summaryrefslogtreecommitdiffstats
path: root/scaddins/source/analysis/analysishelper.cxx
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--scaddins/source/analysis/analysishelper.cxx2813
1 files changed, 2813 insertions, 0 deletions
diff --git a/scaddins/source/analysis/analysishelper.cxx b/scaddins/source/analysis/analysishelper.cxx
new file mode 100644
index 000000000..2e799c9d1
--- /dev/null
+++ b/scaddins/source/analysis/analysishelper.cxx
@@ -0,0 +1,2813 @@
+/* -*- 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 <com/sun/star/util/Date.hpp>
+#include <com/sun/star/util/XNumberFormatTypes.hpp>
+#include <com/sun/star/util/NumberFormatter.hpp>
+
+#include <string.h>
+#include <stdio.h>
+#include <o3tl/any.hxx>
+#include <rtl/math.hxx>
+#include <algorithm>
+#include <memory>
+#include "analysisdefs.hxx"
+#include "analysishelper.hxx"
+#include <analysis.hrc>
+#include <strings.hrc>
+#include "deffuncname.hxx"
+
+using namespace ::com::sun::star;
+using namespace sca::analysis;
+
+#define UNIQUE false // function name does not exist in Calc
+#define DOUBLE true // function name exists in Calc
+
+#define STDPAR false // all parameters are described
+#define INTPAR true // first parameter is internal
+
+#define FUNCDATA( FUNCNAME, DBL, OPT, NUMOFPAR, CAT ) \
+ { "get" #FUNCNAME, ANALYSIS_FUNCNAME_##FUNCNAME, ANALYSIS_##FUNCNAME, DBL, OPT, ANALYSIS_DEFFUNCNAME_##FUNCNAME, NUMOFPAR, CAT, nullptr }
+
+#define FUNCDATAS( FUNCNAME, DBL, OPT, NUMOFPAR, CAT, SUFFIX ) \
+ { "get" #FUNCNAME, ANALYSIS_FUNCNAME_##FUNCNAME, ANALYSIS_##FUNCNAME, DBL, OPT, ANALYSIS_DEFFUNCNAME_##FUNCNAME, NUMOFPAR, CAT, SUFFIX }
+
+const FuncDataBase pFuncDatas[] =
+{
+ // UNIQUE or INTPAR or
+ // function name DOUBLE STDPAR # of param category
+ FUNCDATA( Workday, UNIQUE, INTPAR, 3, FDCategory::DateTime ),
+ FUNCDATA( Yearfrac, UNIQUE, INTPAR, 3, FDCategory::DateTime ),
+ FUNCDATA( Edate, UNIQUE, INTPAR, 2, FDCategory::DateTime ),
+ FUNCDATAS( Weeknum, DOUBLE, INTPAR, 2, FDCategory::DateTime, "_EXCEL2003" ),
+ FUNCDATA( Eomonth, UNIQUE, INTPAR, 2, FDCategory::DateTime ),
+ FUNCDATAS( Networkdays, DOUBLE, INTPAR, 3, FDCategory::DateTime, "_EXCEL2003" ),
+ FUNCDATA( Iseven, DOUBLE, STDPAR, 1, FDCategory::Inf ),
+ FUNCDATA( Isodd, DOUBLE, STDPAR, 1, FDCategory::Inf ),
+ FUNCDATA( Multinomial, UNIQUE, STDPAR, 1, FDCategory::Math ),
+ FUNCDATA( Seriessum, UNIQUE, STDPAR, 4, FDCategory::Math ),
+ FUNCDATA( Quotient, UNIQUE, STDPAR, 2, FDCategory::Math ),
+ FUNCDATA( Mround, UNIQUE, STDPAR, 2, FDCategory::Math ),
+ FUNCDATA( Sqrtpi, UNIQUE, STDPAR, 1, FDCategory::Math ),
+ FUNCDATA( Randbetween, UNIQUE, STDPAR, 2, FDCategory::Math ),
+ FUNCDATAS( Gcd, DOUBLE, INTPAR, 1, FDCategory::Math, "_EXCEL2003" ),
+ FUNCDATAS( Lcm, DOUBLE, INTPAR, 1, FDCategory::Math, "_EXCEL2003" ),
+ FUNCDATA( Besseli, UNIQUE, STDPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Besselj, UNIQUE, STDPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Besselk, UNIQUE, STDPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Bessely, UNIQUE, STDPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Bin2Oct, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Bin2Dec, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Bin2Hex, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Oct2Bin, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Oct2Dec, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Oct2Hex, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Dec2Bin, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Dec2Hex, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Dec2Oct, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Hex2Bin, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Hex2Dec, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Hex2Oct, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Delta, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Erf, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Erfc, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Gestep, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Factdouble, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imabs, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imaginary, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Impower, UNIQUE, STDPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Imargument, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imcos, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imdiv, UNIQUE, STDPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Imexp, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imconjugate, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imln, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imlog10, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imlog2, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Improduct, UNIQUE, INTPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Imreal, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imsin, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imsub, UNIQUE, STDPAR, 2, FDCategory::Tech ),
+ FUNCDATA( Imsqrt, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imsum, UNIQUE, INTPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imtan, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imsec, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imcsc, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imcot, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imsinh, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imcosh, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imsech, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Imcsch, UNIQUE, STDPAR, 1, FDCategory::Tech ),
+ FUNCDATA( Complex, UNIQUE, STDPAR, 3, FDCategory::Tech ),
+ FUNCDATA( Convert, UNIQUE, STDPAR, 3, FDCategory::Tech ),
+ FUNCDATA( Amordegrc, UNIQUE, INTPAR, 7, FDCategory::Finance ),
+ FUNCDATA( Amorlinc, UNIQUE, INTPAR, 7, FDCategory::Finance ),
+ FUNCDATA( Accrint, UNIQUE, INTPAR, 7, FDCategory::Finance ),
+ FUNCDATA( Accrintm, UNIQUE, INTPAR, 5, FDCategory::Finance ),
+ FUNCDATA( Received, UNIQUE, INTPAR, 5, FDCategory::Finance ),
+ FUNCDATA( Disc, UNIQUE, INTPAR, 5, FDCategory::Finance ),
+ FUNCDATA( Duration, UNIQUE, INTPAR, 6, FDCategory::Finance ),
+ FUNCDATA( Effect, DOUBLE, STDPAR, 2, FDCategory::Finance ),
+ FUNCDATA( Cumprinc, DOUBLE, STDPAR, 6, FDCategory::Finance ),
+ FUNCDATA( Cumipmt, DOUBLE, STDPAR, 6, FDCategory::Finance ),
+ FUNCDATA( Price, UNIQUE, INTPAR, 7, FDCategory::Finance ),
+ FUNCDATA( Pricedisc, UNIQUE, INTPAR, 5, FDCategory::Finance ),
+ FUNCDATA( Pricemat, UNIQUE, INTPAR, 6, FDCategory::Finance ),
+ FUNCDATA( Mduration, UNIQUE, INTPAR, 6, FDCategory::Finance ),
+ FUNCDATA( Nominal, DOUBLE, STDPAR, 2, FDCategory::Finance ),
+ FUNCDATA( Dollarfr, UNIQUE, STDPAR, 2, FDCategory::Finance ),
+ FUNCDATA( Dollarde, UNIQUE, STDPAR, 2, FDCategory::Finance ),
+ FUNCDATA( Yield, UNIQUE, INTPAR, 7, FDCategory::Finance ),
+ FUNCDATA( Yielddisc, UNIQUE, INTPAR, 5, FDCategory::Finance ),
+ FUNCDATA( Yieldmat, UNIQUE, INTPAR, 6, FDCategory::Finance ),
+ FUNCDATA( Tbilleq, UNIQUE, INTPAR, 3, FDCategory::Finance ),
+ FUNCDATA( Tbillprice, UNIQUE, INTPAR, 3, FDCategory::Finance ),
+ FUNCDATA( Tbillyield, UNIQUE, INTPAR, 3, FDCategory::Finance ),
+ FUNCDATA( Oddfprice, UNIQUE, INTPAR, 9, FDCategory::Finance ),
+ FUNCDATA( Oddfyield, UNIQUE, INTPAR, 9, FDCategory::Finance ),
+ FUNCDATA( Oddlprice, UNIQUE, INTPAR, 8, FDCategory::Finance ),
+ FUNCDATA( Oddlyield, UNIQUE, INTPAR, 8, FDCategory::Finance ),
+ FUNCDATA( Xirr, UNIQUE, INTPAR, 3, FDCategory::Finance ),
+ FUNCDATA( Xnpv, UNIQUE, STDPAR, 3, FDCategory::Finance ),
+ FUNCDATA( Intrate, UNIQUE, INTPAR, 5, FDCategory::Finance ),
+ FUNCDATA( Coupncd, UNIQUE, INTPAR, 4, FDCategory::Finance ),
+ FUNCDATA( Coupdays, UNIQUE, INTPAR, 4, FDCategory::Finance ),
+ FUNCDATA( Coupdaysnc, UNIQUE, INTPAR, 4, FDCategory::Finance ),
+ FUNCDATA( Coupdaybs, UNIQUE, INTPAR, 4, FDCategory::Finance ),
+ FUNCDATA( Couppcd, UNIQUE, INTPAR, 4, FDCategory::Finance ),
+ FUNCDATA( Coupnum, UNIQUE, INTPAR, 4, FDCategory::Finance ),
+ FUNCDATA( Fvschedule, UNIQUE, STDPAR, 2, FDCategory::Finance )
+};
+#undef FUNCDATA
+
+namespace sca::analysis {
+
+sal_uInt16 DaysInMonth( sal_uInt16 nMonth, sal_uInt16 nYear )
+{
+ if( (nMonth == 2) && IsLeapYear( nYear ) )
+ return 29;
+ static const sal_uInt16 aDaysInMonth[] = { 0, 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 };
+ return aDaysInMonth[ nMonth ];
+}
+
+
+/**
+ * Convert a date to a count of days starting from 01/01/0001
+ *
+ * The internal representation of a Date used in this Addin
+ * is the number of days between 01/01/0001 and the date
+ * this function converts a Day , Month, Year representation
+ * to this internal Date value.
