNumber;
+ bool bSign = std::signbit(fTestNum);
+ if (bSign)
+ {
+ fTestNum = -fTestNum;
+ }
+ if (fTestNum < EXP_LOWER_BOUND)
+ {
+ lcl_GetOutputStringScientific(fNumber, nCharCount, GetFormatter(), rOutString);
+ return true;
+ }
+
+ double fExp = log10(fTestNum);
+ // Values < 1.0 always have one digit before the decimal point.
+ sal_uInt16 nDigitPre = fExp >= 0.0 ? static_cast<sal_uInt16>(ceil(fExp)) : 1;
+
+ if (nDigitPre > 15)
+ {
+ lcl_GetOutputStringScientific(fNumber, nCharCount, GetFormatter(), rOutString);
+ return true;
+ }
+
+ sal_uInt16 nPrec = nCharCount >= nDigitPre ? nCharCount - nDigitPre : 0;
+ if (nPrec && bSign)
+ {
+ // Subtract the negative sign.
+ --nPrec;
+ }
+ if (nPrec)
+ {
+ // Subtract the decimal point.
+ --nPrec;
+ }
+ ImpGetOutputStdToPrecision(fNumber, rOutString, nPrec);
+ if (rOutString.getLength() > nCharCount)
+ {
+ // String still wider than desired. Switch to scientific notation.
+ lcl_GetOutputStringScientific(fNumber, nCharCount, GetFormatter(), rOutString);
+ }
+ return true;
+}
+
+sal_uInt16 SvNumberformat::GetSubformatIndex (double fNumber ) const
+{
+ sal_uInt16 nIx; // Index of the partial format
+ double fLimit_1 = fLimit1;
+ short nCheck = ImpCheckCondition(fNumber, fLimit_1, eOp1);
+ if (nCheck == -1 || nCheck == 1) // Only 1 String or True
+ {
+ nIx = 0;
+ }
+ else
+ {
+ double fLimit_2 = fLimit2;
+ nCheck = ImpCheckCondition(fNumber, fLimit_2, eOp2);
+ if (nCheck == -1 || nCheck == 1)
+ {
+ nIx = 1;
+ }
+ else
+ {
+ nIx = 2;
+ }
+ }
+ return nIx;
+}
+
+bool SvNumberformat::GetOutputString(double fNumber,
+ OUString& OutString,
+ const Color** ppColor)
+{
+ bool bRes = false;
+ OutString.clear();
+ *ppColor = nullptr; // No color change
+ if (eType & SvNumFormatType::LOGICAL && sFormatstring == rScan.GetKeywords()[NF_KEY_BOOLEAN])
+ {
+ if (fNumber)
+ {
+ OutString = rScan.GetTrueString();
+ }
+ else
+ {
+ OutString = rScan.GetFalseString();
+ }
+ return false;
+ }
+ OUStringBuffer sBuff(64);
+ if (eType & SvNumFormatType::TEXT)
+ {
+ ImpGetOutputStandard(fNumber, sBuff);
+ OutString = sBuff.makeStringAndClear();
+ return false;
+ }
+ bool bHadStandard = false;
+ if (bStandard) // Individual standard formats
+ {
+ if (rScan.GetStandardPrec() == SvNumberFormatter::INPUTSTRING_PRECISION) // All number format InputLine
+ {
+ ImpGetOutputInputLine(fNumber, OutString);
+ return false;
+ }
+ switch (eType)
+ {
+ case SvNumFormatType::NUMBER: // Standard number format
+ if (rScan.GetStandardPrec() == SvNumberFormatter::UNLIMITED_PRECISION)
+ {
+ if (std::signbit(fNumber))
+ {
+ if (!(fNumber < 0.0))
+ fNumber = -fNumber; // do not display -0.0
+ }
+ if (fNumber == 0.0)
+ {
+ OutString = "0";
+ }
+ else if (fNumber < 1.0 && fNumber > -1.0)
+ {
+ // Decide whether to display as 0.000000123... or 1.23...e-07
+ bool bFix = (fNumber < -EXP_LOWER_BOUND || EXP_LOWER_BOUND < fNumber);
+ if (!bFix)
+ {
+ // Arbitrary, not too many 0s visually, start E2 at 1E-10.
+ constexpr sal_Int32 kMaxExp = 9;
+ const sal_Int32 nExp = static_cast<sal_Int32>(ceil( -log10( fabs( fNumber))));
+ if (nExp <= kMaxExp && rtl::math::approxEqual(
+ rtl::math::round( fNumber, 16), rtl::math::round( fNumber, nExp + 16)))
+ {
+ // Not too many significant digits or accuracy
+ // artefacts, otherwise leave everything to E2
+ // format.
+ bFix = true;
+ }
+ }
+ if (bFix)
+ OutString = ::rtl::math::doubleToUString( fNumber,
+ rtl_math_StringFormat_F,
+ rtl_math_DecimalPlaces_Max,
+ GetFormatter().GetNumDecimalSep()[0], true);
+ else
+ OutString = ::rtl::math::doubleToUString( fNumber,
+ rtl_math_StringFormat_E2,
+ rtl_math_DecimalPlaces_Max,
+ GetFormatter().GetNumDecimalSep()[0], true);
+ }
+ else
+ {
+ OutString = ::rtl::math::doubleToUString( fNumber,
+ rtl_math_StringFormat_Automatic,
+ rtl_math_DecimalPlaces_Max,
+ GetFormatter().GetNumDecimalSep()[0], true);
+ }
+ return false;
+ }
+ ImpGetOutputStandard(fNumber, sBuff);
+ bHadStandard = true;
+ break;
+ case SvNumFormatType::DATE:
+ bRes |= ImpGetDateOutput(fNumber, 0, sBuff);
+ bHadStandard = true;
+ break;
+ case SvNumFormatType::TIME:
+ bRes |= ImpGetTimeOutput(fNumber, 0, sBuff);
+ bHadStandard = true;
+ break;
+ case SvNumFormatType::DATETIME:
+ bRes |= ImpGetDateTimeOutput(fNumber, 0, sBuff);
+ bHadStandard = true;
+ break;
+ default: break;
+ }
+ }
+ if ( !bHadStandard )
+ {
+ sal_uInt16 nIx = GetSubformatIndex ( fNumber ); // Index of the partial format
+ if (fNumber < 0.0 &&
+ ((nIx == 0 && IsFirstSubformatRealNegative()) || // 1st, usually positive subformat
+ (nIx == 1 && IsSecondSubformatRealNegative()))) // 2nd, usually negative subformat
+ {
+ fNumber = -fNumber; // eliminate sign
+ }
+ *ppColor = NumFor[nIx].GetColor();
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+ if (nCnt == 0 && rInfo.eScannedType == SvNumFormatType::EMPTY)
+ {
+ return false; // Empty => nothing
+ }
+ else if (nCnt == 0) // Else Standard Format
+ {
+ ImpGetOutputStandard(fNumber, sBuff);
+ OutString = sBuff.makeStringAndClear();
+ return false;
+ }
+ switch (rInfo.eScannedType)
+ {
+ case SvNumFormatType::TEXT:
+ case SvNumFormatType::DEFINED:
+ for (sal_uInt16 i = 0; i < nCnt; i++)
+ {
+ switch (rInfo.nTypeArray[i])
+ {
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ bRes = lcl_appendStarFillChar( sBuff, rInfo.sStrArray[i]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[i].getLength() >= 2)
+ InsertBlanks(sBuff, sBuff.getLength(), rInfo.sStrArray[i][1] );
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_CURRENCY:
+ sBuff.append(rInfo.sStrArray[i]);
+ break;
+ case NF_SYMBOLTYPE_THSEP:
+ if (rInfo.nThousand == 0)
+ {
+ sBuff.append(rInfo.sStrArray[i]);
+ }
+ break;
+ default:
+ break;
+ }
+ }
+ break;
+ case SvNumFormatType::DATE:
+ bRes |= ImpGetDateOutput(fNumber, nIx, sBuff);
+ break;
+ case SvNumFormatType::TIME:
+ bRes |= ImpGetTimeOutput(fNumber, nIx, sBuff);
+ break;
+ case SvNumFormatType::DATETIME:
+ bRes |= ImpGetDateTimeOutput(fNumber, nIx, sBuff);
+ break;
+ case SvNumFormatType::NUMBER:
+ case SvNumFormatType::PERCENT:
+ case SvNumFormatType::CURRENCY:
+ bRes |= ImpGetNumberOutput(fNumber, nIx, sBuff);
+ break;
+ case SvNumFormatType::LOGICAL:
+ bRes |= ImpGetLogicalOutput(fNumber, nIx, sBuff);
+ break;
+ case SvNumFormatType::FRACTION:
+ bRes |= ImpGetFractionOutput(fNumber, nIx, sBuff);
+ break;
+ case SvNumFormatType::SCIENTIFIC:
+ bRes |= ImpGetScientificOutput(fNumber, nIx, sBuff);
+ break;
+ default: break;
+ }
+ }
+ OutString = sBuff.makeStringAndClear();
+ return bRes;
+}
+
+bool SvNumberformat::ImpGetScientificOutput(double fNumber,
+ sal_uInt16 nIx,
+ OUStringBuffer& sStr)
+{
+ bool bRes = false;
+ bool bSign = false;
+
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+
+ if (fNumber < 0)
+ {
+ if (nIx == 0) // Not in the ones at the end
+ {
+ bSign = true; // Formats
+ }
+ fNumber = -fNumber;
+ }
+
+ sStr = ::rtl::math::doubleToUString( fNumber,
+ rtl_math_StringFormat_E,
+ rInfo.nCntPre + rInfo.nCntPost - 1, '.' );
+ OUStringBuffer ExpStr;
+ short nExpSign = 1;
+ sal_Int32 nExPos = sStr.indexOf('E');
+ sal_Int32 nDecPos = -1;
+
+ if ( nExPos >= 0 )
+ {
+ // split into mantissa and exponent and get rid of "E+" or "E-"
+ sal_Int32 nExpStart = nExPos + 1;
+
+ switch ( sStr[ nExpStart ] )
+ {
+ case '-' :
+ nExpSign = -1;
+ [[fallthrough]];
+ case '+' :
+ ++nExpStart;
+ break;
+ }
+ ExpStr = sStr.subView( nExpStart ); // part following the "E+"
+ sStr.truncate( nExPos );
+
+ if ( rInfo.nCntPre != 1 ) // rescale Exp
+ {
+ sal_Int32 nExp = OUString::unacquired(ExpStr).toInt32() * nExpSign;
+ sal_Int32 nRescale = (rInfo.nCntPre != 0) ? nExp % static_cast<sal_Int32>(rInfo.nCntPre) : -1;
+ if( nRescale < 0 && rInfo.nCntPre != 0 )
+ nRescale += static_cast<sal_Int32>(rInfo.nCntPre);
+ nExp -= nRescale;
+ if ( nExp < 0 )
+ {
+ nExpSign = -1;
+ nExp = -nExp;
+ }
+ else
+ {
+ nExpSign = 1;
+ }
+ ExpStr = OUString::number( nExp );
+ const sal_Unicode cFirstDigit = sStr[0];
+ // rescale mantissa
+ sStr = ::rtl::math::doubleToUString( fNumber,
+ rtl_math_StringFormat_E,
+ nRescale + rInfo.nCntPost, '.' );
+
+ // sStr now may contain a rounded-up value shifted into the next
+ // magnitude, for example 1.000E+02 (4 digits) for fNumber 99.995
+ // (9.9995E+02 rounded to 3 decimals) but we want the final result
+ // to be 100.00E+00 (5 digits), so for the following fill routines
+ // below to work correctly append a zero decimal.
+ /* TODO: this is awkward, could an engineering notation mode be
+ * introduced to rtl_math_doubleToUString()? */
+ sStr.truncate( sStr.indexOf('E') );
+ if (sStr[0] == '1' && cFirstDigit != '1')
+ sStr.append('0');
+ }
+
+ // cut any decimal delimiter
+ sal_Int32 index = 0;
+
+ while((index = sStr.indexOf('.', index)) >= 0)
+ {
+ if (nDecPos < 0)
+ nDecPos = index;
+ sStr.remove(index, 1);
+ }
+ }
+
+ sal_uInt16 j = nCnt-1; // Last symbol
+ sal_Int32 k; // Position in ExpStr
+ sal_Int32 nZeros = 0; // Erase leading zeros
+
+ bRes |= ImpNumberFill(ExpStr, fNumber, k, j, nIx, NF_SYMBOLTYPE_EXP);
+
+ while (nZeros < k && ExpStr[nZeros] == '0')
+ {
+ ++nZeros;
+ }
+ if (nZeros)
+ {
+ ExpStr.remove( 0, nZeros);
+ }
+
+ bool bCont = true;
+
+ if (rInfo.nTypeArray[j] == NF_SYMBOLTYPE_EXP)
+ {
+ const OUString& rStr = rInfo.sStrArray[j];
+ if (nExpSign == -1)
+ {
+ ExpStr.insert(0, '-');
+ }
+ else if (rStr.getLength() > 1 && rStr[1] == '+')
+ {
+ ExpStr.insert(0, '+');
+ }
+ ExpStr.insert(0, rStr[0]);
+ if ( j )
+ {
+ j--;
+ }
+ else
+ {
+ bCont = false;
+ }
+ }
+ // Continue main number:
+ if ( !bCont )
+ {
+ sStr.truncate();
+ }
+ else
+ {
+ bRes |= ImpDecimalFill(sStr, fNumber, nDecPos, j, nIx, false);
+ }
+
+ if (bSign)
+ {
+ sStr.insert(0, '-');
+ }
+ sStr.append(ExpStr);
+
+ return bRes;
+}
+
+double SvNumberformat::GetRoundFractionValue ( double fNumber ) const
+{
+ sal_uInt16 nIx = GetSubformatIndex ( fNumber );
+ double fIntPart = 0.0; // integer part of fraction
+ sal_Int64 nFrac = 0, nDiv = 1; // numerator and denominator
+ double fSign = (fNumber < 0.0) ? -1.0 : 1.0;
+ // fNumber is modified in ImpGetFractionElements to absolute fractional part
+ ImpGetFractionElements ( fNumber, nIx, fIntPart, nFrac, nDiv );
+ if ( nDiv > 0 )
+ return fSign * ( fIntPart + static_cast<double>(nFrac) / static_cast<double>(nDiv) );
+ else
+ return fSign * fIntPart;
+}
+
+void SvNumberformat::ImpGetFractionElements ( double& fNumber, sal_uInt16 nIx,
+ double& fIntPart, sal_Int64& nFrac, sal_Int64& nDiv ) const
+{
+ if ( fNumber < 0.0 )
+ fNumber = -fNumber;
+ fIntPart = floor(fNumber); // Integral part
+ fNumber -= fIntPart; // Fractional part
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ sal_Int64 nForcedDiv = lcl_GetDenominatorString( rInfo, NumFor[nIx].GetCount() ).toInt32();
+ if( nForcedDiv > 0 )
+ { // Forced Denominator
+ nDiv = nForcedDiv;
+ nFrac = static_cast<sal_Int64>(floor ( fNumber * nDiv ));
+ double fFracNew = static_cast<double>(nFrac) / static_cast<double>(nDiv);
+ double fFracNew1 = static_cast<double>(nFrac + 1) / static_cast<double>(nDiv);
+ double fDiff = fNumber - fFracNew;
+ if( fDiff > ( fFracNew1 - fNumber ) )
+ {
+ nFrac++;
+ }
+ }
+ else // Calculated Denominator
+ {
+ nDiv = 1;
+ sal_Int64 nBasis = static_cast<sal_Int64>(floor( pow(10.0,rInfo.nCntExp))) - 1; // 9, 99, 999 ,...