+ *
+ */
+
+sal_Int32 DateToDays( sal_uInt16 nDay, sal_uInt16 nMonth, sal_uInt16 nYear )
+{
+ sal_Int32 nDays = (static_cast<sal_Int32>(nYear)-1) * 365;
+ nDays += ((nYear-1) / 4) - ((nYear-1) / 100) + ((nYear-1) / 400);
+
+ for( sal_uInt16 i = 1; i < nMonth; i++ )
+ nDays += DaysInMonth(i,nYear);
+ nDays += nDay;
+
+ return nDays;
+}
+
+
+/**
+ * Convert a count of days starting from 01/01/0001 to a date
+ *
+ * The internal representation of a Date used in this Addin
+ * is the number of days between 01/01/0001 and the date
+ * this function converts this internal Date value
+ * to a Day , Month, Year representation of a Date.
+ *
+ */
+
+void DaysToDate( sal_Int32 nDays, sal_uInt16& rDay, sal_uInt16& rMonth, sal_uInt16& rYear )
+{
+ if( nDays < 0 )
+ throw lang::IllegalArgumentException();
+
+ sal_Int32 nTempDays;
+ sal_Int32 i = 0;
+ bool bCalc;
+
+ do
+ {
+ nTempDays = nDays;
+ rYear = static_cast<sal_uInt16>((nTempDays / 365) - i);
+ nTempDays -= (static_cast<sal_Int32>(rYear) -1) * 365;
+ nTempDays -= (( rYear -1) / 4) - (( rYear -1) / 100) + ((rYear -1) / 400);
+ bCalc = false;
+ if ( nTempDays < 1 )
+ {
+ i++;
+ bCalc = true;
+ }
+ else
+ {
+ if ( nTempDays > 365 )
+ {
+ if ( (nTempDays != 366) || !IsLeapYear( rYear ) )
+ {
+ i--;
+ bCalc = true;
+ }
+ }
+ }
+ }
+ while ( bCalc );
+
+ rMonth = 1;
+ while ( nTempDays > DaysInMonth( rMonth, rYear ) )
+ {
+ nTempDays -= DaysInMonth( rMonth, rYear );
+ rMonth++;
+ }
+ rDay = static_cast<sal_uInt16>(nTempDays);
+}
+
+
+/**
+ * Get the null date used by the spreadsheet document
+ *
+ * The internal representation of a Date used in this Addin
+ * is the number of days between 01/01/0001 and the date
+ * this function returns this internal Date value for the document null date
+ *
+ */
+
+sal_Int32 GetNullDate( const uno::Reference< beans::XPropertySet >& xOpt )
+{
+ if( xOpt.is() )
+ {
+ try
+ {
+ uno::Any aAny = xOpt->getPropertyValue( "NullDate" );
+ util::Date aDate;
+ if( aAny >>= aDate )
+ return DateToDays( aDate.Day, aDate.Month, aDate.Year );
+ }
+ catch( uno::Exception& )
+ {
+ }
+ }
+
+ // no null date available -> no calculations possible
+ throw uno::RuntimeException();
+}
+
+
+sal_Int32 GetDiffDate360(
+ sal_uInt16 nDay1, sal_uInt16 nMonth1, sal_uInt16 nYear1, bool bLeapYear1,
+ sal_uInt16 nDay2, sal_uInt16 nMonth2, sal_uInt16 nYear2,
+ bool bUSAMethod )
+{
+ if( nDay1 == 31 )
+ nDay1--;
+ else if( bUSAMethod && ( nMonth1 == 2 && ( nDay1 == 29 || ( nDay1 == 28 && !bLeapYear1 ) ) ) )
+ nDay1 = 30;
+
+ if( nDay2 == 31 )
+ {
+ if( bUSAMethod && nDay1 != 30 )
+ {
+ nDay2 = 1;
+ if( nMonth2 == 12 )
+ {
+ nYear2++;
+ nMonth2 = 1;
+ }
+ else
+ nMonth2++;
+ }
+ else
+ nDay2 = 30;
+ }
+
+ return nDay2 + nMonth2 * 30 + nYear2 * 360 - nDay1 - nMonth1 * 30 - nYear1 * 360;
+}
+
+
+sal_Int32 GetDiffDate360( sal_Int32 nNullDate, sal_Int32 nDate1, sal_Int32 nDate2, bool bUSAMethod )
+{
+ nDate1 += nNullDate;
+ nDate2 += nNullDate;
+
+ sal_uInt16 nDay1, nMonth1, nYear1, nDay2, nMonth2, nYear2;
+
+ DaysToDate( nDate1, nDay1, nMonth1, nYear1 );
+ DaysToDate( nDate2, nDay2, nMonth2, nYear2 );
+
+ return GetDiffDate360( nDay1, nMonth1, nYear1, IsLeapYear( nYear1 ), nDay2, nMonth2, nYear2, bUSAMethod );
+}
+
+
+sal_Int32 GetDaysInYears( sal_uInt16 nYear1, sal_uInt16 nYear2 )
+{
+ sal_uInt16 nLeaps = 0;
+ for( sal_uInt16 n = nYear1 ; n <= nYear2 ; n++ )
+ {
+ if( IsLeapYear( n ) )
+ nLeaps++;
+ }
+
+ sal_uInt32 nSum = 1;
+ nSum += nYear2;
+ nSum -= nYear1;
+ nSum *= 365;
+ nSum += nLeaps;
+
+ return nSum;
+}
+
+
+sal_Int32 GetDiffDate( sal_Int32 nNullDate, sal_Int32 nStartDate, sal_Int32 nEndDate, sal_Int32 nMode,
+ sal_Int32* pOptDaysIn1stYear )
+{
+ bool bNeg = nStartDate > nEndDate;
+
+ if( bNeg )
+ {
+ sal_Int32 n = nEndDate;
+ nEndDate = nStartDate;
+ nStartDate = n;
+ }
+
+ sal_Int32 nRet;
+
+ switch( nMode )
+ {
+ case 0: // 0=USA (NASD) 30/360
+ case 4: // 4=Europe 30/360
+ {
+ sal_uInt16 nD1, nM1, nY1, nD2, nM2, nY2;
+
+ nStartDate += nNullDate;
+ nEndDate += nNullDate;
+
+ DaysToDate( nStartDate, nD1, nM1, nY1 );
+ DaysToDate( nEndDate, nD2, nM2, nY2 );
+
+ bool bLeap = IsLeapYear( nY1 );
+ sal_Int32 nDays, nMonths;
+
+ nMonths = nM2 - nM1;
+ nDays = nD2 - nD1;
+
+ nMonths += ( nY2 - nY1 ) * 12;
+
+ nRet = nMonths * 30 + nDays;
+ if( nMode == 0 && nM1 == 2 && nM2 != 2 && nY1 == nY2 )
+ nRet -= bLeap? 1 : 2;
+
+ if( pOptDaysIn1stYear )
+ *pOptDaysIn1stYear = 360;
+ }
+ break;
+ case 1: // 1=exact/exact
+ if( pOptDaysIn1stYear )
+ {
+ sal_uInt16 nD, nM, nY;
+
+ DaysToDate( nStartDate + nNullDate, nD, nM, nY );
+
+ *pOptDaysIn1stYear = IsLeapYear( nY )? 366 : 365;
+ }
+ nRet = nEndDate - nStartDate;
+ break;
+ case 2: // 2=exact/360
+ nRet = nEndDate - nStartDate;
+ if( pOptDaysIn1stYear )
+ *pOptDaysIn1stYear = 360;
+ break;
+ case 3: //3=exact/365
+ nRet = nEndDate - nStartDate;
+ if( pOptDaysIn1stYear )
+ *pOptDaysIn1stYear = 365;
+ break;
+ default:
+ throw lang::IllegalArgumentException();
+ }
+
+ return bNeg? -nRet : nRet;
+}
+
+
+double GetYearDiff( sal_Int32 nNullDate, sal_Int32 nStartDate, sal_Int32 nEndDate, sal_Int32 nMode )
+{
+ sal_Int32 nDays1stYear;
+ sal_Int32 nTotalDays = GetDiffDate( nNullDate, nStartDate, nEndDate, nMode, &nDays1stYear );
+
+ return double( nTotalDays ) / double( nDays1stYear );
+}
+
+
+sal_Int32 GetDaysInYear( sal_Int32 nNullDate, sal_Int32 nDate, sal_Int32 nMode )
+{
+ switch( nMode )
+ {
+ case 0: // 0=USA (NASD) 30/360
+ case 2: // 2=exact/360
+ case 4: // 4=Europe 30/360
+ return 360;
+ case 1: // 1=exact/exact
+ {
+ sal_uInt16 nD, nM, nY;
+ nDate += nNullDate;
+ DaysToDate( nDate, nD, nM, nY );
+ return IsLeapYear( nY )? 366 : 365;
+ }
+ case 3: //3=exact/365
+ return 365;
+ default:
+ throw lang::IllegalArgumentException();
+ }
+}
+
+
+// tdf69569 making code compliant with change request for ODFF1.2 par 4.11.7.7
+/**
+ * Function GetYearFrac implements YEARFRAC as defined in:
+ * Open Document Format for Office Applications version 1.2 Part 2, par. 6.10.24
+ * The calculations are defined in:
+ * Open Document Format for Office Applications version 1.2 Part 2, par. 4.11.7
+ */
+double GetYearFrac( sal_Int32 nNullDate, sal_Int32 nStartDate, sal_Int32 nEndDate, sal_Int32 nMode )
+{
+ if( nStartDate == nEndDate )
+ return 0.0; // nothing to do...
+
+ if( nStartDate > nEndDate )
+ {
+ sal_Int32 n = nEndDate;
+ nEndDate = nStartDate;
+ nStartDate = n;
+ }
+
+ sal_Int32 nDate1 = nStartDate + nNullDate;
+ sal_Int32 nDate2 = nEndDate + nNullDate;
+
+ sal_uInt16 nDay1, nDay2;
+ sal_uInt16 nMonth1, nMonth2;
+ sal_uInt16 nYear1, nYear2;
+
+ DaysToDate( nDate1, nDay1, nMonth1, nYear1 );
+ DaysToDate( nDate2, nDay2, nMonth2, nYear2 );
+
+ // calculate days between nDate1 and nDate2
+ sal_Int32 nDayDiff;
+ switch( nMode )
+ {
+ case 0: // 0=USA (NASD) 30/360
+ if ( nDay1 == 31 )
+ {
+ nDay1--;
+ }
+ if ( nDay1 == 30 && nDay2 == 31 )
+ {
+ nDay2--;
+ }
+ else
+ {
+ if ( nMonth1 == 2 && nDay1 == ( IsLeapYear( nYear1 ) ? 29 : 28 ) )
+ {
+ nDay1 = 30;
+ if ( nMonth2 == 2 && nDay2 == ( IsLeapYear( nYear2 ) ? 29 : 28 ) )
+ {
+ nDay2 = 30;
+ }
+ }
+ }
+ nDayDiff = ( nYear2 - nYear1 ) * 360 + ( nMonth2 - nMonth1 ) * 30 + ( nDay2 - nDay1 );
+ break;
+ case 1: // 1=exact/exact
+ case 2: // 2=exact/360
+ case 3: // 3=exact/365
+ nDayDiff = nDate2 - nDate1;
+ break;
+ case 4: // 4=Europe 30/360
+ if ( nDay1 == 31 )
+ {
+ nDay1--;
+ }
+ if ( nDay2 == 31 )
+ {
+ nDay2--;
+ }
+ nDayDiff = ( nYear2 - nYear1 ) * 360 + ( nMonth2 - nMonth1 ) * 30 + ( nDay2 - nDay1 );
+ break;
+ default:
+ throw lang::IllegalArgumentException();
+ }
+
+ //calculate days in year
+ double nDaysInYear;
+ switch( nMode )
+ {
+ case 0: // 0=USA (NASD) 30/360
+ case 2: // 2=exact/360
+ case 4: // 4=Europe 30/360
+ nDaysInYear = 360;
+ break;
+ case 1: // 1=exact/exact
+ {
+ const bool isYearDifferent = ( nYear1 != nYear2 );
+ // ODFv1.2 part 2 section 4.11.7.7.7
+ if ( isYearDifferent &&
+ ( ( nYear2 != nYear1 + 1 ) ||
+ ( nMonth1 < nMonth2 ) ||
+ ( nMonth1 == nMonth2 && nDay1 < nDay2 ) ) )
+ {
+ // return average of days in year between nDate1 and nDate2, inclusive
+ sal_Int32 nDayCount = 0;
+ for ( sal_uInt16 i = nYear1; i <= nYear2; i++ )
+ nDayCount += ( IsLeapYear( i ) ? 366 : 365 );
+
+ nDaysInYear = static_cast<double>(nDayCount) / static_cast<double>( nYear2 - nYear1 + 1 );
+ }
+ else
+ {
+ // as a consequence, !isYearDifferent or
+ // nYear2 == nYear + 1 and (nMonth1 > nMonth2 or
+ // (nMonth1 == nMonth2 and nDay1 >= nDay2))
+ assert( ( !isYearDifferent ||
+ ( nYear1 + 1 == nYear2 &&
+ ( nMonth1 > nMonth2 ||
+ ( nMonth1 == nMonth2 || nDay1 >= nDay2 ) ) ) ) );
+
+ // ODFv1.2 part 2 section 4.11.7.7.8 (CHANGE REQUEST PENDING, see tdf6959)
+ if ( !isYearDifferent && IsLeapYear( nYear1 ) )
+ {
+ nDaysInYear = 366;
+ }
+ else
+ {
+ // ODFv1.2 part 2 section 4.11.7.7.9/10 (CHANGE REQUEST PENDING, see tdf69569)
+ // we need to determine whether there is a 29 February
+ // between nDate1 (inclusive) and nDate2 (inclusive)
+ // the case of nYear1 == nYear2 is adequately tested in previous test
+ if( isYearDifferent &&
+ ( ( IsLeapYear( nYear1 ) &&
+ ( ( nMonth1 < 2 ) || ( ( nMonth1 == 2 ) && ( nDay1 <= 29 ) ) ) ) ||
+ ( IsLeapYear( nYear2 ) &&
+ ( nMonth2 > 2 || ( ( nMonth2 == 2 ) && ( nDay2 == 29 ) ) ) ) ) )
+ {
+ nDaysInYear = 366;
+ }
+ else
+ {
+ nDaysInYear = 365;
+ }
+ }
+ }
+ }
+ break;
+ case 3: // 3=exact/365
+ nDaysInYear = 365;
+ break;
+ // coverity[dead_error_begin] - following conditions exist to avoid compiler warning
+ default:
+ throw lang::IllegalArgumentException();
+ }
+
+ return double( nDayDiff ) / nDaysInYear;
+}
+
+double BinomialCoefficient( double n, double k )
+{
+ // This method is a copy of BinomKoeff()
+ // found in sc/source/core/tool/interpr3.cxx
+
+ double nVal = 0.0;
+ k = ::rtl::math::approxFloor(k);
+ if (n < k)
+ nVal = 0.0;
+ else if (k == 0.0)
+ nVal = 1.0;
+ else
+ {
+ nVal = n/k;
+ n--;
+ k--;
+ while (k > 0.0)
+ {
+ nVal *= n/k;
+ k--;
+ n--;
+ }
+ }
+ return nVal;
+}
+
+double GetGcd( double f1, double f2 )
+{