+ sal_Int64 nFracPrev = 1, nDivPrev = 0, nFracNext, nDivNext, nPartialDenom;
+ double fRemainder = fNumber;
+
+ // Use continued fraction representation of fNumber
+ // See https://en.wikipedia.org/wiki/Continued_fraction#Best_rational_approximations
+ while ( fRemainder > 0.0 )
+ {
+ double fTemp = 1.0 / fRemainder; // 64bits precision required when fRemainder is very weak
+ nPartialDenom = static_cast<sal_Int64>(floor(fTemp)); // due to floating point notation with double precision
+ fRemainder = fTemp - static_cast<double>(nPartialDenom);
+ nDivNext = nPartialDenom * nDiv + nDivPrev;
+ if ( nDivNext <= nBasis ) // continue loop
+ {
+ nFracNext = nPartialDenom * nFrac + nFracPrev;
+ nFracPrev = nFrac;
+ nFrac = nFracNext;
+ nDivPrev = nDiv;
+ nDiv = nDivNext;
+ }
+ else // calculate collateral fraction and exit
+ {
+ sal_Int64 nCollat = (nBasis - nDivPrev) / nDiv;
+ if ( 2 * nCollat >= nPartialDenom )
+ {
+ sal_Int64 nFracTest = nCollat * nFrac + nFracPrev;
+ sal_Int64 nDivTest = nCollat * nDiv + nDivPrev;
+ double fSign = (static_cast<double>(nFrac) > fNumber * static_cast<double>(nDiv))?1.0:-1.0;
+ if ( fSign * ( double(nFrac * nDivTest + nDiv * nFracTest) - 2.0 * double(nDiv * nDivTest) * fNumber ) > 0.0 )
+ {
+ nFrac = nFracTest;
+ nDiv = nDivTest;
+ }
+ }
+ fRemainder = 0.0; // exit while loop
+ }
+ }
+ }
+ if (nFrac >= nDiv)
+ {
+ ++fIntPart;
+ nFrac = 0;
+ nDiv = ( nForcedDiv > 0 ) ? nForcedDiv : 1;
+ }
+}
+
+bool SvNumberformat::ImpGetFractionOutput(double fNumber,
+ sal_uInt16 nIx,
+ OUStringBuffer& sBuff)
+{
+ bool bRes = false;
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+ OUStringBuffer sStr, sFrac, sDiv; // Strings, value for Integral part Numerator and denominator
+ bool bSign = ( (fNumber < 0) && (nIx == 0) ); // sign Not in the ones at the end
+ const OUString sIntegerFormat = lcl_GetFractionIntegerString(rInfo, nCnt);
+ const OUString sNumeratorFormat = lcl_GetNumeratorString(rInfo, nCnt);
+ const OUString sDenominatorFormat = lcl_GetDenominatorString(rInfo, nCnt);
+
+ sal_Int64 nFrac = 0, nDiv = 1;
+ double fNum = floor(fNumber); // Integral part
+
+ if (fNum > D_MAX_U_INT32 || rInfo.nCntExp > 9) // Too large
+ {
+ sBuff = ImpSvNumberformatScan::sErrStr;
+ return false;
+ }
+ if (rInfo.nCntExp == 0)
+ {
+ SAL_WARN( "svl.numbers", "SvNumberformat:: Fraction, nCntExp == 0");
+ sBuff.truncate();
+ return false;
+ }
+
+ ImpGetFractionElements( fNumber, nIx, fNum, nFrac, nDiv);
+
+ if (rInfo.nCntPre == 0) // Improper fraction
+ {
+ double fNum1 = fNum * static_cast<double>(nDiv) + static_cast<double>(nFrac);
+
+ if (fNum1 > D_MAX_INTEGER)
+ {
+ sBuff = ImpSvNumberformatScan::sErrStr;
+ return false;
+ }
+ nFrac = static_cast<sal_Int64>(floor(fNum1));
+ }
+ else if (fNum == 0.0 && nFrac != 0)
+ {
+ }
+ else
+ {
+ char aBuf[100];
+ sprintf( aBuf, "%.f", fNum ); // simple rounded integer (#100211# - checked)
+ sStr.appendAscii( aBuf );
+ impTransliterate(sStr, NumFor[nIx].GetNatNum());
+ }
+ bool bHideFraction = (rInfo.nCntPre > 0 && nFrac == 0
+ && (sNumeratorFormat.indexOf('0') < 0)
+ && (sDenominatorFormat.indexOf('0') < 0
+ || sDenominatorFormat.toInt32() > 0) );
+ if ( bHideFraction )
+ {
+ sDiv.truncate();
+ }
+ else // if there are some '0' in format, force display of fraction
+ {
+ sFrac = ImpIntToString( nIx, nFrac );
+ sDiv = ImpIntToString( nIx, nDiv );
+ }
+
+ sal_uInt16 j = nCnt-1; // Last symbol -> backwards
+ sal_Int32 k; // Denominator
+
+ bRes |= ImpNumberFill(sDiv, fNumber, k, j, nIx, NF_SYMBOLTYPE_FRAC, true);
+
+ bool bCont = true;
+ if (rInfo.nTypeArray[j] == NF_SYMBOLTYPE_FRAC)
+ {
+ if ( bHideFraction )
+ { // do not insert blank for fraction if there is no '?'
+ if ( sNumeratorFormat.indexOf('?') >= 0
+ || sDenominatorFormat.indexOf('?') >= 0 )
+ sDiv.insert(0, ' ');
+ }
+ else
+ {
+ sDiv.insert(0, rInfo.sStrArray[j][0]);
+ }
+ if ( j )
+ {
+ j--;
+ }
+ else
+ {
+ bCont = false;
+ }
+ }
+ // Further numerators:
+ if ( !bCont )
+ {
+ sFrac.truncate();
+ }
+ else
+ {
+ bRes |= ImpNumberFill(sFrac, fNumber, k, j, nIx, NF_SYMBOLTYPE_FRACBLANK);
+ bCont = false; // there is no integer part?
+ if (rInfo.nTypeArray[j] == NF_SYMBOLTYPE_FRACBLANK)
+ {
+ if ( j )
+ {
+ if ( bHideFraction )
+ { // '?' in any format force display of blank as delimiter
+ if ( sIntegerFormat.indexOf('?') >= 0
+ || sNumeratorFormat.indexOf('?') >= 0
+ || sDenominatorFormat.indexOf('?') >= 0 )
+ {
+ for (sal_Int32 i = 0; i < rInfo.sStrArray[j].getLength(); i++)
+ sFrac.insert(0, ' ');
+ }
+ }
+ else
+ {
+ if ( fNum != 0.0 || sIntegerFormat.indexOf('0') >= 0 )
+ sFrac.insert(0, rInfo.sStrArray[j]); // insert Blank string only if there are both integer and fraction
+ else
+ {
+ if ( sIntegerFormat.indexOf('?') >= 0
+ || sNumeratorFormat.indexOf('?') >= 0 )
+ {
+ for (sal_Int32 i = 0; i < rInfo.sStrArray[j].getLength(); i++)
+ sFrac.insert(0, ' ');
+ }
+ }
+ }
+ j--;
+ bCont = true; // Yes, there is an integer
+ }
+ else
+ sFrac.insert(0, rInfo.sStrArray[j]);
+ }
+ }
+ // Continue integer part
+ if ( !bCont )
+ {
+ sStr.truncate();
+ }
+ else
+ {
+ k = sStr.getLength(); // After last figure
+ bRes |= ImpNumberFillWithThousands(sStr, fNumber, k, j, nIx,
+ rInfo.nCntPre);
+ }
+ if (bSign && (nFrac != 0 || fNum != 0.0))
+ {
+ sBuff.insert(0, '-'); // Not -0
+ }
+ sBuff.append(sStr);
+ sBuff.append(sFrac);
+ sBuff.append(sDiv);
+ return bRes;
+}
+
+sal_uInt16 SvNumberformat::ImpGetFractionOfSecondString( OUStringBuffer& rBuf, double fFractionOfSecond,
+ int nFractionDecimals, bool bAddOneRoundingDecimal, sal_uInt16 nIx, sal_uInt16 nMinimumInputLineDecimals )
+{
+ if (!nFractionDecimals)
+ return 0;
+
+ // nFractionDecimals+1 to not round up what Time::GetClock() carefully
+ // truncated.
+ rBuf.append( rtl::math::doubleToUString( fFractionOfSecond, rtl_math_StringFormat_F,
+ (bAddOneRoundingDecimal ? nFractionDecimals + 1 : nFractionDecimals), '.'));
+ rBuf.stripStart('0');
+ rBuf.stripStart('.');
+ if (bAddOneRoundingDecimal && rBuf.getLength() > nFractionDecimals)
+ rBuf.truncate( nFractionDecimals); // the digit appended because of nFractionDecimals+1
+ if (nMinimumInputLineDecimals)
+ {
+ rBuf.stripEnd('0');
+ for (sal_Int32 index = rBuf.getLength(); index < nMinimumInputLineDecimals; ++index)
+ {
+ rBuf.append('0');
+ }
+ impTransliterate(rBuf, NumFor[nIx].GetNatNum());
+ nFractionDecimals = rBuf.getLength();
+ }
+ else
+ {
+ impTransliterate(rBuf, NumFor[nIx].GetNatNum());
+ }
+ return static_cast<sal_uInt16>(nFractionDecimals);
+}
+
+bool SvNumberformat::ImpGetTimeOutput(double fNumber,
+ sal_uInt16 nIx,
+ OUStringBuffer& sBuff)
+{
+ using namespace ::com::sun::star::i18n;
+ bool bCalendarSet = false;
+ const double fNumberOrig = fNumber;
+ bool bRes = false;
+ bool bSign = false;
+ if (fNumber < 0.0)
+ {
+ fNumber = -fNumber;
+ if (nIx == 0)
+ {
+ bSign = true;
+ }
+ }
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ bool bInputLine;
+ sal_Int32 nCntPost;
+ if ( rScan.GetStandardPrec() == SvNumberFormatter::INPUTSTRING_PRECISION &&
+ 0 < rInfo.nCntPost && rInfo.nCntPost < 7 )
+ { // round at 7 decimals (+5 of 86400 == 12 significant digits)
+ bInputLine = true;
+ nCntPost = 7;
+ }
+ else
+ {
+ bInputLine = false;
+ nCntPost = rInfo.nCntPost;
+ }
+
+ OUStringBuffer sSecStr;
+ sal_Int32 nSecPos = 0; // For figure by figure processing
+ sal_uInt32 nHour, nMin, nSec;
+ if (!rInfo.bThousand) // No [] format
+ {
+ sal_uInt16 nCHour, nCMinute, nCSecond;
+ double fFractionOfSecond;
+ tools::Time::GetClock( fNumberOrig, nCHour, nCMinute, nCSecond, fFractionOfSecond, nCntPost);
+ nHour = nCHour;
+ nMin = nCMinute;
+ nSec = nCSecond;
+ nCntPost = ImpGetFractionOfSecondString( sSecStr, fFractionOfSecond, nCntPost, true, nIx,
+ (bInputLine ? rInfo.nCntPost : 0));
+ }
+ else
+ {
+ const double fTime = rtl::math::round( fNumber * 86400.0, int(nCntPost));
+ if (bSign && fTime == 0.0)
+ {
+ bSign = false; // Not -00:00:00
+ }
+ if (fTime > D_MAX_U_INT32)
+ {
+ sBuff = ImpSvNumberformatScan::sErrStr;
+ return false;
+ }
+ sal_uInt32 nSeconds = static_cast<sal_uInt32>(fTime);
+
+ nCntPost = ImpGetFractionOfSecondString( sSecStr, fTime - nSeconds, nCntPost, false, nIx,
+ (bInputLine ? rInfo.nCntPost : 0));
+
+ if (rInfo.nThousand == 3) // [ss]
+ {
+ nHour = 0;
+ nMin = 0;
+ nSec = nSeconds;
+ }
+ else if (rInfo.nThousand == 2) // [mm]:ss
+ {
+ nHour = 0;
+ nMin = nSeconds / 60;
+ nSec = nSeconds % 60;
+ }
+ else if (rInfo.nThousand == 1) // [hh]:mm:ss
+ {
+ nHour = nSeconds / 3600;
+ nMin = (nSeconds%3600) / 60;
+ nSec = nSeconds%60;
+ }
+ else
+ {
+ // TODO What should these be set to?
+ nHour = 0;
+ nMin = 0;
+ nSec = 0;
+ }
+ }
+
+ sal_Unicode cAmPm = ' '; // a or p
+ if (rInfo.nCntExp) // AM/PM
+ {
+ if (nHour == 0)
+ {
+ nHour = 12;
+ cAmPm = 'a';
+ }
+ else if (nHour < 12)
+ {
+ cAmPm = 'a';
+ }
+ else
+ {
+ cAmPm = 'p';
+ if (nHour > 12)
+ {
+ nHour -= 12;
+ }
+ }
+ }
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+ for (sal_uInt16 i = 0; i < nCnt; i++)
+ {
+ sal_Int32 nLen;
+ switch (rInfo.nTypeArray[i])
+ {
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ bRes = lcl_appendStarFillChar( sBuff, rInfo.sStrArray[i]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[i].getLength() >= 2)
+ InsertBlanks(sBuff, sBuff.getLength(), rInfo.sStrArray[i][1] );
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_CURRENCY:
+ case NF_SYMBOLTYPE_DATESEP:
+ case NF_SYMBOLTYPE_TIMESEP:
+ case NF_SYMBOLTYPE_TIME100SECSEP:
+ sBuff.append(rInfo.sStrArray[i]);
+ break;
+ case NF_SYMBOLTYPE_DIGIT:
+ nLen = ( bInputLine && i > 0 &&
+ (rInfo.nTypeArray[i-1] == NF_SYMBOLTYPE_STRING ||
+ rInfo.nTypeArray[i-1] == NF_SYMBOLTYPE_TIME100SECSEP) ?
+ nCntPost : rInfo.sStrArray[i].getLength() );
+ for (sal_Int32 j = 0; j < nLen && nSecPos < nCntPost && nSecPos < sSecStr.getLength(); ++j)
+ {
+ sBuff.append(sSecStr[nSecPos]);
+ nSecPos++;
+ }
+ break;
+ case NF_KEY_AMPM: // AM/PM
+ if ( !bCalendarSet )
+ {
+ double fDiff = DateTime(rScan.GetNullDate()) - GetCal().getEpochStart();
+ fDiff += fNumberOrig;
+ GetCal().setLocalDateTime( fDiff );
+ bCalendarSet = true;
+ }
+ if (cAmPm == 'a')
+ {
+ sBuff.append(GetCal().getDisplayName(
+ CalendarDisplayIndex::AM_PM, AmPmValue::AM, 0 ));
+ }
+ else
+ {
+ sBuff.append(GetCal().getDisplayName(
+ CalendarDisplayIndex::AM_PM, AmPmValue::PM, 0 ));
+ }
+ break;
+ case NF_KEY_AP: // A/P
+ if (cAmPm == 'a')
+ {
+ sBuff.append('a');
+ }
+ else
+ {
+ sBuff.append('p');
+ }
+ break;
+ case NF_KEY_MI: // M
+ sBuff.append(ImpIntToString( nIx, nMin ));
+ break;
+ case NF_KEY_MMI: // MM
+ sBuff.append(ImpIntToString( nIx, nMin, 2 ));
+ break;
+ case NF_KEY_H: // H
+ sBuff.append(ImpIntToString( nIx, nHour ));
+ break;
+ case NF_KEY_HH: // HH
+ sBuff.append(ImpIntToString( nIx, nHour, 2 ));
+ break;
+ case NF_KEY_S: // S
+ sBuff.append(ImpIntToString( nIx, nSec ));
+ break;
+ case NF_KEY_SS: // SS
+ sBuff.append(ImpIntToString( nIx, nSec, 2 ));
+ break;
+ default:
+ break;
+ }
+ }
+ if (bSign && rInfo.bThousand)
+ {
+ sBuff.insert(0, '-');
+ }
+ return bRes;
+}
+
+
+/** If a day of month occurs within the format, the month name is in possessive
+ genitive case if the day follows the month, and partitive case if the day
+ precedes the month. If there is no day of month the nominative case (noun)
+ is returned. Also if the month is immediately preceded or followed by a
+ literal string other than space and not followed by a comma, the nominative
+ name is used, this prevents duplicated casing for MMMM\t\a and such in
+ documents imported from (e.g. Finnish) Excel or older LibO/OOo releases.
+ */
+
+// IDEA: instead of eCodeType pass the index to nTypeArray and restrict
+// inspection of month name around that one, that would enable different month
+// cases in one format. Though probably the most rare use case ever...