+ double f = fmod( f1, f2 );
+ while( f > 0.0 )
+ {
+ f1 = f2;
+ f2 = f;
+ f = fmod( f1, f2 );
+ }
+
+ return f2;
+}
+
+
+double ConvertToDec( const OUString& aStr, sal_uInt16 nBase, sal_uInt16 nCharLim )
+{
+ if ( nBase < 2 || nBase > 36 )
+ throw lang::IllegalArgumentException();
+
+ sal_uInt32 nStrLen = aStr.getLength();
+ if( nStrLen > nCharLim )
+ throw lang::IllegalArgumentException();
+ else if( !nStrLen )
+ return 0.0;
+
+ double fVal = 0.0;
+
+ const sal_Unicode* p = aStr.getStr();
+
+ sal_uInt16 nFirstDig = 0;
+ bool bFirstDig = true;
+ double fBase = nBase;
+
+ while ( *p )
+ {
+ sal_uInt16 n;
+
+ if( '0' <= *p && *p <= '9' )
+ n = *p - '0';
+ else if( 'A' <= *p && *p <= 'Z' )
+ n = 10 + ( *p - 'A' );
+ else if ( 'a' <= *p && *p <= 'z' )
+ n = 10 + ( *p - 'a' );
+ else
+ n = nBase;
+
+ if( n >= nBase )
+ throw lang::IllegalArgumentException(); // illegal char!
+
+ if( bFirstDig )
+ {
+ bFirstDig = false;
+ nFirstDig = n;
+ }
+ fVal = fVal * fBase + double( n );
+
+ p++;
+
+ }
+
+ if( nStrLen == nCharLim && !bFirstDig && (nFirstDig >= nBase / 2) )
+ { // handling negative values
+ fVal = ( pow( double( nBase ), double( nCharLim ) ) - fVal ); // complement
+ fVal *= -1.0;
+ }
+
+ return fVal;
+}
+
+
+static char GetMaxChar( sal_uInt16 nBase )
+{
+ const char* const c = "--123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
+ return c[ nBase ];
+}
+
+
+OUString ConvertFromDec( double fNum, double fMin, double fMax, sal_uInt16 nBase,
+ sal_Int32 nPlaces, sal_Int32 nMaxPlaces, bool bUsePlaces )
+{
+ fNum = ::rtl::math::approxFloor( fNum );
+ fMin = ::rtl::math::approxFloor( fMin );
+ fMax = ::rtl::math::approxFloor( fMax );
+
+ if( fNum < fMin || fNum > fMax || ( bUsePlaces && ( nPlaces <= 0 || nPlaces > nMaxPlaces ) ) )
+ throw lang::IllegalArgumentException();
+
+ sal_Int64 nNum = static_cast< sal_Int64 >( fNum );
+ bool bNeg = nNum < 0;
+ if( bNeg )
+ nNum = sal_Int64( pow( double( nBase ), double( nMaxPlaces ) ) ) + nNum;
+
+ OUString aRet(OUString::number(nNum, nBase).toAsciiUpperCase());
+
+
+ if( bUsePlaces )
+ {
+ sal_Int32 nLen = aRet.getLength();
+ if( !bNeg && nLen > nPlaces )
+ {
+ throw lang::IllegalArgumentException();
+ }
+ else if( ( bNeg && nLen < nMaxPlaces ) || ( !bNeg && nLen < nPlaces ) )
+ {
+ sal_Int32 nLeft = nPlaces - nLen;
+ std::unique_ptr<char[]> p( new char[ nLeft + 1 ] );
+ memset( p.get(), bNeg ? GetMaxChar( nBase ) : '0', nLeft );
+ p[ nLeft ] = 0x00;
+ aRet = OUString( p.get(), nLeft, RTL_TEXTENCODING_MS_1252 ) + aRet;
+ }
+ }
+
+ return aRet;
+}
+
+// implementation moved to module sal, see #i97091#
+double Erf( double x )
+{
+ return ::rtl::math::erf(x);
+}
+
+// implementation moved to module sal, see #i97091#
+double Erfc( double x )
+{
+ return ::rtl::math::erfc(x);
+}
+
+static bool IsNum( sal_Unicode c )
+{
+ return c >= '0' && c <= '9';
+}
+
+
+static bool IsComma( sal_Unicode c )
+{
+ return c == '.' || c == ',';
+}
+
+
+static bool IsExpStart( sal_Unicode c )
+{
+ return c == 'e' || c == 'E';
+}
+
+
+static bool IsImagUnit( sal_Unicode c )
+{
+ return c == 'i' || c == 'j';
+}
+
+
+static sal_uInt16 GetVal( sal_Unicode c )
+{
+ return sal_uInt16( c - '0' );
+}
+
+
+bool ParseDouble( const sal_Unicode*& rp, double& rRet )
+{
+ double fInt = 0.0;
+ double fFrac = 0.0;
+ double fMult = 0.1; // multiplier to multiply digits with, when adding fractional ones
+ sal_Int32 nExp = 0;
+ sal_Int32 nMaxExp = 307;
+ sal_uInt16 nDigCnt = 18; // max. number of digits to read in, rest doesn't matter
+
+ enum State { S_End = 0, S_Sign, S_IntStart, S_Int, S_IgnoreIntDigs, S_Frac, S_IgnoreFracDigs, S_ExpSign, S_Exp };
+
+ State eS = S_Sign;
+
+ bool bNegNum = false;
+ bool bNegExp = false;
+
+ const sal_Unicode* p = rp;
+ sal_Unicode c;
+
+ while( eS )
+ {
+ c = *p;
+ switch( eS )
+ {
+ case S_Sign:
+ if( IsNum( c ) )
+ {
+ fInt = GetVal( c );
+ nDigCnt--;
+ eS = S_Int;
+ }
+ else if( c == '-' )
+ {
+ bNegNum = true;
+ eS = S_IntStart;
+ }
+ else if( c == '+' )
+ eS = S_IntStart;
+ else if( IsComma( c ) )
+ eS = S_Frac;
+ else
+ return false;
+ break;
+ case S_IntStart:
+ if( IsNum( c ) )
+ {
+ fInt = GetVal( c );
+ nDigCnt--;
+ eS = S_Int;
+ }
+ else if( IsComma( c ) )
+ eS = S_Frac;
+ else if( IsImagUnit( c ) )
+ {
+ rRet = 0.0;
+ return true;
+ }
+ else
+ return false;
+ break;
+ case S_Int:
+ if( IsNum( c ) )
+ {
+ fInt *= 10.0;
+ fInt += double( GetVal( c ) );
+ nDigCnt--;
+ if( !nDigCnt )
+ eS = S_IgnoreIntDigs;
+ }
+ else if( IsComma( c ) )
+ eS = S_Frac;
+ else if( IsExpStart( c ) )
+ eS = S_ExpSign;
+ else
+ eS = S_End;
+ break;
+ case S_IgnoreIntDigs:
+ if( IsNum( c ) )
+ nExp++; // just multiply num with 10... ;-)
+ else if( IsComma( c ) )
+ eS = S_Frac;
+ else if( IsExpStart( c ) )
+ eS = S_ExpSign;
+ else
+ eS = S_End;
+ break;
+ case S_Frac:
+ if( IsNum( c ) )
+ {
+ fFrac += double( GetVal( c ) ) * fMult;
+ nDigCnt--;
+ if( nDigCnt )
+ fMult *= 0.1;
+ else
+ eS = S_IgnoreFracDigs;
+ }
+ else if( IsExpStart( c ) )
+ eS = S_ExpSign;
+ else
+ eS = S_End;
+ break;