+
+sal_Int32 SvNumberformat::ImpUseMonthCase( int & io_nState, const ImpSvNumFor& rNumFor, NfKeywordIndex eCodeType )
+{
+ using namespace ::com::sun::star::i18n;
+ if (!io_nState)
+ {
+ bool bMonthSeen = false;
+ bool bDaySeen = false;
+ const ImpSvNumberformatInfo& rInfo = rNumFor.Info();
+ const sal_uInt16 nCount = rNumFor.GetCount();
+ for (sal_uInt16 i = 0; i < nCount && io_nState == 0; ++i)
+ {
+ sal_Int32 nLen;
+ switch (rInfo.nTypeArray[i])
+ {
+ case NF_KEY_D :
+ case NF_KEY_DD :
+ if (bMonthSeen)
+ {
+ io_nState = 2;
+ }
+ else
+ {
+ bDaySeen = true;
+ }
+ break;
+ case NF_KEY_MMM:
+ case NF_KEY_MMMM:
+ case NF_KEY_MMMMM:
+ if ((i < nCount-1 &&
+ rInfo.nTypeArray[i+1] == NF_SYMBOLTYPE_STRING &&
+ // Literal following, not empty, space nor comma.
+ !rInfo.sStrArray[i+1].isEmpty() &&
+ rInfo.sStrArray[i+1][0] != ' ' && rInfo.sStrArray[i+1][0] != ',') ||
+ (i > 0 && rInfo.nTypeArray[i-1] == NF_SYMBOLTYPE_STRING &&
+ ((nLen = rInfo.sStrArray[i-1].getLength()) > 0) &&
+ // Literal preceding, not space.
+ rInfo.sStrArray[i-1][nLen-1] != ' '))
+ {
+ io_nState = 1;
+ }
+ else if (bDaySeen)
+ {
+ io_nState = 3;
+ }
+ else
+ {
+ bMonthSeen = true;
+ }
+ break;
+ }
+ }
+ if (io_nState == 0)
+ {
+ io_nState = 1; // No day of month
+ }
+ }
+ switch (io_nState)
+ {
+ case 1:
+ // No day of month or forced nominative
+ switch (eCodeType)
+ {
+ case NF_KEY_MMM:
+ return CalendarDisplayCode::SHORT_MONTH_NAME;
+ case NF_KEY_MMMM:
+ return CalendarDisplayCode::LONG_MONTH_NAME;
+ case NF_KEY_MMMMM:
+ return CalendarDisplayCode::NARROW_MONTH_NAME;
+ default:
+ ; // nothing
+ }
+ break;
+ case 2:
+ // Day of month follows month (the month's 17th)
+ switch (eCodeType)
+ {
+ case NF_KEY_MMM:
+ return CalendarDisplayCode::SHORT_GENITIVE_MONTH_NAME;
+ case NF_KEY_MMMM:
+ return CalendarDisplayCode::LONG_GENITIVE_MONTH_NAME;
+ case NF_KEY_MMMMM:
+ return CalendarDisplayCode::NARROW_GENITIVE_MONTH_NAME;
+ default:
+ ; // Nothing
+ }
+ break;
+ case 3:
+ // Day of month precedes month (17 of month)
+ switch (eCodeType)
+ {
+ case NF_KEY_MMM:
+ return CalendarDisplayCode::SHORT_PARTITIVE_MONTH_NAME;
+ case NF_KEY_MMMM:
+ return CalendarDisplayCode::LONG_PARTITIVE_MONTH_NAME;
+ case NF_KEY_MMMMM:
+ return CalendarDisplayCode::NARROW_PARTITIVE_MONTH_NAME;
+ default:
+ ; // nothing
+ }
+ break;
+ }
+ SAL_WARN( "svl.numbers", "ImpUseMonthCase: unhandled keyword index eCodeType");
+ return CalendarDisplayCode::LONG_MONTH_NAME;
+}
+
+
+bool SvNumberformat::ImpIsOtherCalendar( const ImpSvNumFor& rNumFor ) const
+{
+ if ( GetCal().getUniqueID() != GREGORIAN )
+ {
+ return false;
+ }
+ const ImpSvNumberformatInfo& rInfo = rNumFor.Info();
+ const sal_uInt16 nCnt = rNumFor.GetCount();
+ sal_uInt16 i;
+ for ( i = 0; i < nCnt; i++ )
+ {
+ switch ( rInfo.nTypeArray[i] )
+ {
+ case NF_SYMBOLTYPE_CALENDAR :
+ return false;
+ case NF_KEY_EC :
+ case NF_KEY_EEC :
+ case NF_KEY_R :
+ case NF_KEY_RR :
+ case NF_KEY_AAA :
+ case NF_KEY_AAAA :
+ case NF_KEY_G :
+ case NF_KEY_GG :
+ case NF_KEY_GGG :
+ return true;
+ }
+ }
+ return false;
+}
+
+void SvNumberformat::SwitchToOtherCalendar( OUString& rOrgCalendar,
+ double& fOrgDateTime ) const
+{
+ CalendarWrapper& rCal = GetCal();
+ if ( rCal.getUniqueID() != GREGORIAN )
+ return;
+
+ using namespace ::com::sun::star::i18n;
+ const css::uno::Sequence< OUString > xCals = rCal.getAllCalendars(
+ rLoc().getLanguageTag().getLocale() );
+ sal_Int32 nCnt = xCals.getLength();
+ if ( nCnt <= 1 )
+ return;
+
+ auto pCal = std::find_if(xCals.begin(), xCals.end(),
+ [](const OUString& rCalName) { return rCalName != GREGORIAN; });
+ if (pCal == xCals.end())
+ return;
+
+ if ( !rOrgCalendar.getLength() )
+ {
+ rOrgCalendar = rCal.getUniqueID();
+ fOrgDateTime = rCal.getDateTime();
+ }
+ rCal.loadCalendar( *pCal, rLoc().getLanguageTag().getLocale() );
+ rCal.setDateTime( fOrgDateTime );
+}
+
+void SvNumberformat::SwitchToGregorianCalendar( const OUString& rOrgCalendar,
+ double fOrgDateTime ) const
+{
+ CalendarWrapper& rCal = GetCal();
+ if ( rOrgCalendar.getLength() && rCal.getUniqueID() != GREGORIAN )
+ {
+ rCal.loadCalendar( GREGORIAN, rLoc().getLanguageTag().getLocale() );
+ rCal.setDateTime( fOrgDateTime );
+ }
+}
+
+bool SvNumberformat::ImpFallBackToGregorianCalendar( OUString& rOrgCalendar, double& fOrgDateTime )
+{
+ using namespace ::com::sun::star::i18n;
+ CalendarWrapper& rCal = GetCal();
+ if ( rCal.getUniqueID() != GREGORIAN )
+ {
+ sal_Int16 nVal = rCal.getValue( CalendarFieldIndex::ERA );
+ if ( nVal == 0 && rCal.getLoadedCalendar().Eras[0].ID == "Dummy" )
+ {
+ if ( !rOrgCalendar.getLength() )
+ {
+ rOrgCalendar = rCal.getUniqueID();
+ fOrgDateTime = rCal.getDateTime();
+ }
+ else if ( rOrgCalendar == GREGORIAN )
+ {
+ rOrgCalendar.clear();
+ }
+ rCal.loadCalendar( GREGORIAN, rLoc().getLanguageTag().getLocale() );
+ rCal.setDateTime( fOrgDateTime );
+ return true;
+ }
+ }
+ return false;
+}
+
+
+#ifdef THE_FUTURE
+/* XXX NOTE: even if the ImpSwitchToSpecifiedCalendar method is currently
+ * unused please don't remove it, it would be needed by
+ * SwitchToSpecifiedCalendar(), see comment in
+ * ImpSvNumberInputScan::GetDateRef() */
+
+bool SvNumberformat::ImpSwitchToSpecifiedCalendar( OUString& rOrgCalendar,
+ double& fOrgDateTime,
+ const ImpSvNumFor& rNumFor ) const
+{
+ const ImpSvNumberformatInfo& rInfo = rNumFor.Info();
+ const sal_uInt16 nCnt = rNumFor.GetCount();
+ for ( sal_uInt16 i = 0; i < nCnt; i++ )
+ {
+ if ( rInfo.nTypeArray[i] == NF_SYMBOLTYPE_CALENDAR )
+ {
+ CalendarWrapper& rCal = GetCal();
+ if ( !rOrgCalendar.getLength() )
+ {
+ rOrgCalendar = rCal.getUniqueID();
+ fOrgDateTime = rCal.getDateTime();
+ }
+ rCal.loadCalendar( rInfo.sStrArray[i], rLoc().getLocale() );
+ rCal.setDateTime( fOrgDateTime );
+ return true;
+ }
+ }
+ return false;
+}
+#endif
+
+// static
+void SvNumberformat::ImpAppendEraG( OUStringBuffer& OutString,
+ const CalendarWrapper& rCal,
+ sal_Int16 nNatNum )
+{
+ using namespace ::com::sun::star::i18n;
+ if ( rCal.getUniqueID() == "gengou" )
+ {
+ sal_Unicode cEra;
+ sal_Int16 nVal = rCal.getValue( CalendarFieldIndex::ERA );
+ switch ( nVal )
+ {
+ case 1:
+ cEra = 'M';
+ break;
+ case 2:
+ cEra = 'T';
+ break;
+ case 3:
+ cEra = 'S';
+ break;
+ case 4:
+ cEra = 'H';
+ break;
+ case 5:
+ cEra = 'R';
+ break;
+ default:
+ cEra = '?';
+ break;
+ }
+ OutString.append(cEra);
+ }
+ else
+ {
+ OutString.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_ERA, nNatNum ));
+ }
+}
+
+bool SvNumberformat::ImpIsIso8601( const ImpSvNumFor& rNumFor ) const
+{
+ bool bIsIso = false;
+ if (eType & SvNumFormatType::DATE)
+ {
+ enum State
+ {
+ eNone,
+ eAtYear,
+ eAtSep1,
+ eAtMonth,
+ eAtSep2,
+ eNotIso
+ };
+ State eState = eNone;
+ auto & rTypeArray = rNumFor.Info().nTypeArray;
+ sal_uInt16 nCnt = rNumFor.GetCount();
+ for (sal_uInt16 i=0; i < nCnt && !bIsIso && eState != eNotIso; ++i)
+ {
+ switch ( rTypeArray[i] )
+ {
+ case NF_KEY_YY: // two digits not strictly ISO 8601
+ case NF_KEY_YYYY:
+ if (eState != eNone)
+ {
+ eState = eNotIso;
+ }
+ else
+ {
+ eState = eAtYear;
+ }
+ break;
+ case NF_KEY_M: // single digit not strictly ISO 8601
+ case NF_KEY_MM:
+ if (eState != eAtSep1)
+ {
+ eState = eNotIso;
+ }
+ else
+ {
+ eState = eAtMonth;
+ }
+ break;
+ case NF_KEY_D: // single digit not strictly ISO 8601
+ case NF_KEY_DD:
+ if (eState != eAtSep2)
+ {
+ eState = eNotIso;
+ }
+ else
+ {
+ bIsIso = true;
+ }
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_DATESEP:
+ if (rNumFor.Info().sStrArray[i] == "-")
+ {
+ if (eState == eAtYear)
+ {
+ eState = eAtSep1;
+ }
+ else if (eState == eAtMonth)
+ {
+ eState = eAtSep2;
+ }
+ else
+ {
+ eState = eNotIso;
+ }
+ }
+ else
+ {
+ eState = eNotIso;
+ }
+ break;
+ default:
+ eState = eNotIso;
+ }
+ }
+ }
+ else
+ {
+ SAL_WARN( "svl.numbers", "SvNumberformat::ImpIsIso8601: no date" );
+ }
+ return bIsIso;
+}
+
+static bool lcl_hasEra( const ImpSvNumFor& rNumFor )
+{
+ const ImpSvNumberformatInfo& rInfo = rNumFor.Info();
+ const sal_uInt16 nCnt = rNumFor.GetCount();
+ for ( sal_uInt16 i = 0; i < nCnt; i++ )
+ {
+ switch ( rInfo.nTypeArray[i] )
+ {
+ case NF_KEY_RR :
+ case NF_KEY_G :
+ case NF_KEY_GG :
+ case NF_KEY_GGG :
+ return true;
+ }
+ }
+ return false;
+}
+
+static bool lcl_isSignedYear( const CalendarWrapper& rCal, const ImpSvNumFor& rNumFor )
+{
+ return rCal.getValue( css::i18n::CalendarFieldIndex::ERA ) == 0 &&
+ rCal.getUniqueID() == GREGORIAN && !lcl_hasEra( rNumFor );
+}
+
+/* XXX: if needed this could be stripped from rEpochStart and diff adding and
+ * moved to tools' DateTime to be reused elsewhere. */
+static bool lcl_getValidDate( const DateTime& rNullDate, const DateTime& rEpochStart, double& fNumber )
+{
+ static const DateTime aCE( Date(1,1,1));
+ static const DateTime aMin( Date(1,1, SAL_MIN_INT16));
+ static const DateTime aMax( Date(31,12, SAL_MAX_INT16), tools::Time(23,59,59, tools::Time::nanoSecPerSec - 1));
+ static const double fMin = aMin - aCE;
+ static const double fMax = aMax - aCE;
+ // Value must be representable in our tools::Date proleptic Gregorian
+ // calendar as well.
+ const double fOff = (rNullDate - aCE) + fNumber;
+ // Add diff between epochs to serial date number.
+ const double fDiff = rNullDate - rEpochStart;
+ fNumber += fDiff;
+ return fMin <= fOff && fOff <= fMax;
+}
+
+bool SvNumberformat::ImpGetDateOutput(double fNumber,
+ sal_uInt16 nIx,
+ OUStringBuffer& sBuff)
+{
+ using namespace ::com::sun::star::i18n;
+ bool bRes = false;
+
+ CalendarWrapper& rCal = GetCal();
+ if (!lcl_getValidDate( rScan.GetNullDate(), rCal.getEpochStart(), fNumber))
+ {
+ sBuff = ImpSvNumberformatScan::sErrStr;
+ return false;
+ }
+ rCal.setLocalDateTime( fNumber );
+ int nUseMonthCase = 0; // Not decided yet
+ OUString aOrgCalendar; // empty => not changed yet
+
+ double fOrgDateTime(0.0);
+ bool bOtherCalendar = ImpIsOtherCalendar( NumFor[nIx] );
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ if ( ImpFallBackToGregorianCalendar( aOrgCalendar, fOrgDateTime ) )
+ {
+ bOtherCalendar = false;
+ }
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+ sal_Int16 nNatNum = NumFor[nIx].GetNatNum().GetNatNum();
+ OUString aStr;
+
+ // NatNum12: if the date format contains more than a date
+ // field, it needs to specify in NatNum12 argument
+ // which date element needs special formatting:
+ //
+ // '[NatNum12 ordinal-number]D' -> "1st"
+ // '[NatNum12 D=ordinal-number]D" of "MMMM' -> "1st of April"
+ // '[NatNum12 D=ordinal]D" of "MMMM' -> "first of April"
+ // '[NatNum12 YYYY=year,D=ordinal]D" of "MMMM", "YYYY' -> "first of April, nineteen ninety"
+ //
+ // Note: set only for YYYY, MMMM, M, DDDD, D and NNN/AAAA in date formats.
+ // Additionally for MMMMM, MMM, DDD and NN/AA to support at least
+ // capitalize, upper, lower, title.
+ // XXX It's possible to extend this for other keywords and date + time
+ // combinations, as required.
+
+ bool bUseSpellout = NatNumTakesParameters(nNatNum) &&
+ (nCnt == 1 || NumFor[nIx].GetNatNum().GetParams().indexOf('=') > -1);
+
+ for (sal_uInt16 i = 0; i < nCnt; i++)
+ {
+ switch (rInfo.nTypeArray[i])
+ {
+ case NF_SYMBOLTYPE_CALENDAR :
+ if ( !aOrgCalendar.getLength() )
+ {
+ aOrgCalendar = rCal.getUniqueID();
+ fOrgDateTime = rCal.getDateTime();
+ }
+ rCal.loadCalendar( rInfo.sStrArray[i], rLoc().getLanguageTag().getLocale() );
+ rCal.setDateTime( fOrgDateTime );
+ ImpFallBackToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ break;
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ bRes = lcl_appendStarFillChar( sBuff, rInfo.sStrArray[i]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[i].getLength() >= 2)
+ InsertBlanks( sBuff, sBuff.getLength(), rInfo.sStrArray[i][1] );
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_CURRENCY:
+ case NF_SYMBOLTYPE_DATESEP:
+ case NF_SYMBOLTYPE_TIMESEP:
+ case NF_SYMBOLTYPE_TIME100SECSEP:
+ sBuff.append(rInfo.sStrArray[i]);
+ break;
+ case NF_KEY_M: // M
+ aStr = rCal.getDisplayString( CalendarDisplayCode::SHORT_MONTH, nNatNum );
+ // NatNum12: support variants of preposition, suffixation or article
+ // for example, Catalan "de març", but "d'abril" etc.