+ case S_IgnoreFracDigs:
+ if( IsExpStart( c ) )
+ eS = S_ExpSign;
+ else if( !IsNum( c ) )
+ eS = S_End;
+ break;
+ case S_ExpSign:
+ if( IsNum( c ) )
+ {
+ nExp = GetVal( c );
+ eS = S_Exp;
+ }
+ else if( c == '-' )
+ {
+ bNegExp = true;
+ eS = S_Exp;
+ }
+ else if( c != '+' )
+ eS = S_End;
+ break;
+ case S_Exp:
+ if( IsNum( c ) )
+ {
+ nExp *= 10;
+ nExp += GetVal( c );
+ if( nExp > nMaxExp )
+ return false;
+ }
+ else
+ eS = S_End;
+ break;
+ // coverity[dead_error_begin] - following conditions exist to avoid compiler warning
+ case S_End:
+ break;
+ }
+
+ p++;
+ }
+
+ p--; // set pointer back to last
+ rp = p;
+
+ fInt += fFrac;
+
+ if (fInt != 0.0) // exact check; log10(0.0) may entail a pole error
+ {
+ sal_Int32 nLog10 = sal_Int32( log10( fInt ) );
+
+ if( bNegExp )
+ nExp = -nExp;
+
+ if( nLog10 + nExp > nMaxExp )
+ return false;
+
+ fInt = ::rtl::math::pow10Exp( fInt, nExp );
+ }
+
+ if( bNegNum )
+ fInt = -fInt;
+
+ rRet = fInt;
+
+ return true;
+}
+
+
+OUString GetString( double f, bool bLeadingSign, sal_uInt16 nMaxDig )
+{
+ const int nBuff = 256;
+ char aBuff[ nBuff + 1 ];
+ const char* pFormStr = bLeadingSign? "%+.*g" : "%.*g";
+ int nLen = snprintf( aBuff, nBuff, pFormStr, int( nMaxDig ), f );
+ // you never know which underlying implementation you get ...
+ aBuff[nBuff] = 0;
+ if ( nLen < 0 || nLen > nBuff )
+ nLen = strlen( aBuff );
+
+ OUString aRet( aBuff, nLen, RTL_TEXTENCODING_MS_1252 );
+
+ return aRet;
+}
+
+
+double GetAmordegrc( sal_Int32 nNullDate, double fCost, sal_Int32 nDate, sal_Int32 nFirstPer,
+ double fRestVal, double fPer, double fRate, sal_Int32 nBase )
+{
+ sal_uInt32 nPer = sal_uInt32( fPer );
+ double fUsePer = 1.0 / fRate;
+ double fAmorCoeff;
+
+ if( fUsePer < 3.0 )
+ fAmorCoeff = 1.0;
+ else if( fUsePer < 5.0 )
+ fAmorCoeff = 1.5;
+ else if( fUsePer <= 6.0 )
+ fAmorCoeff = 2.0;
+ else
+ fAmorCoeff = 2.5;
+
+ fRate *= fAmorCoeff;
+ double fNRate = ::rtl::math::round( GetYearFrac( nNullDate, nDate, nFirstPer, nBase ) * fRate * fCost );
+ fCost -= fNRate;
+ double fRest = fCost - fRestVal; // aboriginal cost - residual value - sum of all write-downs
+
+ for( sal_uInt32 n = 0 ; n < nPer ; n++ )
+ {
+ fNRate = ::rtl::math::round( fRate * fCost );
+ fRest -= fNRate;
+
+ if( fRest < 0.0 )
+ {
+ switch( nPer - n )
+ {
+ case 0:
+ case 1:
+ return ::rtl::math::round( fCost * 0.5 );
+ default:
+ return 0.0;
+ }
+ }
+
+ fCost -= fNRate;
+ }
+
+ return fNRate;
+}
+
+
+double GetAmorlinc( sal_Int32 nNullDate, double fCost, sal_Int32 nDate, sal_Int32 nFirstPer,
+ double fRestVal, double fPer, double fRate, sal_Int32 nBase )
+{
+ sal_uInt32 nPer = sal_uInt32( fPer );
+ double fOneRate = fCost * fRate;
+ double fCostDelta = fCost - fRestVal;
+ double f0Rate = GetYearFrac( nNullDate, nDate, nFirstPer, nBase ) * fRate * fCost;
+ sal_uInt32 nNumOfFullPeriods = sal_uInt32( ( fCost - fRestVal - f0Rate) / fOneRate );
+
+ double fResult = 0.0;
+ if( nPer == 0 )
+ fResult = f0Rate;
+ else if( nPer <= nNumOfFullPeriods )
+ fResult = fOneRate;
+ else if( nPer == nNumOfFullPeriods + 1 )
+ fResult = fCostDelta - fOneRate * nNumOfFullPeriods - f0Rate;
+
+ if ( fResult > 0.0 )
+ return fResult;
+ else
+ return 0.0;
+}
+
+
+double GetDuration( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, double fCoup,
+ double fYield, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ double fYearfrac = GetYearFrac( nNullDate, nSettle, nMat, nBase );
+ double fNumOfCoups = GetCoupnum( nNullDate, nSettle, nMat, nFreq, nBase );
+ double fDur = 0.0;
+ const double f100 = 100.0;
+ fCoup *= f100 / double( nFreq ); // fCoup is used as cash flow
+ fYield /= nFreq;
+ fYield += 1.0;
+
+ double nDiff = fYearfrac * nFreq - fNumOfCoups;
+
+ double t;
+
+ for( t = 1.0 ; t < fNumOfCoups ; t++ )
+ fDur += ( t + nDiff ) * fCoup / pow( fYield, t + nDiff );
+
+ fDur += ( fNumOfCoups + nDiff ) * ( fCoup + f100 ) / pow( fYield, fNumOfCoups + nDiff );
+
+ double p = 0.0;
+ for( t = 1.0 ; t < fNumOfCoups ; t++ )
+ p += fCoup / pow( fYield, t + nDiff );
+
+ p += ( fCoup + f100 ) / pow( fYield, fNumOfCoups + nDiff );
+
+ fDur /= p;
+ fDur /= double( nFreq );
+
+ return fDur;
+}
+
+
+double GetYieldmat( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nIssue,
+ double fRate, double fPrice, sal_Int32 nBase )
+{
+ double fIssMat = GetYearFrac( nNullDate, nIssue, nMat, nBase );
+ double fIssSet = GetYearFrac( nNullDate, nIssue, nSettle, nBase );
+ double fSetMat = GetYearFrac( nNullDate, nSettle, nMat, nBase );
+
+ double y = 1.0 + fIssMat * fRate;
+ y /= fPrice / 100.0 + fIssSet * fRate;
+ y--;
+ y /= fSetMat;
+
+ return y;
+}
+
+
+double GetOddfprice( sal_Int32 /*nNullDate*/, sal_Int32 /*nSettle*/, sal_Int32 /*nMat*/, sal_Int32 /*nIssue*/,
+ sal_Int32 /*nFirstCoup*/, double /*fRate*/, double /*fYield*/, double /*fRedemp*/, sal_Int32 /*nFreq*/,
+ sal_Int32 /*nBase*/ )
+{
+ // If you change this to not unconditionally throw, the
+ // SAL_WNOUNREACHABLE_CODE_PUSH/POP around the caller in
+ // financial.cxx can be removed.