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ break;
+ case NF_KEY_MM: // MM
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_MONTH, nNatNum ));
+ break;
+ case NF_KEY_MMM: // MMM
+ case NF_KEY_MMMM: // MMMM
+ case NF_KEY_MMMMM: // MMMMM
+ // NatNum12: support variants of preposition, suffixation or
+ // article, or capitalize, upper, lower, title.
+ // Note: result of the "spell out" conversion can depend from the optional
+ // PartitiveMonths or GenitiveMonths defined in the locale data,
+ // see description of ImpUseMonthCase(), and locale data in
+ // i18npool/source/localedata/data/ and libnumbertext
+ aStr = rCal.getDisplayString( ImpUseMonthCase( nUseMonthCase, NumFor[nIx],
+ static_cast<NfKeywordIndex>(rInfo.nTypeArray[i])),
+ nNatNum);
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ break;
+ case NF_KEY_Q: // Q
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_QUARTER, nNatNum ));
+ break;
+ case NF_KEY_QQ: // QQ
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_QUARTER, nNatNum ));
+ break;
+ case NF_KEY_D: // D
+ aStr = rCal.getDisplayString( CalendarDisplayCode::SHORT_DAY, nNatNum );
+ // NatNum12: support variants of preposition, suffixation or article
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ break;
+ case NF_KEY_DD: // DD
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_DAY, nNatNum ));
+ break;
+ case NF_KEY_DDD: // DDD
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ aStr = rCal.getDisplayString( CalendarDisplayCode::SHORT_DAY_NAME, nNatNum );
+ // NatNum12: support at least capitalize, upper, lower, title
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_DDDD: // DDDD
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ aStr = rCal.getDisplayString( CalendarDisplayCode::LONG_DAY_NAME, nNatNum );
+ // NatNum12: support variants of preposition, suffixation or article
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_YY: // YY
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ // Prepend a minus sign if Gregorian BCE and era is not displayed.
+ if (lcl_isSignedYear( rCal, NumFor[nIx] ))
+ {
+ sBuff.append('-');
+ }
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_YEAR, nNatNum ));
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_YYYY: // YYYY
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ // Prepend a minus sign if Gregorian BCE and era is not displayed.
+ if (lcl_isSignedYear( rCal, NumFor[nIx] ))
+ {
+ sBuff.append('-');
+ }
+ aStr = rCal.getDisplayString( CalendarDisplayCode::LONG_YEAR, nNatNum );
+ if (aStr.getLength() < 4 && !lcl_hasEra(NumFor[nIx]))
+ {
+ using namespace comphelper::string;
+ // Ensure that year consists of at least 4 digits, so it
+ // can be distinguished from 2 digits display and edited
+ // without suddenly being hit by the 2-digit year magic.
+ OUStringBuffer aBuf;
+ padToLength(aBuf, 4 - aStr.getLength(), '0');
+ impTransliterate(aBuf, NumFor[nIx].GetNatNum());
+ aBuf.append(aStr);
+ aStr = aBuf.makeStringAndClear();
+ }
+ // NatNum12: support variants of preposition, suffixation or article
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_EC: // E
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_YEAR, nNatNum ));
+ break;
+ case NF_KEY_EEC: // EE
+ case NF_KEY_R: // R
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_YEAR, nNatNum ));
+ break;
+ case NF_KEY_NN: // NN
+ case NF_KEY_AAA: // AAA
+ aStr = rCal.getDisplayString( CalendarDisplayCode::SHORT_DAY_NAME, nNatNum );
+ // NatNum12: support at least capitalize, upper, lower, title
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ break;
+ case NF_KEY_NNN: // NNN
+ case NF_KEY_AAAA: // AAAA
+ aStr = rCal.getDisplayString( CalendarDisplayCode::LONG_DAY_NAME, nNatNum );
+ // NatNum12: support variants of preposition, suffixation or article
+ if ( bUseSpellout )
+ {
+ aStr = impTransliterate(aStr, NumFor[nIx].GetNatNum(), rInfo.nTypeArray[i]);
+ }
+ sBuff.append(aStr);
+ break;
+ case NF_KEY_NNNN: // NNNN
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_DAY_NAME, nNatNum ));
+ sBuff.append(rLoc().getLongDateDayOfWeekSep());
+ break;
+ case NF_KEY_WW : // WW
+ sBuff.append(ImpIntToString( nIx,
+ rCal.getValue( CalendarFieldIndex::WEEK_OF_YEAR )));
+ break;
+ case NF_KEY_G: // G
+ ImpAppendEraG(sBuff, rCal, nNatNum );
+ break;
+ case NF_KEY_GG: // GG
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_ERA, nNatNum ));
+ break;
+ case NF_KEY_GGG: // GGG
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_ERA, nNatNum ));
+ break;
+ case NF_KEY_RR: // RR => GGGEE
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_YEAR_AND_ERA, nNatNum ));
+ break;
+ }
+ }
+ if ( aOrgCalendar.getLength() )
+ {
+ rCal.loadCalendar( aOrgCalendar, rLoc().getLanguageTag().getLocale() ); // restore calendar
+ }
+ return bRes;
+}
+
+bool SvNumberformat::ImpGetDateTimeOutput(double fNumber,
+ sal_uInt16 nIx,
+ OUStringBuffer& sBuff)
+{
+ using namespace ::com::sun::star::i18n;
+ bool bRes = false;
+
+ CalendarWrapper& rCal = GetCal();
+ if (!lcl_getValidDate( rScan.GetNullDate(), rCal.getEpochStart(), fNumber))
+ {
+ sBuff = ImpSvNumberformatScan::sErrStr;
+ return false;
+ }
+
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ bool bInputLine;
+ sal_Int32 nCntPost;
+ if ( rScan.GetStandardPrec() == SvNumberFormatter::INPUTSTRING_PRECISION &&
+ 0 < rInfo.nCntPost && rInfo.nCntPost < 7 )
+ {
+ // round at 7 decimals (+5 of 86400 == 12 significant digits)
+ bInputLine = true;
+ nCntPost = 7;
+ }
+ else
+ {
+ bInputLine = false;
+ nCntPost = rInfo.nCntPost;
+ }
+ double fTime = (fNumber - floor( fNumber )) * 86400.0;
+ fTime = ::rtl::math::round( fTime, int(nCntPost) );
+ if (fTime >= 86400.0)
+ {
+ // result of fNumber==x.999999999... rounded up, use correct date/time
+ fTime -= 86400.0;
+ fNumber = floor( fNumber + 0.5) + fTime;
+ }
+ rCal.setLocalDateTime( fNumber );
+
+ int nUseMonthCase = 0; // Not decided yet
+ OUString aOrgCalendar; // empty => not changed yet
+ double fOrgDateTime(0.0);
+ bool bOtherCalendar = ImpIsOtherCalendar( NumFor[nIx] );
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ if ( ImpFallBackToGregorianCalendar( aOrgCalendar, fOrgDateTime ) )
+ {
+ bOtherCalendar = false;
+ }
+ sal_Int16 nNatNum = NumFor[nIx].GetNatNum().GetNatNum();
+
+ OUStringBuffer sSecStr;
+ sal_Int32 nSecPos = 0; // For figure by figure processing
+ sal_uInt32 nHour, nMin, nSec;
+ if (!rInfo.bThousand) // No [] format
+ {
+ sal_uInt16 nCHour, nCMinute, nCSecond;
+ double fFractionOfSecond;
+ tools::Time::GetClock( fNumber, nCHour, nCMinute, nCSecond, fFractionOfSecond, nCntPost);
+ nHour = nCHour;
+ nMin = nCMinute;
+ nSec = nCSecond;
+ nCntPost = ImpGetFractionOfSecondString( sSecStr, fFractionOfSecond, nCntPost, true, nIx,
+ (bInputLine ? rInfo.nCntPost : 0));
+ }
+ else
+ {
+ sal_uInt32 nSeconds = static_cast<sal_uInt32>(floor( fTime ));
+
+ nCntPost = ImpGetFractionOfSecondString( sSecStr, fTime - nSeconds, nCntPost, false, nIx,
+ (bInputLine ? rInfo.nCntPost : 0));
+
+ if (rInfo.nThousand == 3) // [ss]
+ {
+ nHour = 0;
+ nMin = 0;
+ nSec = nSeconds;
+ }
+ else if (rInfo.nThousand == 2) // [mm]:ss
+ {
+ nHour = 0;
+ nMin = nSeconds / 60;
+ nSec = nSeconds % 60;
+ }
+ else if (rInfo.nThousand == 1) // [hh]:mm:ss
+ {
+ nHour = nSeconds / 3600;
+ nMin = (nSeconds%3600) / 60;
+ nSec = nSeconds%60;
+ }
+ else
+ {
+ nHour = 0; // TODO What should these values be?
+ nMin = 0;
+ nSec = 0;
+ }
+ }
+ sal_Unicode cAmPm = ' '; // a or p
+ if (rInfo.nCntExp) // AM/PM
+ {
+ if (nHour == 0)
+ {
+ nHour = 12;
+ cAmPm = 'a';
+ }
+ else if (nHour < 12)
+ {
+ cAmPm = 'a';
+ }
+ else
+ {
+ cAmPm = 'p';
+ if (nHour > 12)
+ {
+ nHour -= 12;
+ }
+ }
+ }
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+ sal_Int32 nLen;
+ OUString aYear;
+ for (sal_uInt16 i = 0; i < nCnt; i++)
+ {
+ switch (rInfo.nTypeArray[i])
+ {
+ case NF_SYMBOLTYPE_CALENDAR :
+ if ( !aOrgCalendar.getLength() )
+ {
+ aOrgCalendar = rCal.getUniqueID();
+ fOrgDateTime = rCal.getDateTime();
+ }
+ rCal.loadCalendar( rInfo.sStrArray[i], rLoc().getLanguageTag().getLocale() );
+ rCal.setDateTime( fOrgDateTime );
+ ImpFallBackToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ break;
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ bRes = lcl_appendStarFillChar( sBuff, rInfo.sStrArray[i]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[i].getLength() >= 2)
+ InsertBlanks( sBuff, sBuff.getLength(), rInfo.sStrArray[i][1] );
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_CURRENCY:
+ case NF_SYMBOLTYPE_DATESEP:
+ case NF_SYMBOLTYPE_TIMESEP:
+ case NF_SYMBOLTYPE_TIME100SECSEP:
+ sBuff.append(rInfo.sStrArray[i]);
+ break;
+ case NF_SYMBOLTYPE_DIGIT:
+ nLen = ( bInputLine && i > 0 &&
+ (rInfo.nTypeArray[i-1] == NF_SYMBOLTYPE_STRING ||
+ rInfo.nTypeArray[i-1] == NF_SYMBOLTYPE_TIME100SECSEP) ?
+ nCntPost : rInfo.sStrArray[i].getLength() );
+ for (sal_Int32 j = 0; j < nLen && nSecPos < nCntPost && nSecPos < sSecStr.getLength(); ++j)
+ {
+ sBuff.append(sSecStr[ nSecPos ]);
+ nSecPos++;
+ }
+ break;
+ case NF_KEY_AMPM: // AM/PM
+ if (cAmPm == 'a')
+ {
+ sBuff.append(rCal.getDisplayName( CalendarDisplayIndex::AM_PM,
+ AmPmValue::AM, 0 ));
+ }
+ else
+ {
+ sBuff.append(rCal.getDisplayName( CalendarDisplayIndex::AM_PM,
+ AmPmValue::PM, 0 ));
+ }
+ break;
+ case NF_KEY_AP: // A/P
+ if (cAmPm == 'a')
+ {
+ sBuff.append('a');
+ }
+ else
+ {
+ sBuff.append('p');
+ }
+ break;
+ case NF_KEY_MI: // M
+ sBuff.append(ImpIntToString( nIx, nMin ));
+ break;
+ case NF_KEY_MMI: // MM
+ sBuff.append(ImpIntToString( nIx, nMin, 2 ));
+ break;
+ case NF_KEY_H: // H
+ sBuff.append(ImpIntToString( nIx, nHour ));
+ break;
+ case NF_KEY_HH: // HH
+ sBuff.append(ImpIntToString( nIx, nHour, 2 ));
+ break;
+ case NF_KEY_S: // S
+ sBuff.append(ImpIntToString( nIx, nSec ));
+ break;
+ case NF_KEY_SS: // SS
+ sBuff.append(ImpIntToString( nIx, nSec, 2 ));
+ break;
+ case NF_KEY_M: // M
+ sBuff.append(rCal.getDisplayString(
+ CalendarDisplayCode::SHORT_MONTH, nNatNum ));
+ break;
+ case NF_KEY_MM: // MM
+ sBuff.append(rCal.getDisplayString(
+ CalendarDisplayCode::LONG_MONTH, nNatNum ));
+ break;
+ case NF_KEY_MMM: // MMM
+ sBuff.append(rCal.getDisplayString( ImpUseMonthCase( nUseMonthCase, NumFor[nIx],
+ static_cast<NfKeywordIndex>(rInfo.nTypeArray[i])),
+ nNatNum));
+ break;
+ case NF_KEY_MMMM: // MMMM
+ sBuff.append(rCal.getDisplayString( ImpUseMonthCase( nUseMonthCase, NumFor[nIx],
+ static_cast<NfKeywordIndex>(rInfo.nTypeArray[i])),
+ nNatNum));
+ break;
+ case NF_KEY_MMMMM: // MMMMM
+ sBuff.append(rCal.getDisplayString( ImpUseMonthCase( nUseMonthCase, NumFor[nIx],
+ static_cast<NfKeywordIndex>(rInfo.nTypeArray[i])),
+ nNatNum));
+ break;
+ case NF_KEY_Q: // Q
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_QUARTER, nNatNum ));
+ break;
+ case NF_KEY_QQ: // QQ
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_QUARTER, nNatNum ));
+ break;
+ case NF_KEY_D: // D
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_DAY, nNatNum ));
+ break;
+ case NF_KEY_DD: // DD
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_DAY, nNatNum ));
+ break;
+ case NF_KEY_DDD: // DDD
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_DAY_NAME, nNatNum ));
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_DDDD: // DDDD
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_DAY_NAME, nNatNum ));
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_YY: // YY
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ // Prepend a minus sign if Gregorian BCE and era is not displayed.
+ if (lcl_isSignedYear( rCal, NumFor[nIx] ))
+ {
+ sBuff.append('-');
+ }
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_YEAR, nNatNum ));
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_YYYY: // YYYY
+ if ( bOtherCalendar )
+ {
+ SwitchToGregorianCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ // Prepend a minus sign if Gregorian BCE and era is not displayed.
+ if (lcl_isSignedYear( rCal, NumFor[nIx] ))
+ {
+ sBuff.append('-');
+ }
+ aYear = rCal.getDisplayString( CalendarDisplayCode::LONG_YEAR, nNatNum );
+ if (aYear.getLength() < 4 && !lcl_hasEra(NumFor[nIx]))
+ {
+ using namespace comphelper::string;
+ // Ensure that year consists of at least 4 digits, so it
+ // can be distinguished from 2 digits display and edited
+ // without suddenly being hit by the 2-digit year magic.