+ throw uno::RuntimeException();
+}
+
+
+double getYield_( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, double fCoup, double fPrice,
+ double fRedemp, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ double fRate = fCoup;
+ double fPriceN = 0.0;
+ double fYield1 = 0.0;
+ double fYield2 = 1.0;
+ double fPrice1 = getPrice_( nNullDate, nSettle, nMat, fRate, fYield1, fRedemp, nFreq, nBase );
+ double fPrice2 = getPrice_( nNullDate, nSettle, nMat, fRate, fYield2, fRedemp, nFreq, nBase );
+ double fYieldN = ( fYield2 - fYield1 ) * 0.5;
+
+ for( sal_uInt32 nIter = 0 ; nIter < 100 && !rtl::math::approxEqual(fPriceN, fPrice) ; nIter++ )
+ {
+ fPriceN = getPrice_( nNullDate, nSettle, nMat, fRate, fYieldN, fRedemp, nFreq, nBase );
+
+ if( rtl::math::approxEqual(fPrice, fPrice1) )
+ return fYield1;
+ else if( rtl::math::approxEqual(fPrice, fPrice2) )
+ return fYield2;
+ else if( rtl::math::approxEqual(fPrice, fPriceN) )
+ return fYieldN;
+ else if( fPrice < fPrice2 )
+ {
+ fYield2 *= 2.0;
+ fPrice2 = getPrice_( nNullDate, nSettle, nMat, fRate, fYield2, fRedemp, nFreq, nBase );
+
+ fYieldN = ( fYield2 - fYield1 ) * 0.5;
+ }
+ else
+ {
+ if( fPrice < fPriceN )
+ {
+ fYield1 = fYieldN;
+ fPrice1 = fPriceN;
+ }
+ else
+ {
+ fYield2 = fYieldN;
+ fPrice2 = fPriceN;
+ }
+
+ fYieldN = fYield2 - ( fYield2 - fYield1 ) * ( ( fPrice - fPrice2 ) / ( fPrice1 - fPrice2 ) );
+ }
+ }
+
+ if( fabs( fPrice - fPriceN ) > fPrice / 100.0 )
+ throw lang::IllegalArgumentException(); // result not precise enough
+
+ return fYieldN;
+}
+
+
+double getPrice_( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, double fRate, double fYield,
+ double fRedemp, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ double fFreq = nFreq;
+
+ double fE = GetCoupdays( nNullDate, nSettle, nMat, nFreq, nBase );
+ double fDSC_E = GetCoupdaysnc( nNullDate, nSettle, nMat, nFreq, nBase ) / fE;
+ double fN = GetCoupnum( nNullDate, nSettle, nMat, nFreq, nBase );
+ double fA = GetCoupdaybs( nNullDate, nSettle, nMat, nFreq, nBase );
+
+ double fRet = fRedemp / ( pow( 1.0 + fYield / fFreq, fN - 1.0 + fDSC_E ) );
+ fRet -= 100.0 * fRate / fFreq * fA / fE;
+
+ double fT1 = 100.0 * fRate / fFreq;
+ double fT2 = 1.0 + fYield / fFreq;
+
+ for( double fK = 0.0 ; fK < fN ; fK++ )
+ fRet += fT1 / pow( fT2, fK + fDSC_E );
+
+ return fRet;
+}
+
+
+double GetOddfyield( sal_Int32 /*nNullDate*/, sal_Int32 /*nSettle*/, sal_Int32 /*nMat*/, sal_Int32 /*nIssue*/,
+ sal_Int32 /*nFirstCoup*/, double /*fRate*/, double /*fPrice*/, double /*fRedemp*/, sal_Int32 /*nFreq*/,
+ sal_Int32 /*nBase*/ )
+{
+ // If you change this to not unconditionally throw, the
+ // SAL_WNOUNREACHABLE_CODE_PUSH/POP around the caller in
+ // financial.cxx can be removed.
+ throw uno::RuntimeException();
+}
+
+
+double GetOddlprice( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nLastCoup,
+ double fRate, double fYield, double fRedemp, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ double fFreq = double( nFreq );
+ double fDCi = GetYearFrac( nNullDate, nLastCoup, nMat, nBase ) * fFreq;
+ double fDSCi = GetYearFrac( nNullDate, nSettle, nMat, nBase ) * fFreq;
+ double fAi = GetYearFrac( nNullDate, nLastCoup, nSettle, nBase ) * fFreq;
+
+ double p = fRedemp + fDCi * 100.0 * fRate / fFreq;
+ p /= fDSCi * fYield / fFreq + 1.0;
+ p -= fAi * 100.0 * fRate / fFreq;
+
+ return p;
+}
+
+
+double GetOddlyield( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nLastCoup,
+ double fRate, double fPrice, double fRedemp, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ double fFreq = double( nFreq );
+ double fDCi = GetYearFrac( nNullDate, nLastCoup, nMat, nBase ) * fFreq;
+ double fDSCi = GetYearFrac( nNullDate, nSettle, nMat, nBase ) * fFreq;
+ double fAi = GetYearFrac( nNullDate, nLastCoup, nSettle, nBase ) * fFreq;
+
+ double y = fRedemp + fDCi * 100.0 * fRate / fFreq;
+ y /= fPrice + fAi * 100.0 * fRate / fFreq;
+ y--;
+ y *= fFreq / fDSCi;
+
+ return y;
+}
+
+
+double GetPmt( double fRate, double fNper, double fPv, double fFv, sal_Int32 nPayType )
+{
+ double fPmt;
+ if( fRate == 0.0 )
+ fPmt = ( fPv + fFv ) / fNper;
+ else
+ {
+ double fTerm = pow( 1.0 + fRate, fNper );
+ if( nPayType > 0 )
+ fPmt = ( fFv * fRate / ( fTerm - 1.0 ) + fPv * fRate / ( 1.0 - 1.0 / fTerm ) ) / ( 1.0 + fRate );
+ else
+ fPmt = fFv * fRate / ( fTerm - 1.0 ) + fPv * fRate / ( 1.0 - 1.0 / fTerm );
+ }
+
+ return -fPmt;
+}
+
+
+double GetFv( double fRate, double fNper, double fPmt, double fPv, sal_Int32 nPayType )
+{
+ double fFv;
+ if( fRate == 0.0 )
+ fFv = fPv + fPmt * fNper;
+ else
+ {
+ double fTerm = pow( 1.0 + fRate, fNper );
+ if( nPayType > 0 )
+ fFv = fPv * fTerm + fPmt * ( 1.0 + fRate ) * ( fTerm - 1.0 ) / fRate;
+ else
+ fFv = fPv * fTerm + fPmt * ( fTerm - 1.0 ) / fRate;
+ }
+
+ return -fFv;
+}
+
+// financial functions COUP***
+
+// COUPPCD: find last coupon date before settlement (can be equal to settlement)
+/// @throws css::lang::IllegalArgumentException
+static void lcl_GetCouppcd( ScaDate& rDate, const ScaDate& rSettle, const ScaDate& rMat, sal_Int32 nFreq )
+{
+ rDate = rMat;
+ rDate.setYear( rSettle.getYear() );
+ if( rDate < rSettle )
+ rDate.addYears( 1 );
+ while( rDate > rSettle )
+ rDate.addMonths( -12 / nFreq );
+}
+
+double GetCouppcd( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ if( nSettle >= nMat || CHK_Freq )
+ throw lang::IllegalArgumentException();
+
+ ScaDate aDate;
+ lcl_GetCouppcd( aDate, ScaDate( nNullDate, nSettle, nBase ), ScaDate( nNullDate, nMat, nBase ), nFreq );