+ OUStringBuffer aBuf;
+ padToLength(aBuf, 4 - aYear.getLength(), '0');
+ impTransliterate(aBuf, NumFor[nIx].GetNatNum());
+ aBuf.append(aYear);
+ sBuff.append(aBuf);
+ }
+ else
+ {
+ sBuff.append(aYear);
+ }
+ if ( bOtherCalendar )
+ {
+ SwitchToOtherCalendar( aOrgCalendar, fOrgDateTime );
+ }
+ break;
+ case NF_KEY_EC: // E
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_YEAR, nNatNum ));
+ break;
+ case NF_KEY_EEC: // EE
+ case NF_KEY_R: // R
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_YEAR, nNatNum ));
+ break;
+ case NF_KEY_NN: // NN
+ case NF_KEY_AAA: // AAA
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_DAY_NAME, nNatNum ));
+ break;
+ case NF_KEY_NNN: // NNN
+ case NF_KEY_AAAA: // AAAA
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_DAY_NAME, nNatNum ));
+ break;
+ case NF_KEY_NNNN: // NNNN
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_DAY_NAME, nNatNum ));
+ sBuff.append(rLoc().getLongDateDayOfWeekSep());
+ break;
+ case NF_KEY_WW : // WW
+ sBuff.append(ImpIntToString( nIx, rCal.getValue( CalendarFieldIndex::WEEK_OF_YEAR )));
+ break;
+ case NF_KEY_G: // G
+ ImpAppendEraG( sBuff, rCal, nNatNum );
+ break;
+ case NF_KEY_GG: // GG
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::SHORT_ERA, nNatNum ));
+ break;
+ case NF_KEY_GGG: // GGG
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_ERA, nNatNum ));
+ break;
+ case NF_KEY_RR: // RR => GGGEE
+ sBuff.append(rCal.getDisplayString( CalendarDisplayCode::LONG_YEAR_AND_ERA, nNatNum ));
+ break;
+ }
+ }
+ if ( aOrgCalendar.getLength() )
+ {
+ rCal.loadCalendar( aOrgCalendar, rLoc().getLanguageTag().getLocale() ); // restore calendar
+ }
+ return bRes;
+}
+
+bool SvNumberformat::ImpGetLogicalOutput(double fNumber,
+ sal_uInt16 nIx,
+ OUStringBuffer& sStr)
+{
+ bool bRes = false;
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ const sal_uInt16 nCnt = NumFor[nIx].GetCount();
+ for (sal_uInt16 j = 0; j < nCnt; ++j)
+ {
+ switch (rInfo.nTypeArray[j])
+ {
+ case NF_KEY_BOOLEAN:
+ sStr.append( fNumber ? rScan.GetTrueString() : rScan.GetFalseString());
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ sStr.append( rInfo.sStrArray[j]);
+ break;
+ }
+ }
+ impTransliterate(sStr, NumFor[nIx].GetNatNum());
+ return bRes;
+}
+
+bool SvNumberformat::ImpGetNumberOutput(double fNumber,
+ sal_uInt16 nIx,
+ OUStringBuffer& sStr)
+{
+ bool bRes = false;
+ bool bSign;
+ if (fNumber < 0.0)
+ {
+ bSign = (nIx == 0); // Not in the ones at the back;
+ fNumber = -fNumber;
+ }
+ else
+ {
+ bSign = false;
+ if ( std::signbit( fNumber ) )
+ {
+ fNumber = -fNumber; // yes, -0.0 is possible, eliminate '-'
+ }
+ }
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ if (rInfo.eScannedType == SvNumFormatType::PERCENT)
+ {
+ if (fNumber < D_MAX_D_BY_100)
+ {
+ fNumber *= 100.0;
+ }
+ else
+ {
+ sStr = ImpSvNumberformatScan::sErrStr;
+ return false;
+ }
+ }
+ sal_uInt16 i, j;
+ sal_Int32 nDecPos = -1;
+ bool bInteger = false;
+ if ( rInfo.nThousand != FLAG_STANDARD_IN_FORMAT )
+ {
+ // Special formatting only if no GENERAL keyword in format code
+ const sal_uInt16 nThousand = rInfo.nThousand;
+ tools::Long nPrecExp;
+ for (i = 0; i < nThousand; i++)
+ {
+ if (fNumber > D_MIN_M_BY_1000)
+ {
+ fNumber /= 1000.0;
+ }
+ else
+ {
+ fNumber = 0.0;
+ }
+ }
+ if (fNumber > 0.0)
+ {
+ nPrecExp = GetPrecExp( fNumber );
+ }
+ else
+ {
+ nPrecExp = 0;
+ }
+ if (rInfo.nCntPost) // Decimal places
+ {
+ if ((rInfo.nCntPost + nPrecExp) > 15 && nPrecExp < 15)
+ {
+ sStr = ::rtl::math::doubleToUString( fNumber, rtl_math_StringFormat_F, 15-nPrecExp, '.');
+ for (tools::Long l = 15-nPrecExp; l < static_cast<tools::Long>(rInfo.nCntPost); l++)
+ {
+ sStr.append('0');
+ }
+ }
+ else
+ {
+ sStr = ::rtl::math::doubleToUString( fNumber, rtl_math_StringFormat_F, rInfo.nCntPost, '.' );
+ }
+ sStr.stripStart('0'); // Strip leading zeros
+ }
+ else if (fNumber == 0.0) // Null
+ {
+ // Nothing to be done here, keep empty string sStr,
+ // ImpNumberFillWithThousands does the rest
+ }
+ else // Integer
+ {
+ sStr = ::rtl::math::doubleToUString( fNumber, rtl_math_StringFormat_F, 0, '.');
+ sStr.stripStart('0'); // Strip leading zeros
+ }
+ nDecPos = sStr.indexOf('.' );
+ if ( nDecPos >= 0)
+ {
+ const sal_Unicode* p = sStr.getStr() + nDecPos;
+ while ( *++p == '0' )
+ ;
+ if ( !*p )
+ {
+ bInteger = true;
+ }
+ sStr.remove( nDecPos, 1 ); // Remove .
+ }
+ if (bSign && (sStr.isEmpty() || checkForAll0s(sStr))) // Only 00000
+ {
+ bSign = false; // Not -0.00
+ }
+ } // End of != FLAG_STANDARD_IN_FORMAT
+
+ // Edit backwards:
+ j = NumFor[nIx].GetCount()-1; // Last symbol
+ // Decimal places:
+ bRes |= ImpDecimalFill( sStr, fNumber, nDecPos, j, nIx, bInteger );
+ if (bSign)
+ {
+ sStr.insert(0, '-');
+ }
+ impTransliterate(sStr, NumFor[nIx].GetNatNum());
+ return bRes;
+}
+
+bool SvNumberformat::ImpDecimalFill( OUStringBuffer& sStr, // number string
+ double& rNumber, // number
+ sal_Int32 nDecPos, // decimals start
+ sal_uInt16 j, // symbol index within format code
+ sal_uInt16 nIx, // subformat index
+ bool bInteger) // is integer
+{
+ bool bRes = false;
+ bool bFilled = false; // Was filled?
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ sal_Int32 k = sStr.getLength(); // After last figure
+ // Decimal places:
+ if (rInfo.nCntPost > 0)
+ {
+ bool bTrailing = true; // Trailing zeros?
+ short nType;
+ while (j > 0 && // Backwards
+ (nType = rInfo.nTypeArray[j]) != NF_SYMBOLTYPE_DECSEP)
+ {
+ switch ( nType )
+ {
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ bRes = lcl_insertStarFillChar( sStr, k, rInfo.sStrArray[j]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[j].getLength() >= 2)
+ /*k = */ InsertBlanks(sStr, k, rInfo.sStrArray[j][1] );
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_CURRENCY:
+ case NF_SYMBOLTYPE_PERCENT:
+ sStr.insert(k, rInfo.sStrArray[j]);
+ break;
+ case NF_SYMBOLTYPE_THSEP:
+ if (rInfo.nThousand == 0)
+ {
+ sStr.insert(k, rInfo.sStrArray[j]);
+ }
+ break;
+ case NF_SYMBOLTYPE_DIGIT:
+ {
+ const OUString& rStr = rInfo.sStrArray[j];
+ const sal_Unicode* p1 = rStr.getStr();
+ const sal_Unicode* p = p1 + rStr.getLength();
+ // In case the number of decimals passed are less than the
+ // "digits" given, append trailing '0' characters, which here
+ // means insert them because literal strings may have been
+ // appended already. If they weren't to be '0' characters
+ // they'll be changed below, as if decimals with trailing zeros
+ // were passed.
+ if (nDecPos >= 0 && nDecPos <= k)
+ {
+ sal_Int32 nAppend = rStr.getLength() - (k - nDecPos);
+ while (nAppend-- > 0)
+ {
+ sStr.insert( k++, '0');
+ }
+ }
+ while (k && p1 < p--)
+ {
+ const sal_Unicode c = *p;
+ k--;
+ if ( sStr[k] != '0' )
+ {
+ bTrailing = false;
+ bFilled = true;
+ }
+ if (bTrailing)
+ {
+ if ( c == '0' )
+ {
+ bFilled = true;
+ }
+ else if ( c == '-' )
+ {
+ if ( bInteger )
+ {
+ sStr[ k ] = '-';
+ }
+ bFilled = true;
+ }
+ else if ( c == '?' )
+ {
+ sStr[ k ] = ' ';
+ bFilled = true;
+ }
+ else if ( !bFilled ) // #
+ {
+ sStr.remove(k,1);
+ }
+ }
+ } // of for
+ break;
+ } // of case digi
+ case NF_KEY_CCC: // CCC currency
+ sStr.insert(k, rScan.GetCurAbbrev());
+ break;
+ case NF_KEY_GENERAL: // Standard in the String
+ {
+ OUStringBuffer sNum;
+ ImpGetOutputStandard(rNumber, sNum);
+ sNum.stripStart('-');
+ sStr.insert(k, sNum.makeStringAndClear());
+ break;
+ }
+ default:
+ break;
+ } // of switch
+ j--;
+ } // of while
+ } // of decimal places
+
+ bRes |= ImpNumberFillWithThousands(sStr, rNumber, k, j, nIx, // Fill with . if needed
+ rInfo.nCntPre, bFilled );
+
+ return bRes;
+}
+
+bool SvNumberformat::ImpNumberFillWithThousands( OUStringBuffer& sBuff, // number string
+ double& rNumber, // number
+ sal_Int32 k, // position within string
+ sal_uInt16 j, // symbol index within format code
+ sal_uInt16 nIx, // subformat index
+ sal_Int32 nDigCnt, // count of integer digits in format
+ bool bAddDecSep) // add decimal separator if necessary
+{
+ bool bRes = false;
+ sal_Int32 nLeadingStringChars = 0; // inserted StringChars before number
+ sal_Int32 nDigitCount = 0; // count of integer digits from the right
+ bool bStop = false;
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ // no normal thousands separators if number divided by thousands
+ bool bDoThousands = (rInfo.nThousand == 0);
+ utl::DigitGroupingIterator aGrouping( GetFormatter().GetLocaleData()->getDigitGrouping());
+
+ while (!bStop) // backwards
+ {
+ if (j == 0)
+ {
+ bStop = true;
+ }
+ switch (rInfo.nTypeArray[j])
+ {
+ case NF_SYMBOLTYPE_DECSEP:
+ aGrouping.reset();
+ [[fallthrough]];
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_CURRENCY:
+ case NF_SYMBOLTYPE_PERCENT:
+ if ( rInfo.nTypeArray[j] != NF_SYMBOLTYPE_DECSEP || bAddDecSep )
+ sBuff.insert(k, rInfo.sStrArray[j]);
+ if ( k == 0 )
+ {
+ nLeadingStringChars = nLeadingStringChars + rInfo.sStrArray[j].getLength();
+ }
+ break;
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ bRes = lcl_insertStarFillChar( sBuff, k, rInfo.sStrArray[j]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[j].getLength() >= 2)
+ /*k = */ InsertBlanks(sBuff, k, rInfo.sStrArray[j][1] );
+ break;
+ case NF_SYMBOLTYPE_THSEP:
+ // #i7284# #102685# Insert separator also if number is divided
+ // by thousands and the separator is specified somewhere in
+ // between and not only at the end.
+ // #i12596# But do not insert if it's a parenthesized negative
+ // format like (#,)
+ // In fact, do not insert if divided and regex [0#,],[^0#] and
+ // no other digit symbol follows (which was already detected
+ // during scan of format code, otherwise there would be no
+ // division), else do insert. Same in ImpNumberFill() below.
+ if ( !bDoThousands && j < NumFor[nIx].GetCount()-1 )
+ {
+ bDoThousands = ((j == 0) ||
+ (rInfo.nTypeArray[j-1] != NF_SYMBOLTYPE_DIGIT &&
+ rInfo.nTypeArray[j-1] != NF_SYMBOLTYPE_THSEP) ||
+ (rInfo.nTypeArray[j+1] == NF_SYMBOLTYPE_DIGIT));
+ }
+ if ( bDoThousands )
+ {
+ if (k > 0)
+ {
+ sBuff.insert(k, rInfo.sStrArray[j]);
+ }
+ else if (nDigitCount < nDigCnt)
+ {
+ // Leading '#' displays nothing (e.g. no leading
+ // separator for numbers <1000 with #,##0 format).
+ // Leading '?' displays blank.
+ // Everything else, including nothing, displays the
+ // separator.
+ sal_Unicode cLeader = 0;
+ if (j > 0 && rInfo.nTypeArray[j-1] == NF_SYMBOLTYPE_DIGIT)
+ {
+ const OUString& rStr = rInfo.sStrArray[j-1];
+ sal_Int32 nLen = rStr.getLength();
+ if (nLen)
+ {
+ cLeader = rStr[ nLen - 1 ];
+ }
+ }
+ switch (cLeader)
+ {
+ case '#':
+ ; // nothing
+ break;
+ case '?':
+ // replace thousand separator with blank
+ sBuff.insert(k, ' ');
+ break;
+ default:
+ sBuff.insert(k, rInfo.sStrArray[j]);
+ }
+ }
+ aGrouping.advance();
+ }
+ break;
+ case NF_SYMBOLTYPE_DIGIT:
+ {
+ const OUString& rStr = rInfo.sStrArray[j];
+ const sal_Unicode* p1 = rStr.getStr();
+ const sal_Unicode* p = p1 + rStr.getLength();
+ while ( p1 < p-- )
+ {
+ nDigitCount++;
+ if (k > 0)
+ {
+ k--;
+ }
+ else
+ {
+ switch (*p)
+ {
+ case '0':
+ sBuff.insert(0, '0');
+ break;
+ case '?':
+ sBuff.insert(0, ' ');
+ break;
+ }
+ }
+ if (nDigitCount == nDigCnt && k > 0)
+ {
+ // more digits than specified
+ ImpDigitFill(sBuff, 0, k, nIx, nDigitCount, aGrouping);
+ }
+ }
+ break;
+ }
+ case NF_KEY_CCC: // CCC currency
+ sBuff.insert(k, rScan.GetCurAbbrev());
+ break;
+ case NF_KEY_GENERAL: // "General" in string
+ {
+ OUStringBuffer sNum;
+ ImpGetOutputStandard(rNumber, sNum);
+ sNum.stripStart('-');
+ sBuff.insert(k, sNum.makeStringAndClear());
+ break;
+ }
+ default:
+ break;
+ } // switch
+ j--; // next format code string
+ } // while
+
+ k = k + nLeadingStringChars; // MSC converts += to int and then warns, so ...