+ return aDate.getDate( nNullDate );
+}
+
+// COUPNCD: find first coupon date after settlement (is never equal to settlement)
+/// @throws css::lang::IllegalArgumentException
+static void lcl_GetCoupncd( ScaDate& rDate, const ScaDate& rSettle, const ScaDate& rMat, sal_Int32 nFreq )
+{
+ rDate = rMat;
+ rDate.setYear( rSettle.getYear() );
+ if( rDate > rSettle )
+ rDate.addYears( -1 );
+ while( rDate <= rSettle )
+ rDate.addMonths( 12 / nFreq );
+}
+
+double GetCoupncd( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ if( nSettle >= nMat || CHK_Freq )
+ throw lang::IllegalArgumentException();
+
+ ScaDate aDate;
+ lcl_GetCoupncd( aDate, ScaDate( nNullDate, nSettle, nBase ), ScaDate( nNullDate, nMat, nBase ), nFreq );
+ return aDate.getDate( nNullDate );
+}
+
+// COUPDAYBS: get day count: coupon date before settlement <-> settlement
+double GetCoupdaybs( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ if( nSettle >= nMat || CHK_Freq )
+ throw lang::IllegalArgumentException();
+
+ ScaDate aSettle( nNullDate, nSettle, nBase );
+ ScaDate aDate;
+ lcl_GetCouppcd( aDate, aSettle, ScaDate( nNullDate, nMat, nBase ), nFreq );
+ return ScaDate::getDiff( aDate, aSettle );
+}
+
+// COUPDAYSNC: get day count: settlement <-> coupon date after settlement
+double GetCoupdaysnc( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ if( nSettle >= nMat || CHK_Freq )
+ throw lang::IllegalArgumentException();
+
+ if( (nBase != 0) && (nBase != 4) )
+ {
+ ScaDate aSettle( nNullDate, nSettle, nBase );
+ ScaDate aDate;
+ lcl_GetCoupncd( aDate, aSettle, ScaDate( nNullDate, nMat, nBase ), nFreq );
+ return ScaDate::getDiff( aSettle, aDate );
+ }
+ return GetCoupdays( nNullDate, nSettle, nMat, nFreq, nBase ) - GetCoupdaybs( nNullDate, nSettle, nMat, nFreq, nBase );
+}
+
+// COUPDAYS: get day count: coupon date before settlement <-> coupon date after settlement
+double GetCoupdays( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ if( nSettle >= nMat || CHK_Freq )
+ throw lang::IllegalArgumentException();
+
+ if( nBase == 1 )
+ {
+ ScaDate aDate;
+ lcl_GetCouppcd( aDate, ScaDate( nNullDate, nSettle, nBase ), ScaDate( nNullDate, nMat, nBase ), nFreq );
+ ScaDate aNextDate( aDate );
+ aNextDate.addMonths( 12 / nFreq );
+ return ScaDate::getDiff( aDate, aNextDate );
+ }
+ return static_cast< double >( GetDaysInYear( 0, 0, nBase ) ) / nFreq;
+}
+
+// COUPNUM: get count of coupon dates
+double GetCoupnum( sal_Int32 nNullDate, sal_Int32 nSettle, sal_Int32 nMat, sal_Int32 nFreq, sal_Int32 nBase )
+{
+ if( nSettle >= nMat || CHK_Freq )
+ throw lang::IllegalArgumentException();
+
+ ScaDate aMat( nNullDate, nMat, nBase );
+ ScaDate aDate;
+ lcl_GetCouppcd( aDate, ScaDate( nNullDate, nSettle, nBase ), aMat, nFreq );
+ sal_uInt16 nMonths = (aMat.getYear() - aDate.getYear()) * 12 + aMat.getMonth() - aDate.getMonth();
+ return static_cast< double >( nMonths * nFreq / 12 );
+}
+
+FuncData::FuncData(const FuncDataBase& r) :
+ aIntName( OUString::createFromAscii( r.pIntName ) ),
+ pUINameID( r.pUINameID ),
+ pDescrID( r.pDescrID ),
+ bDouble( r.bDouble ),
+ bWithOpt( r.bWithOpt ),
+ nParam( r.nNumOfParams ),
+ eCat( r.eCat )
+{
+ if (r.pSuffix)
+ aSuffix = OUString::createFromAscii(r.pSuffix);
+
+ aCompList.resize(2);
+ aCompList[0] = OUString(r.pCompListID[0], strlen(r.pCompListID[0]), RTL_TEXTENCODING_UTF8);
+ aCompList[1] = OUString(r.pCompListID[1], strlen(r.pCompListID[1]), RTL_TEXTENCODING_UTF8);
+}
+
+sal_uInt16 FuncData::GetStrIndex( sal_uInt16 nParamNum ) const
+{
+ if( !bWithOpt )
+ nParamNum++;
+
+ if( nParamNum > nParam )
+ return nParam * 2;
+ else
+ return nParamNum * 2;
+}
+
+void InitFuncDataList(FuncDataList& rList)
+{
+ for(const auto & rFuncData : pFuncDatas)
+ rList.push_back(FuncData(rFuncData));
+}
+
+SortedIndividualInt32List::SortedIndividualInt32List()
+{
+}
+
+
+SortedIndividualInt32List::~SortedIndividualInt32List()
+{
+}
+
+
+void SortedIndividualInt32List::Insert( sal_Int32 nDay )
+{
+ sal_uInt32 nIndex = Count();
+ while( nIndex )
+ {
+ nIndex--;
+ sal_Int32 nRef = Get( nIndex );
+ if( nDay == nRef )
+ return;
+ else if( nDay > nRef )
+ {
+ maVector.insert( maVector.begin() + nIndex + 1, nDay );
+ return;
+ }
+ }
+ maVector.insert( maVector.begin(), nDay );
+}
+
+
+void SortedIndividualInt32List::Insert( sal_Int32 nDay, sal_Int32 nNullDate, bool bInsertOnWeekend )
+{
+ if( !nDay )
+ return;
+
+ nDay += nNullDate;
+ if( bInsertOnWeekend || (GetDayOfWeek( nDay ) < 5) )
+ Insert( nDay );
+}
+
+
+void SortedIndividualInt32List::Insert(
+ double fDay, sal_Int32 nNullDate, bool bInsertOnWeekend )
+{
+ if( (fDay < -2147483648.0) || (fDay > 2147483649.0) )
+ throw lang::IllegalArgumentException();
+ Insert( static_cast< sal_Int32 >( fDay ), nNullDate, bInsertOnWeekend );
+}
+
+
+bool SortedIndividualInt32List::Find( sal_Int32 nVal ) const
+{
+ sal_uInt32 nE = Count();
+
+ if( !nE || nVal < Get( 0 ) || nVal > Get( nE - 1 ) )
+ return false;
+
+ // linear search
+
+ for( sal_uInt32 n = 0 ; n < nE ; n++ )
+ {
+ sal_Int32 nRef = Get( n );
+
+ if( nRef == nVal )
+ return true;
+ else if( nRef > nVal )
+ return false;
+ }
+ return false;
+}
+
+
+void SortedIndividualInt32List::InsertHolidayList(
+ const ScaAnyConverter& rAnyConv,
+ const uno::Any& rHolAny,
+ sal_Int32 nNullDate,
+ bool bInsertOnWeekend )
+{
+ double fDay;
+ if( rAnyConv.getDouble( fDay, rHolAny ) )
+ Insert( fDay, nNullDate, bInsertOnWeekend );
+}
+
+
+void SortedIndividualInt32List::InsertHolidayList(
+ ScaAnyConverter& rAnyConv,
+ const uno::Reference< beans::XPropertySet >& xOptions,