+ if (k > nLeadingStringChars)
+ {
+ ImpDigitFill(sBuff, nLeadingStringChars, k, nIx, nDigitCount, aGrouping);
+ }
+ return bRes;
+}
+
+void SvNumberformat::ImpDigitFill(OUStringBuffer& sStr, // number string
+ sal_Int32 nStart, // start of digits
+ sal_Int32 & k, // position within string
+ sal_uInt16 nIx, // subformat index
+ sal_Int32 & nDigitCount, // count of integer digits from the right so far
+ utl::DigitGroupingIterator & rGrouping ) // current grouping
+{
+ if (NumFor[nIx].Info().bThousand) // Only if grouping fill in separators
+ {
+ const OUString& rThousandSep = GetFormatter().GetNumThousandSep();
+ while (k > nStart)
+ {
+ if (nDigitCount == rGrouping.getPos())
+ {
+ sStr.insert( k, rThousandSep );
+ rGrouping.advance();
+ }
+ nDigitCount++;
+ k--;
+ }
+ }
+ else // simply skip
+ {
+ k = nStart;
+ }
+}
+
+bool SvNumberformat::ImpNumberFill( OUStringBuffer& sBuff, // number string
+ double& rNumber, // number for "General" format
+ sal_Int32& k, // position within string
+ sal_uInt16& j, // symbol index within format code
+ sal_uInt16 nIx, // subformat index
+ short eSymbolType, // type of stop condition
+ bool bInsertRightBlank)// insert blank on right for denominator (default = false)
+{
+ bool bRes = false;
+ bool bStop = false;
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+ // no normal thousands separators if number divided by thousands
+ bool bDoThousands = (rInfo.nThousand == 0);
+ bool bFoundNumber = false;
+ short nType;
+
+ k = sBuff.getLength(); // behind last digit
+
+ while (!bStop && (nType = rInfo.nTypeArray[j]) != eSymbolType ) // Backwards
+ {
+ switch ( nType )
+ {
+ case NF_SYMBOLTYPE_STAR:
+ if( bStarFlag )
+ {
+ if ( bFoundNumber && eSymbolType != NF_SYMBOLTYPE_EXP )
+ k = 0; // tdf#100842 jump to beginning of number before inserting something else
+ bRes = lcl_insertStarFillChar( sBuff, k, rInfo.sStrArray[j]);
+ }
+ break;
+ case NF_SYMBOLTYPE_BLANK:
+ if (rInfo.sStrArray[j].getLength() >= 2)
+ {
+ if ( bFoundNumber && eSymbolType != NF_SYMBOLTYPE_EXP )
+ k = 0; // tdf#100842 jump to beginning of number before inserting something else
+ k = InsertBlanks(sBuff, k, rInfo.sStrArray[j][1] );
+ }
+ break;
+ case NF_SYMBOLTYPE_THSEP:
+ // Same as in ImpNumberFillWithThousands() above, do not insert
+ // if divided and regex [0#,],[^0#] and no other digit symbol
+ // follows (which was already detected during scan of format
+ // code, otherwise there would be no division), else do insert.
+ if ( !bDoThousands && j < NumFor[nIx].GetCount()-1 )
+ {
+ bDoThousands = ((j == 0) ||
+ (rInfo.nTypeArray[j-1] != NF_SYMBOLTYPE_DIGIT &&
+ rInfo.nTypeArray[j-1] != NF_SYMBOLTYPE_THSEP) ||
+ (rInfo.nTypeArray[j+1] == NF_SYMBOLTYPE_DIGIT));
+ }
+ if ( bDoThousands && k > 0 )
+ {
+ sBuff.insert(k, rInfo.sStrArray[j]);
+ }
+ break;
+ case NF_SYMBOLTYPE_DIGIT:
+ {
+ bFoundNumber = true;
+ sal_uInt16 nPosInsertBlank = bInsertRightBlank ? k : 0; // left alignment of denominator
+ const OUString& rStr = rInfo.sStrArray[j];
+ const sal_Unicode* p1 = rStr.getStr();
+ const sal_Unicode* p = p1 + rStr.getLength();
+ while ( p1 < p-- )
+ {
+ if (k > 0)
+ {
+ k--;
+ }
+ else
+ {
+ switch (*p)
+ {
+ case '0':
+ sBuff.insert(0, '0');
+ break;
+ case '?':
+ sBuff.insert(nPosInsertBlank, ' ');
+ break;
+ }
+ }
+ }
+ }
+ break;
+ case NF_KEY_CCC: // CCC currency
+ sBuff.insert(k, rScan.GetCurAbbrev());
+ break;
+ case NF_KEY_GENERAL: // Standard in the String
+ {
+ OUStringBuffer sNum;
+ bFoundNumber = true;
+ ImpGetOutputStandard(rNumber, sNum);
+ sNum.stripStart('-');
+ sBuff.insert(k, sNum.makeStringAndClear());
+ }
+ break;
+ case NF_SYMBOLTYPE_FRAC_FDIV: // Do Nothing
+ if (k > 0)
+ {
+ k--;
+ }
+ break;
+
+ default:
+ if ( bFoundNumber && eSymbolType != NF_SYMBOLTYPE_EXP )
+ k = 0; // tdf#100842 jump to beginning of number before inserting something else
+ sBuff.insert(k, rInfo.sStrArray[j]);
+ break;
+ } // of switch
+ if ( j )
+ j--; // Next String
+ else
+ bStop = true;
+ } // of while
+ return bRes;
+}
+
+void SvNumberformat::GetFormatSpecialInfo(bool& bThousand,
+ bool& IsRed,
+ sal_uInt16& nPrecision,
+ sal_uInt16& nLeadingCnt) const
+{
+ // as before: take info from nNumFor=0 for whole format (for dialog etc.)
+
+ SvNumFormatType nDummyType;
+ GetNumForInfo( 0, nDummyType, bThousand, nPrecision, nLeadingCnt );
+
+ // "negative in red" is only useful for the whole format
+
+ const Color* pColor = NumFor[1].GetColor();
+ IsRed = fLimit1 == 0.0 && fLimit2 == 0.0 && pColor
+ && (*pColor == ImpSvNumberformatScan::GetRedColor());
+}
+
+void SvNumberformat::GetNumForInfo( sal_uInt16 nNumFor, SvNumFormatType& rScannedType,
+ bool& bThousand, sal_uInt16& nPrecision, sal_uInt16& nLeadingCnt ) const
+{
+ // take info from a specified sub-format (for XML export)
+
+ if ( nNumFor > 3 )
+ {
+ return; // invalid
+ }
+
+ const ImpSvNumberformatInfo& rInfo = NumFor[nNumFor].Info();
+ rScannedType = rInfo.eScannedType;
+ bThousand = rInfo.bThousand;
+ nPrecision = (rInfo.eScannedType == SvNumFormatType::FRACTION)
+ ? rInfo.nCntExp // number of denominator digits for fraction
+ : rInfo.nCntPost;
+ sal_Int32 nPosHash = 1;
+ if ( rInfo.eScannedType == SvNumFormatType::FRACTION &&
+ ( (nPosHash += GetDenominatorString(nNumFor).indexOf('#')) > 0 ) )
+ nPrecision -= nPosHash;
+ if (bStandard && rInfo.eScannedType == SvNumFormatType::NUMBER)
+ {
+ // StandardFormat
+ nLeadingCnt = 1;
+ }
+ else
+ {
+ nLeadingCnt = 0;
+ bool bStop = false;
+ sal_uInt16 i = 0;
+ const sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ while (!bStop && i < nCnt)
+ {
+ short nType = rInfo.nTypeArray[i];
+ if ( nType == NF_SYMBOLTYPE_DIGIT)
+ {
+ const sal_Unicode* p = rInfo.sStrArray[i].getStr();
+ while ( *p == '#' )
+ {
+ p++;
+ }
+ while ( *p++ == '0' )
+ {
+ nLeadingCnt++;
+ }
+ }
+ else if (nType == NF_SYMBOLTYPE_DECSEP
+ || nType == NF_SYMBOLTYPE_EXP
+ || nType == NF_SYMBOLTYPE_FRACBLANK) // Fraction: stop after integer part,
+ { // do not count '0' of fraction
+ bStop = true;
+ }
+ i++;
+ }
+ }
+}
+
+const OUString* SvNumberformat::GetNumForString( sal_uInt16 nNumFor, sal_uInt16 nPos,
+ bool bString /* = false */ ) const
+{
+ if ( nNumFor > 3 )
+ {
+ return nullptr;
+ }
+ sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ if ( !nCnt )
+ {
+ return nullptr;
+ }
+ if ( nPos == 0xFFFF )
+ {
+ nPos = nCnt - 1;
+ if ( bString )
+ { // Backwards
+ short const * pType = NumFor[nNumFor].Info().nTypeArray.data() + nPos;
+ while ( nPos > 0 && (*pType != NF_SYMBOLTYPE_STRING) &&
+ (*pType != NF_SYMBOLTYPE_CURRENCY) )
+ {
+ pType--;
+ nPos--;
+ }
+ if ( (*pType != NF_SYMBOLTYPE_STRING) && (*pType != NF_SYMBOLTYPE_CURRENCY) )
+ {
+ return nullptr;
+ }
+ }
+ }
+ else if ( nPos > nCnt - 1 )
+ {
+ return nullptr;
+ }
+ else if ( bString )
+ {
+ // forward
+ short const * pType = NumFor[nNumFor].Info().nTypeArray.data() + nPos;
+ while ( nPos < nCnt && (*pType != NF_SYMBOLTYPE_STRING) &&
+ (*pType != NF_SYMBOLTYPE_CURRENCY) )
+ {
+ pType++;
+ nPos++;
+ }
+ if ( nPos >= nCnt || ((*pType != NF_SYMBOLTYPE_STRING) &&
+ (*pType != NF_SYMBOLTYPE_CURRENCY)) )
+ {
+ return nullptr;
+ }
+ }
+ return &NumFor[nNumFor].Info().sStrArray[nPos];
+}
+
+short SvNumberformat::GetNumForType( sal_uInt16 nNumFor, sal_uInt16 nPos ) const
+{
+ if ( nNumFor > 3 )
+ {
+ return 0;
+ }
+ sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ if ( !nCnt )
+ {
+ return 0;
+ }
+ if ( nPos == 0xFFFF )
+ {
+ nPos = nCnt - 1;
+ }
+ else if ( nPos > nCnt - 1 )
+ {
+ return 0;
+ }
+ return NumFor[nNumFor].Info().nTypeArray[nPos];
+}
+
+bool SvNumberformat::IsNegativeWithoutSign() const
+{
+ if ( IsSecondSubformatRealNegative() )
+ {
+ const OUString* pStr = GetNumForString( 1, 0, true );
+ if ( pStr )
+ {
+ return !HasStringNegativeSign( *pStr );
+ }
+ }
+ return false;
+}
+
+bool SvNumberformat::IsNegativeInBracket() const
+{
+ sal_uInt16 nCnt = NumFor[1].GetCount();
+ if (!nCnt)
+ {
+ return false;
+ }
+ auto& tmp = NumFor[1].Info().sStrArray;
+ return tmp[0] == "(" && tmp[nCnt-1] == ")";
+}
+
+bool SvNumberformat::HasPositiveBracketPlaceholder() const
+{
+ sal_uInt16 nCnt = NumFor[0].GetCount();
+ return NumFor[0].Info().sStrArray[nCnt-1] == "_)";
+}
+
+DateOrder SvNumberformat::GetDateOrder() const
+{
+ if ( eType & SvNumFormatType::DATE )
+ {
+ auto& rTypeArray = NumFor[0].Info().nTypeArray;
+ sal_uInt16 nCnt = NumFor[0].GetCount();
+ for ( sal_uInt16 j=0; j<nCnt; j++ )
+ {
+ switch ( rTypeArray[j] )
+ {
+ case NF_KEY_D :
+ case NF_KEY_DD :
+ return DateOrder::DMY;
+ case NF_KEY_M :
+ case NF_KEY_MM :
+ case NF_KEY_MMM :
+ case NF_KEY_MMMM :
+ case NF_KEY_MMMMM :
+ return DateOrder::MDY;
+ case NF_KEY_YY :
+ case NF_KEY_YYYY :
+ case NF_KEY_EC :
+ case NF_KEY_EEC :
+ case NF_KEY_R :
+ case NF_KEY_RR :
+ return DateOrder::YMD;
+ }
+ }
+ }
+ else
+ {
+ SAL_WARN( "svl.numbers", "SvNumberformat::GetDateOrder: no date" );
+ }
+ return rLoc().getDateOrder();
+}
+
+sal_uInt32 SvNumberformat::GetExactDateOrder() const
+{
+ sal_uInt32 nRet = 0;
+ if ( !(eType & SvNumFormatType::DATE) )
+ {
+ SAL_WARN( "svl.numbers", "SvNumberformat::GetExactDateOrder: no date" );
+ return nRet;
+ }
+ auto& rTypeArray = NumFor[0].Info().nTypeArray;
+ sal_uInt16 nCnt = NumFor[0].GetCount();
+ int nShift = 0;
+ for ( sal_uInt16 j=0; j<nCnt && nShift < 3; j++ )
+ {
+ switch ( rTypeArray[j] )
+ {
+ case NF_KEY_D :
+ case NF_KEY_DD :
+ nRet = (nRet << 8) | 'D';
+ ++nShift;
+ break;
+ case NF_KEY_M :
+ case NF_KEY_MM :
+ case NF_KEY_MMM :
+ case NF_KEY_MMMM :
+ case NF_KEY_MMMMM :
+ nRet = (nRet << 8) | 'M';
+ ++nShift;
+ break;
+ case NF_KEY_YY :
+ case NF_KEY_YYYY :
+ case NF_KEY_EC :
+ case NF_KEY_EEC :
+ case NF_KEY_R :
+ case NF_KEY_RR :
+ nRet = (nRet << 8) | 'Y';
+ ++nShift;
+ break;
+ }
+ }
+ return nRet;
+}
+
+void SvNumberformat::GetConditions( SvNumberformatLimitOps& rOper1, double& rVal1,
+ SvNumberformatLimitOps& rOper2, double& rVal2 ) const
+{
+ rOper1 = eOp1;
+ rOper2 = eOp2;
+ rVal1 = fLimit1;
+ rVal2 = fLimit2;
+}
+
+const Color* SvNumberformat::GetColor( sal_uInt16 nNumFor ) const
+{
+ if ( nNumFor > 3 )
+ {
+ return nullptr;
+ }
+ return NumFor[nNumFor].GetColor();
+}
+
+static void lcl_SvNumberformat_AddLimitStringImpl( OUString& rStr,
+ SvNumberformatLimitOps eOp,
+ double fLimit, std::u16string_view rDecSep )
+{
+ if ( eOp == NUMBERFORMAT_OP_NO )
+ return;
+
+ switch ( eOp )
+ {
+ case NUMBERFORMAT_OP_EQ :
+ rStr = "[=";
+ break;
+ case NUMBERFORMAT_OP_NE :
+ rStr = "[<>";
+ break;
+ case NUMBERFORMAT_OP_LT :
+ rStr = "[<";
+ break;
+ case NUMBERFORMAT_OP_LE :
+ rStr = "[<=";
+ break;
+ case NUMBERFORMAT_OP_GT :
+ rStr = "[>";
+ break;
+ case NUMBERFORMAT_OP_GE :
+ rStr = "[>=";
+ break;
+ default:
+ SAL_WARN( "svl.numbers", "unsupported number format" );
+ break;
+ }
+ rStr += ::rtl::math::doubleToUString( fLimit,
+ rtl_math_StringFormat_Automatic, rtl_math_DecimalPlaces_Max,
+ rDecSep[0], true);
+ rStr += "]";
+}
+
+static void lcl_insertLCID( OUStringBuffer& rFormatStr, sal_uInt32 nLCID, sal_Int32 nPosInsertLCID, bool bDBNumInserted )
+{
+ if ( nLCID == 0 )
+ return;
+ if (nPosInsertLCID == rFormatStr.getLength() && !bDBNumInserted)
+ // No format code, no locale.
+ return;
+
+ OUStringBuffer aLCIDString = OUString::number( nLCID , 16 ).toAsciiUpperCase();
+ // Search for only last DBNum which is the last element before insertion position
+ if ( bDBNumInserted && nPosInsertLCID >= 8
+ && aLCIDString.getLength() > 4
+ && rFormatStr.indexOf( "[DBNum", nPosInsertLCID-8) == nPosInsertLCID-8 )
+ { // remove DBNumX code if long LCID
+ nPosInsertLCID -= 8;
+ rFormatStr.remove( nPosInsertLCID, 8 );
+ }
+ aLCIDString.insert( 0, "[$-" );
+ aLCIDString.append( "]" );
+ rFormatStr.insert( nPosInsertLCID, aLCIDString.makeStringAndClear() );
+}
+
+/** Increment nAlphabetID for CJK numerals
+ * +1 for financial numerals [NatNum2]
+ * +2 for Arabic fullwidth numerals [NatNum3]
+ * */
+static void lcl_incrementAlphabetWithNatNum ( sal_uInt32& nAlphabetID, sal_uInt32 nNatNum )
+{
+ if ( nNatNum == 2) // financial
+ nAlphabetID += 1;
+ else if ( nNatNum == 3)
+ nAlphabetID += 2;
+ nAlphabetID = nAlphabetID << 24;
+}
+
+OUString SvNumberformat::GetMappedFormatstring( const NfKeywordTable& rKeywords,
+ const LocaleDataWrapper& rLocWrp,
+ LanguageType nOriginalLang /* =LANGUAGE_DONTKNOW */,
+ bool bSystemLanguage /* =false */ ) const
+{
+ OUStringBuffer aStr;
+ if (maLocale.meSubstitute != LocaleType::Substitute::NONE)
+ {
+ // XXX: theoretically this could clash with the first subformat's
+ // lcl_insertLCID() below, in practice as long as it is used for system
+ // time and date modifiers it shouldn't (i.e. there is no calendar or
+ // numeral specified as well).
+ aStr.append("[$-" + maLocale.generateCode() + "]");
+ }
+ bool bDefault[4];
+ // 1 subformat matches all if no condition specified,
+ bDefault[0] = ( NumFor[1].GetCount() == 0 && eOp1 == NUMBERFORMAT_OP_NO );
+ // with 2 subformats [>=0];[<0] is implied if no condition specified
+ bDefault[1] = ( !bDefault[0] && NumFor[2].GetCount() == 0 &&
+ eOp1 == NUMBERFORMAT_OP_GE && fLimit1 == 0.0 &&
+ eOp2 == NUMBERFORMAT_OP_NO && fLimit2 == 0.0 );
+ // with 3 or more subformats [>0];[<0];[=0] is implied if no condition specified,
+ // note that subformats may be empty (;;;) and NumFor[2].GetCount()>0 is not checked.
+ bDefault[2] = ( !bDefault[0] && !bDefault[1] &&
+ eOp1 == NUMBERFORMAT_OP_GT && fLimit1 == 0.0 &&
+ eOp2 == NUMBERFORMAT_OP_LT && fLimit2 == 0.0 );
+ bool bDefaults = bDefault[0] || bDefault[1] || bDefault[2];
+ // from now on bDefault[] values are used to append empty subformats at the end
+ bDefault[3] = false;
+ if ( !bDefaults )
+ {
+ // conditions specified
+ if ( eOp1 != NUMBERFORMAT_OP_NO && eOp2 == NUMBERFORMAT_OP_NO )
+ {
+ bDefault[0] = bDefault[1] = true; // [];x
+ }
+ else if ( eOp1 != NUMBERFORMAT_OP_NO && eOp2 != NUMBERFORMAT_OP_NO &&
+ NumFor[2].GetCount() == 0 )
+ {
+ bDefault[0] = bDefault[1] = bDefault[2] = bDefault[3] = true; // [];[];;
+ }
+ // nothing to do if conditions specified for every subformat
+ }
+ else if ( bDefault[0] )
+ {
+ bDefault[0] = false; // a single unconditional subformat is never delimited
+ }
+ else
+ {
+ if ( bDefault[2] && NumFor[2].GetCount() == 0 && NumFor[1].GetCount() > 0 )
+ {
+ bDefault[3] = true; // special cases x;x;; and ;x;;
+ }
+ for ( int i=0; i<3 && !bDefault[i]; ++i )
+ {
+ bDefault[i] = true;
+ }
+ }
+ int nSem = 0; // needed ';' delimiters
+ int nSub = 0; // subformats delimited so far
+ for ( int n=0; n<4; n++ )
+ {
+ if ( n > 0 && NumFor[n].Info().eScannedType != SvNumFormatType::UNDEFINED )
+ {
+ nSem++;
+ }
+ OUString aPrefix;
+
+ if ( !bDefaults )
+ {
+ switch ( n )
+ {
+ case 0 :
+ lcl_SvNumberformat_AddLimitStringImpl( aPrefix, eOp1,
+ fLimit1, rLocWrp.getNumDecimalSep() );
+ break;
+ case 1 :
+ lcl_SvNumberformat_AddLimitStringImpl( aPrefix, eOp2,
+ fLimit2, rLocWrp.getNumDecimalSep() );
+ break;
+ }
+ }
+
+ const OUString& rColorName = NumFor[n].GetColorName();
+ if ( !rColorName.isEmpty() )
+ {
+ const NfKeywordTable & rKey = rScan.GetKeywords();
+ for ( int j = NF_KEY_FIRSTCOLOR; j <= NF_KEY_LASTCOLOR; j++ )
+ {
+ if ( rKey[j] == rColorName )
+ {
+ aPrefix += "[" + rKeywords[j] + "]";
+ break; // for
+ }
+ }
+ }
+
+ SvNumberNatNum aNatNum = NumFor[n].GetNatNum();
+ bool bDBNumInserted = false;
+ if (aNatNum.IsComplete() && (aNatNum.GetDBNum() > 0 || nOriginalLang != LANGUAGE_DONTKNOW))
+ { // GetFormatStringForExcel() may have changed language to en_US
+ if (aNatNum.GetLang() == LANGUAGE_ENGLISH_US && nOriginalLang != LANGUAGE_DONTKNOW)
+ aNatNum.SetLang( nOriginalLang );
+ if ( aNatNum.GetDBNum() > 0 )
+ {
+ aPrefix += "[DBNum" + OUString::number( aNatNum.GetDBNum() ) + "]";
+ bDBNumInserted = true;
+ }
+ }
+
+ sal_uInt16 nCnt = NumFor[n].GetCount();
+ if ( nSem && (nCnt || !aPrefix.isEmpty()) )
+ {
+ for ( ; nSem; --nSem )
+ {
+ aStr.append( ';' );
+ }
+ for ( ; nSub <= n; ++nSub )
+ {
+ bDefault[nSub] = false;
+ }
+ }
+
+ if ( !aPrefix.isEmpty() )
+ {
+ aStr.append( aPrefix );
+ }
+ sal_Int32 nPosHaveLCID = -1;
+ sal_Int32 nPosInsertLCID = aStr.getLength();
+ sal_uInt32 nCalendarID = 0x0000000; // Excel ID of calendar used in sub-format see tdf#36038
+ constexpr sal_uInt32 kCalGengou = 0x0030000;
+ if ( nCnt )
+ {
+ auto& rTypeArray = NumFor[n].Info().nTypeArray;
+ auto& rStrArray = NumFor[n].Info().sStrArray;
+ for ( sal_uInt16 j=0; j<nCnt; j++ )
+ {
+ if ( 0 <= rTypeArray[j] && rTypeArray[j] < NF_KEYWORD_ENTRIES_COUNT )
+ {
+ aStr.append( rKeywords[rTypeArray[j]] );
+ if( NF_KEY_NNNN == rTypeArray[j] )
+ {
+ aStr.append( rLocWrp.getLongDateDayOfWeekSep() );
+ }
+ switch (rTypeArray[j])
+ {
+ case NF_KEY_EC:
+ case NF_KEY_EEC:
+ case NF_KEY_R:
+ case NF_KEY_RR:
+ // Implicit secondary (non-gregorian) calendar.
+ // Currently only for ja-JP.
+ /* TODO: same for all locales in
+ * LocaleDataWrapper::doesSecondaryCalendarUseEC() ?
+ * Should split the locales off that then. */
+ if (!nCalendarID)
+ {
+ const LanguageType nLang = MsLangId::getRealLanguage( nOriginalLang);
+ if (nLang == LANGUAGE_JAPANESE)
+ nCalendarID = kCalGengou;
+ }
+ break;
+ default:
+ ; // nothing
+ }
+ }
+ else
+ {
+ switch ( rTypeArray[j] )
+ {
+ case NF_SYMBOLTYPE_DECSEP :
+ aStr.append( rLocWrp.getNumDecimalSep() );
+ break;
+ case NF_SYMBOLTYPE_THSEP :
+ aStr.append( rLocWrp.getNumThousandSep() );
+ break;
+ case NF_SYMBOLTYPE_EXP :
+ aStr.append( rKeywords[NF_KEY_E] );
+ if ( rStrArray[j].getLength() > 1 && rStrArray[j][1] == '+' )
+ aStr.append( "+" );
+ else
+ // tdf#102370: Excel code for exponent without sign
+ aStr.append( "-" );
+ break;
+ case NF_SYMBOLTYPE_DATESEP :
+ aStr.append( rLocWrp.getDateSep() );
+ break;
+ case NF_SYMBOLTYPE_TIMESEP :
+ aStr.append( rLocWrp.getTimeSep() );
+ break;
+ case NF_SYMBOLTYPE_TIME100SECSEP :
+ aStr.append( rLocWrp.getTime100SecSep() );
+ break;
+ case NF_SYMBOLTYPE_FRACBLANK :
+ case NF_SYMBOLTYPE_STRING :
+ if ( rStrArray[j].getLength() == 1 )
+ {
+ if ( rTypeArray[j] == NF_SYMBOLTYPE_STRING )
+ aStr.append( '\\' );
+ aStr.append( rStrArray[j] );
+ }
+ else
+ {
+ aStr.append( '"' );
+ aStr.append( rStrArray[j] );
+ aStr.append( '"' );
+ }
+ break;
+ case NF_SYMBOLTYPE_CALDEL :
+ if (j + 1 >= nCnt)
+ break;
+ if ( rStrArray[j+1] == "gengou" )
+ {
+ nCalendarID = kCalGengou;
+ }
+ else if ( rStrArray[j+1] == "hijri" )
+ {
+ nCalendarID = 0x0060000;
+ }
+ else if ( rStrArray[j+1] == "buddhist" )
+ {
+ nCalendarID = 0x0070000;
+ }
+ else if ( rStrArray[j+1] == "jewish" )
+ {
+ nCalendarID = 0x0080000;
+ }
+ // Other calendars (see tdf#36038) not corresponding between LibO and XL.
+ // However, skip any calendar modifier and don't write
+ // as format code (if not as literal string).
+ j += 2;
+ break;
+ case NF_SYMBOLTYPE_CURREXT :
+ nPosHaveLCID = aStr.getLength();
+ aStr.append( rStrArray[j] );
+ break;
+ default:
+ aStr.append( rStrArray[j] );
+ }
+ }
+ }
+ }
+ sal_uInt32 nAlphabetID = 0x0000000; // Excel ID of alphabet used for numerals see tdf#36038
+ LanguageType nLanguageID = LANGUAGE_SYSTEM;
+ if ( aNatNum.IsComplete() )
+ {
+ nLanguageID = MsLangId::getRealLanguage( aNatNum.GetLang());
+ if ( aNatNum.GetNatNum() == 0 )
+ {
+ nAlphabetID = 0x01000000; // Arabic-european numerals
+ }
+ else if ( nCalendarID > 0 || aNatNum.GetDBNum() == 0 || aNatNum.GetDBNum() == aNatNum.GetNatNum() )
+ { // if no DBNum code then use long LCID
+ // if DBNum value != NatNum value, use DBNum and not extended LCID
+ // if calendar, then DBNum will be removed
+ LanguageType pri = primary(nLanguageID);
+ if ( pri == LANGUAGE_ARABIC_PRIMARY_ONLY )
+ nAlphabetID = 0x02000000; // Arabic-indic numerals
+ else if ( pri == primary(LANGUAGE_FARSI) )
+ nAlphabetID = 0x03000000; // Farsi numerals
+ else if ( pri.anyOf(
+ primary(LANGUAGE_HINDI),
+ primary(LANGUAGE_MARATHI),
+ primary(LANGUAGE_NEPALI) ))
+ nAlphabetID = 0x04000000; // Devanagari numerals
+ else if ( pri == primary(LANGUAGE_BENGALI) )
+ nAlphabetID = 0x05000000; // Bengali numerals
+ else if ( pri == primary(LANGUAGE_PUNJABI) )
+ {
+ if ( nLanguageID == LANGUAGE_PUNJABI_ARABIC_LSO )
+ nAlphabetID = 0x02000000; // Arabic-indic numerals
+ else
+ nAlphabetID = 0x06000000; // Punjabi numerals
+ }
+ else if ( pri == primary(LANGUAGE_GUJARATI) )
+ nAlphabetID = 0x07000000; // Gujarati numerals
+ else if ( pri == primary(LANGUAGE_ODIA))
+ nAlphabetID = 0x08000000; // Odia (Oriya) numerals
+ else if ( pri == primary(LANGUAGE_TAMIL))
+ nAlphabetID = 0x09000000; // Tamil numerals
+ else if ( pri == primary(LANGUAGE_TELUGU))
+ nAlphabetID = 0x0A000000; // Telugu numerals
+ else if ( pri == primary(LANGUAGE_KANNADA))
+ nAlphabetID = 0x0B000000; // Kannada numerals
+ else if ( pri == primary(LANGUAGE_MALAYALAM))
+ nAlphabetID = 0x0C000000; // Malayalam numerals
+ else if ( pri == primary(LANGUAGE_THAI))
+ {
+ // The Thai T NatNum modifier during Xcl export.
+ if ( rKeywords[NF_KEY_THAI_T] == "T" )
+ nAlphabetID = 0x0D000000; // Thai numerals
+ }
+ else if ( pri == primary(LANGUAGE_LAO))
+ nAlphabetID = 0x0E000000; // Lao numerals
+ else if ( pri == primary(LANGUAGE_TIBETAN))
+ nAlphabetID = 0x0F000000; // Tibetan numerals
+ else if ( pri == primary(LANGUAGE_BURMESE))
+ nAlphabetID = 0x10000000; // Burmese numerals
+ else if ( pri == primary(LANGUAGE_TIGRIGNA_ETHIOPIA))
+ nAlphabetID = 0x11000000; // Tigrigna numerals
+ else if ( pri == primary(LANGUAGE_KHMER))
+ nAlphabetID = 0x12000000; // Khmer numerals
+ else if ( pri == primary(LANGUAGE_MONGOLIAN_MONGOLIAN_MONGOLIA))
+ {
+ if ( nLanguageID != LANGUAGE_MONGOLIAN_CYRILLIC_MONGOLIA
+ && nLanguageID != LANGUAGE_MONGOLIAN_CYRILLIC_LSO )
+ nAlphabetID = 0x13000000; // Mongolian numerals
+ }
+ // CJK numerals
+ else if ( pri == primary(LANGUAGE_JAPANESE))
+ {
+ nAlphabetID = 0x1B;
+ lcl_incrementAlphabetWithNatNum ( nAlphabetID, aNatNum.GetNatNum() );
+ }
+ else if ( pri == primary(LANGUAGE_CHINESE))
+ {
+ if ( nLanguageID == LANGUAGE_CHINESE_TRADITIONAL
+ || nLanguageID == LANGUAGE_CHINESE_HONGKONG
+ || nLanguageID == LANGUAGE_CHINESE_MACAU )
+ {
+ nAlphabetID = 0x21;
+ lcl_incrementAlphabetWithNatNum ( nAlphabetID, aNatNum.GetNatNum() );
+ }
+ else // LANGUAGE_CHINESE_SIMPLIFIED
+ {
+ nAlphabetID = 0x1E;
+ lcl_incrementAlphabetWithNatNum ( nAlphabetID, aNatNum.GetNatNum() );
+ }
+ }
+ else if ( pri == primary(LANGUAGE_KOREAN))
+ {
+ if ( aNatNum.GetNatNum() == 9 ) // Hangul
+ {
+ nAlphabetID = 0x27000000;
+ }
+ else
+ {
+ nAlphabetID = 0x24;
+ lcl_incrementAlphabetWithNatNum ( nAlphabetID, aNatNum.GetNatNum() );
+ }
+ }
+ }
+ // Add LCID to DBNum
+ if ( aNatNum.GetDBNum() > 0 && nLanguageID == LANGUAGE_SYSTEM )
+ nLanguageID = MsLangId::getRealLanguage( aNatNum.GetLang());
+ }
+ else if (nPosHaveLCID < 0)
+ {
+ // Do not insert a duplicated LCID that was already given with a
+ // currency format as [$R-1C09]
+ if (!bSystemLanguage && nOriginalLang != LANGUAGE_DONTKNOW)
+ {
+ // Explicit locale, write only to the first subformat.
+ if (n == 0)
+ nLanguageID = MsLangId::getRealLanguage( nOriginalLang);
+ }
+ else if (bSystemLanguage && maLocale.meLanguageWithoutLocaleData != LANGUAGE_DONTKNOW)
+ {
+ // Explicit locale but no locale data thus assigned to system
+ // locale, preserve for roundtrip, write only to the first
+ // subformat.
+ if (n == 0)
+ nLanguageID = maLocale.meLanguageWithoutLocaleData;
+ }
+ }
+ if ( nCalendarID > 0 )
+ { // Add alphabet and language to calendar
+ if ( nAlphabetID == 0 )
+ nAlphabetID = 0x01000000;
+ if ( nLanguageID == LANGUAGE_SYSTEM && nOriginalLang != LANGUAGE_DONTKNOW )
+ nLanguageID = nOriginalLang;
+ }
+ lcl_insertLCID( aStr, nAlphabetID + nCalendarID + static_cast<sal_uInt16>(nLanguageID), nPosInsertLCID,
+ bDBNumInserted);
+ }
+ for ( ; nSub<4 && bDefault[nSub]; ++nSub )
+ { // append empty subformats
+ aStr.append( ';' );
+ }
+ return aStr.makeStringAndClear();
+}
+
+OUString SvNumberformat::ImpGetNatNumString( const SvNumberNatNum& rNum,
+ sal_Int64 nVal, sal_uInt16 nMinDigits ) const
+{
+ OUString aStr;
+ if ( nMinDigits )
+ {
+ if ( nMinDigits == 2 )
+ {
+ // speed up the most common case
+ if ( 0 <= nVal && nVal < 10 )
+ {
+ sal_Unicode aBuf[2];
+ aBuf[0] = '0';
+ aBuf[1] = '0' + nVal;
+ aStr = OUString(aBuf, SAL_N_ELEMENTS(aBuf));
+ }
+ else
+ {
+ aStr = OUString::number( nVal );
+ }
+ }
+ else
+ {
+ OUString aValStr( OUString::number( nVal ) );
+ if ( aValStr.getLength() >= nMinDigits )
+ {
+ aStr = aValStr;
+ }
+ else
+ {
+ OUStringBuffer aBuf;
+ for(sal_Int32 index = 0; index < nMinDigits - aValStr.getLength(); ++index)
+ {
+ aBuf.append('0');
+ }
+ aBuf.append(aValStr);
+ aStr = aBuf.makeStringAndClear();
+ }
+ }
+ }
+ else
+ {
+ aStr = OUString::number( nVal );
+ }
+ return impTransliterate(aStr, rNum);
+}
+
+OUString SvNumberformat::impTransliterateImpl(const OUString& rStr,
+ const SvNumberNatNum& rNum ) const
+{
+ css::lang::Locale aLocale( LanguageTag( rNum.GetLang() ).getLocale() );
+ return GetFormatter().GetNatNum()->getNativeNumberStringParams(rStr, aLocale, rNum.GetNatNum(),
+ rNum.GetParams());
+}
+
+void SvNumberformat::impTransliterateImpl(OUStringBuffer& rStr,
+ const SvNumberNatNum& rNum ) const
+{
+ css::lang::Locale aLocale( LanguageTag( rNum.GetLang() ).getLocale() );
+
+ rStr = GetFormatter().GetNatNum()->getNativeNumberStringParams(
+ OUString::unacquired(rStr), aLocale, rNum.GetNatNum(), rNum.GetParams());
+}
+
+OUString SvNumberformat::impTransliterateImpl(const OUString& rStr,
+ const SvNumberNatNum& rNum,
+ const sal_uInt16 nDateKey) const
+{
+ // no KEYWORD=argument list in NatNum12
+ if (rNum.GetParams().indexOf('=') == -1)
+ return impTransliterateImpl( rStr, rNum);
+
+ const NfKeywordTable & rKeywords = rScan.GetKeywords();
+
+ // Format: KEYWORD=numbertext_prefix, ..., for example:
+ // [NatNum12 YYYY=title ordinal,MMMM=article, D=ordinal-number]
+ sal_Int32 nField = -1;
+ do
+ {
+ nField = rNum.GetParams().indexOf(OUStringConcatenation(rKeywords[nDateKey] + "="), ++nField);
+ }
+ while (nField != -1 && nField != 0 &&
+ (rNum.GetParams()[nField - 1] != ',' &&
+ rNum.GetParams()[nField - 1] != ' '));
+
+ // no format specified for actual keyword
+ if (nField == -1)
+ return rStr;
+
+ sal_Int32 nKeywordLen = rKeywords[nDateKey].getLength() + 1;
+ sal_Int32 nFieldEnd = rNum.GetParams().indexOf(',', nField);
+
+ if (nFieldEnd == -1)
+ nFieldEnd = rNum.GetParams().getLength();
+
+ css::lang::Locale aLocale( LanguageTag( rNum.GetLang() ).getLocale() );
+
+ return GetFormatter().GetNatNum()->getNativeNumberStringParams(
+ rStr, aLocale, rNum.GetNatNum(),
+ rNum.GetParams().copy(nField + nKeywordLen, nFieldEnd - nField - nKeywordLen));
+}
+
+void SvNumberformat::GetNatNumXml( css::i18n::NativeNumberXmlAttributes2& rAttr,
+ sal_uInt16 nNumFor ) const
+{
+ if ( nNumFor <= 3 )
+ {
+ const SvNumberNatNum& rNum = NumFor[nNumFor].GetNatNum();
+ if ( rNum.IsSet() )
+ {
+ css::lang::Locale aLocale(
+ LanguageTag( rNum.GetLang() ).getLocale() );
+
+ /* TODO: a new XNativeNumberSupplier2::convertToXmlAttributes()
+ * should rather return NativeNumberXmlAttributes2 and places
+ * adapted, and whether to fill Spellout or something different
+ * should be internal there. */
+ css::i18n::NativeNumberXmlAttributes aTmp(
+ GetFormatter().GetNatNum()->convertToXmlAttributes(
+ aLocale, rNum.GetNatNum()));
+ rAttr.Locale = aTmp.Locale;
+ rAttr.Format = aTmp.Format;
+ rAttr.Style = aTmp.Style;
+ if ( NatNumTakesParameters(rNum.GetNatNum()) )
+ {
+ // NatNum12 spell out numbers, dates and money amounts
+ rAttr.Spellout = rNum.GetParams();
+ // Mutually exclusive.
+ rAttr.Format.clear();
+ rAttr.Style.clear();
+ }
+ else
+ {
+ rAttr.Spellout.clear();
+ }
+ }
+ else
+ {
+ rAttr = css::i18n::NativeNumberXmlAttributes2();
+ }
+ }
+ else
+ {
+ rAttr = css::i18n::NativeNumberXmlAttributes2();
+ }
+}
+
+// static
+bool SvNumberformat::HasStringNegativeSign( const OUString& rStr )
+{
+ // For Sign '-' needs to be at the start or at the end of the string (blanks ignored)
+ sal_Int32 nLen = rStr.getLength();
+ if ( !nLen )
+ {
+ return false;
+ }
+ const sal_Unicode* const pBeg = rStr.getStr();
+ const sal_Unicode* const pEnd = pBeg + nLen;
+ const sal_Unicode* p = pBeg;
+ do
+ { // Start
+ if ( *p == '-' )
+ {
+ return true;
+ }
+ }
+ while ( *p == ' ' && ++p < pEnd );
+
+ p = pEnd - 1;
+
+ do
+ { // End
+ if ( *p == '-' )
+ {
+ return true;
+ }
+ }
+ while ( *p == ' ' && pBeg < --p );
+ return false;
+}
+
+// static
+bool SvNumberformat::IsInQuote( const OUString& rStr, sal_Int32 nPos,
+ sal_Unicode cQuote, sal_Unicode cEscIn, sal_Unicode cEscOut )
+{
+ sal_Int32 nLen = rStr.getLength();
+ if ( nPos >= nLen )
+ {
+ return false;
+ }
+ const sal_Unicode* p0 = rStr.getStr();
+ const sal_Unicode* p = p0;
+ const sal_Unicode* p1 = p0 + nPos;
+ bool bQuoted = false;
+ while ( p <= p1 )
+ {
+ if ( *p == cQuote )
+ {
+ if ( p == p0 )
+ {
+ bQuoted = true;
+ }
+ else if ( bQuoted )
+ {
+ if ( *(p-1) != cEscIn )
+ {
+ bQuoted = false;
+ }
+ }
+ else
+ {
+ if ( *(p-1) != cEscOut )
+ {
+ bQuoted = true;
+ }
+ }
+ }
+ p++;
+ }
+ return bQuoted;
+}
+
+// static
+sal_Int32 SvNumberformat::GetQuoteEnd( const OUString& rStr, sal_Int32 nPos,
+ sal_Unicode cQuote, sal_Unicode cEscIn )
+{
+ if ( nPos < 0 )
+ {
+ return -1;
+ }
+ sal_Int32 nLen = rStr.getLength();
+ if ( nPos >= nLen )
+ {
+ return -1;
+ }
+ if ( !IsInQuote( rStr, nPos, cQuote, cEscIn ) )
+ {
+ if ( rStr[ nPos ] == cQuote )
+ {
+ return nPos; // Closing cQuote
+ }
+ return -1;
+ }
+ const sal_Unicode* p0 = rStr.getStr();
+ const sal_Unicode* p = p0 + nPos;
+ const sal_Unicode* p1 = p0 + nLen;
+ while ( p < p1 )
+ {
+ if ( *p == cQuote && p > p0 && *(p-1) != cEscIn )
+ {
+ return sal::static_int_cast< sal_Int32 >(p - p0);
+ }
+ p++;
+ }
+ return nLen; // End of String
+}
+
+sal_uInt16 SvNumberformat::GetNumForNumberElementCount( sal_uInt16 nNumFor ) const
+{
+ if ( nNumFor < 4 )
+ {
+ sal_uInt16 nCnt = NumFor[nNumFor].GetCount();
+ return nCnt - ImpGetNumForStringElementCount( nNumFor );
+ }
+ return 0;
+}
+
+sal_uInt16 SvNumberformat::ImpGetNumForStringElementCount( sal_uInt16 nNumFor ) const
+{
+ sal_uInt16 nCnt = 0;
+ sal_uInt16 nNumForCnt = NumFor[nNumFor].GetCount();
+ auto& rTypeArray = NumFor[nNumFor].Info().nTypeArray;
+ for ( sal_uInt16 j=0; j<nNumForCnt; ++j )
+ {
+ switch ( rTypeArray[j] )
+ {
+ case NF_SYMBOLTYPE_STRING:
+ case NF_SYMBOLTYPE_CURRENCY:
+ case NF_SYMBOLTYPE_DATESEP:
+ case NF_SYMBOLTYPE_TIMESEP:
+ case NF_SYMBOLTYPE_TIME100SECSEP:
+ case NF_SYMBOLTYPE_PERCENT:
+ ++nCnt;
+ break;
+ }
+ }
+ return nCnt;
+}
+
+bool SvNumberformat::IsMinuteSecondFormat() const
+{
+ if (GetMaskedType() != SvNumFormatType::TIME)
+ return false;
+
+ constexpr sal_uInt16 k00 = 0x00; // Nada, Nilch
+ constexpr sal_uInt16 kLB = 0x01; // '[' Left Bracket
+ constexpr sal_uInt16 kRB = 0x02; // ']' Right Bracket
+ constexpr sal_uInt16 kMM = 0x04; // M or MM
+ constexpr sal_uInt16 kTS = 0x08; // Time Separator
+ constexpr sal_uInt16 kSS = 0x10; // S or SS
+#define HAS_MINUTE_SECOND(state) ((state) == (kMM|kTS|kSS) || (state) == (kLB|kMM|kRB|kTS|kSS))
+ // Also (kMM|kTS|kLB|kSS|kRB) but those are the same bits.
+
+ sal_uInt16 nState = k00;
+ bool bSep = false;
+ sal_uInt16 nNumForCnt = NumFor[0].GetCount();
+ auto const & rTypeArray = NumFor[0].Info().nTypeArray;
+ for (sal_uInt16 j=0; j < nNumForCnt; ++j)
+ {
+ switch (rTypeArray[j])
+ {
+ case NF_SYMBOLTYPE_DEL:
+ {
+ // '[' or ']' before/after MM or SS
+ const OUString& rStr = NumFor[0].Info().sStrArray[j];
+ if (rStr == "[")
+ {
+ if (nState != k00 && nState != (kMM|kTS))
+ return false;
+ nState |= kLB;
+ }
+ else if (rStr == "]")
+ {
+ if (nState != (kLB|kMM) && nState != (kMM|kTS|kLB|kSS))
+ return false;
+ nState |= kRB;
+ }
+ else
+ return false;
+ }
+ break;
+ case NF_KEY_MI:
+ case NF_KEY_MMI:
+ if (nState != k00 && nState != kLB)
+ return false;
+ nState |= kMM;
+ break;
+ case NF_SYMBOLTYPE_TIMESEP:
+ if (nState != kMM && nState != (kLB|kMM|kRB))
+ return false;
+ nState |= kTS;
+ break;
+ case NF_KEY_S:
+ case NF_KEY_SS:
+ if (nState != (kMM|kTS) && nState != (kLB|kMM|kRB|kTS) && nState != (kMM|kTS|kLB))
+ return false;
+ nState |= kSS;
+ break;
+ case NF_SYMBOLTYPE_TIME100SECSEP:
+ // Trailing fraction of seconds allowed.
+ if (!HAS_MINUTE_SECOND(nState))
+ return false;
+ bSep = true;
+ break;
+ case NF_SYMBOLTYPE_DIGIT:
+ if (!bSep)
+ return false;
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ // nothing, display literal
+ break;
+ default:
+ return false;
+ }
+ }
+ return HAS_MINUTE_SECOND(nState);
+#undef HAS_MINUTE_SECOND
+}
+
+OUString SvNumberformat::GetFormatStringForTimePrecision( int nPrecision ) const
+{
+ OUStringBuffer sString;
+ using comphelper::string::padToLength;
+
+ sal_uInt16 nNumForCnt = NumFor[0].GetCount();
+ auto const & rTypeArray = NumFor[0].Info().nTypeArray;
+ for (sal_uInt16 j=0; j < nNumForCnt; ++j)
+ {
+ switch (rTypeArray[j])
+ {
+ case NF_KEY_S :
+ case NF_KEY_SS:
+ sString.append( NumFor[0].Info().sStrArray[j] );
+ if ( j > 0 && rTypeArray[j-1] == NF_SYMBOLTYPE_DEL && j < nNumForCnt-1 )
+ {
+ j++;
+ sString.append( NumFor[0].Info().sStrArray[j] );
+ }
+ if (nPrecision > 0)
+ {
+ sString.append( rLoc().getTime100SecSep() );
+ padToLength(sString, sString.getLength() + nPrecision, '0');
+ }
+ break;
+ case NF_SYMBOLTYPE_TIME100SECSEP:
+ case NF_SYMBOLTYPE_DIGIT:
+ break;
+ case NF_SYMBOLTYPE_STRING:
+ sString.append( "\"" );
+ [[fallthrough]];
+ default:
+ sString.append( NumFor[0].Info().sStrArray[j] );
+ if (rTypeArray[j] == NF_SYMBOLTYPE_STRING)
+ {
+ sString.append( "\"" );
+ }
+ }
+ }
+
+ return sString.makeStringAndClear();
+}
+
+sal_uInt16 SvNumberformat::GetThousandDivisorPrecision( sal_uInt16 nIx ) const
+{
+ if (nIx >= 4)
+ return 0;
+
+ const ImpSvNumberformatInfo& rInfo = NumFor[nIx].Info();
+
+ if (rInfo.eScannedType != SvNumFormatType::NUMBER && rInfo.eScannedType != SvNumFormatType::CURRENCY)
+ return 0;
+
+ if (rInfo.nThousand == FLAG_STANDARD_IN_FORMAT)
+ return SvNumberFormatter::UNLIMITED_PRECISION;
+
+ return rInfo.nThousand * 3;
+}
+
+const CharClass& SvNumberformat::rChrCls() const
+{
+ return rScan.GetChrCls();
+}
+
+const LocaleDataWrapper& SvNumberformat::rLoc() const
+{
+ return rScan.GetLoc();
+}
+
+CalendarWrapper& SvNumberformat::GetCal() const
+{
+ return rScan.GetCal();
+}
+
+const SvNumberFormatter& SvNumberformat::GetFormatter() const
+{
+ return *rScan.GetNumberformatter();
+}
+
+/* vim:set shiftwidth=4 softtabstop=4 expandtab: */