+ const uno::Any& rHolAny,
+ sal_Int32 nNullDate )
+{
+ rAnyConv.init( xOptions );
+ if( rHolAny.getValueTypeClass() == uno::TypeClass_SEQUENCE )
+ {
+ uno::Sequence< uno::Sequence< uno::Any > > aAnySeq;
+ if( !(rHolAny >>= aAnySeq) )
+ throw lang::IllegalArgumentException();
+
+ for( const uno::Sequence< uno::Any >& rSubSeq : std::as_const(aAnySeq) )
+ {
+ for( const uno::Any& rAny : rSubSeq )
+ InsertHolidayList( rAnyConv, rAny, nNullDate, false/*bInsertOnWeekend*/ );
+ }
+ }
+ else
+ InsertHolidayList( rAnyConv, rHolAny, nNullDate, false/*bInsertOnWeekend*/ );
+}
+
+
+void ScaDoubleList::Append(
+ const uno::Sequence< uno::Sequence< double > >& rValueSeq )
+{
+ for( const uno::Sequence< double >& rSubSeq : rValueSeq )
+ {
+ for( const double fValue : rSubSeq )
+ Append( fValue );
+ }
+}
+
+
+void ScaDoubleList::Append(
+ const uno::Sequence< uno::Sequence< sal_Int32 > >& rValueSeq )
+{
+ for( const uno::Sequence< sal_Int32 >& rSubSeq : rValueSeq )
+ {
+ for( const sal_Int32 nValue : rSubSeq )
+ Append( nValue );
+ }
+}
+
+void ScaDoubleList::Append(
+ const ScaAnyConverter& rAnyConv,
+ const uno::Any& rAny,
+ bool bIgnoreEmpty )
+{
+ if( auto s = o3tl::tryAccess<
+ css::uno::Sequence<css::uno::Sequence<css::uno::Any>>>(rAny) )
+ Append( rAnyConv, *s, bIgnoreEmpty );
+ else
+ {
+ double fValue;
+ if( rAnyConv.getDouble( fValue, rAny ) )
+ Append( fValue );
+ else if( !bIgnoreEmpty )
+ Append( 0.0 );
+ }
+}
+
+
+void ScaDoubleList::Append(
+ const ScaAnyConverter& rAnyConv,
+ const uno::Sequence< uno::Any >& rAnySeq,
+ bool bIgnoreEmpty )
+{
+ for( const uno::Any& rAny : rAnySeq )
+ Append( rAnyConv, rAny, bIgnoreEmpty );
+}
+
+
+void ScaDoubleList::Append(
+ const ScaAnyConverter& rAnyConv,
+ const uno::Sequence< uno::Sequence< uno::Any > >& rAnySeq,
+ bool bIgnoreEmpty )
+{
+ for( const uno::Sequence< uno::Any >& rArray : rAnySeq )
+ Append( rAnyConv, rArray, bIgnoreEmpty );
+}
+
+void ScaDoubleList::Append(
+ ScaAnyConverter& rAnyConv,
+ const uno::Reference< beans::XPropertySet >& xOpt,
+ const uno::Sequence< uno::Any >& rAnySeq )
+{
+ rAnyConv.init( xOpt );
+ Append( rAnyConv, rAnySeq, true/*bIgnoreEmpty*/ );
+}
+
+
+bool ScaDoubleList::CheckInsert( double ) const
+{
+ return true;
+}
+
+
+bool ScaDoubleListGT0::CheckInsert( double fValue ) const
+{
+ if( fValue < 0.0 )
+ throw lang::IllegalArgumentException();
+ return fValue > 0.0;
+}
+
+
+bool ScaDoubleListGE0::CheckInsert( double fValue ) const
+{
+ if( fValue < 0.0 )
+ throw lang::IllegalArgumentException();
+ return true;
+}
+
+
+Complex::Complex( const OUString& rStr )
+{
+ if( !ParseString( rStr, *this ) )
+ throw lang::IllegalArgumentException();
+}
+
+
+inline bool Complex::IsImagUnit( sal_Unicode c )
+{
+ return c == 'i' || c == 'j';
+}
+
+bool Complex::ParseString( const OUString& rStr, Complex& rCompl )
+{
+ rCompl.c = '\0'; // do not force a symbol, if only real part present
+
+ const sal_Unicode* pStr = rStr.getStr();
+
+ if( IsImagUnit( *pStr ) && rStr.getLength() == 1)
+ {
+ rCompl.r = 0.0;
+ rCompl.i = 1.0;
+ rCompl.c = *pStr;
+ return true;
+ }
+
+ double f;
+
+ if( !ParseDouble( pStr, f ) )
+ return false;
+
+ switch( *pStr )
+ {
+ case '-': // imag part follows
+ case '+':
+ {
+ double r = f;
+ if( IsImagUnit( pStr[ 1 ] ) )
+ {
+ rCompl.c = pStr[ 1 ];
+ if( pStr[ 2 ] == 0 )
+ {
+ rCompl.r = f;
+ rCompl.i = ( *pStr == '+' )? 1.0 : -1.0;
+ return true;
+ }
+ }
+ else if( ParseDouble( pStr, f ) && IsImagUnit( *pStr ) )
+ {
+ rCompl.c = *pStr;
+ pStr++;
+ if( *pStr == 0 )
+ {
+ rCompl.r = r;
+ rCompl.i = f;
+ return true;
+ }
+ }
+ }
+ break;
+ case 'j':
+ case 'i':
+ rCompl.c = *pStr;
+ pStr++;
+ if( *pStr == 0 )
+ {
+ rCompl.i = f;
+ rCompl.r = 0.0;
+ return true;
+ }
+ break;
+ case 0: // only real-part
+ rCompl.r = f;
+ rCompl.i = 0.0;
+ return true;
+ }
+
+ return false;
+}
+
+
+OUString Complex::GetString() const
+{
+ CHK_FINITE(r);
+ CHK_FINITE(i);
+ OUStringBuffer aRet;
+
+ bool bHasImag = i != 0.0;
+ bool bHasReal = !bHasImag || (r != 0.0);
+
+ if( bHasReal )
+ aRet.append(::GetString( r, false ));
+ if( bHasImag )
+ {
+ if( i == 1.0 )
+ {
+ if( bHasReal )
+ aRet.append('+');
+ }
+ else if( i == -1.0 )
+ aRet.append('-');
+ else
+ aRet.append(::GetString( i, bHasReal ));
+ aRet.append((c != 'j') ? 'i' : 'j');
+ }
+
+ return aRet.makeStringAndClear();
+}
+
+
+double Complex::Arg() const
+{
+ if( r == 0.0 && i == 0.0 )
+ throw lang::IllegalArgumentException();
+
+ double phi = acos( r / Abs() );
+
+ if( i < 0.0 )
+ phi = -phi;
+
+ return phi;
+}
+
+
+void Complex::Power( double fPower )
+{
+ if( r == 0.0 && i == 0.0 )
+ {
+ if( fPower <= 0 )
+ throw lang::IllegalArgumentException();
+ r = i = 0.0;
+ return;
+ }
+
+ double p, phi;
+
+ p = Abs();
+
+ phi = acos( r / p );
+ if( i < 0.0 )
+ phi = -phi;
+
+ p = pow( p, fPower );
+ phi *= fPower;
+
+ r = cos( phi ) * p;
+ i = sin( phi ) * p;
+}
+
+
+void Complex::Sqrt()
+{
+ static const double fMultConst = 0.7071067811865475; // ...2440084436210485 = 1/sqrt(2)
+ double p = Abs();
+ double i_ = sqrt( p - r ) * fMultConst;
+
+ r = sqrt( p + r ) * fMultConst;
+ i = ( i < 0.0 )? -i_ : i_;
+}
+
+
+void Complex::Sin()
+{
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+
+ if( i )
+ {
+ double r_;
+
+ r_ = sin( r ) * cosh( i );
+ i = cos( r ) * sinh( i );
+ r = r_;
+ }
+ else
+ r = sin( r );
+}
+
+
+void Complex::Cos()
+{
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+
+ if( i )
+ {
+ double r_;
+
+ r_ = cos( r ) * cosh( i );
+ i = -( sin( r ) * sinh( i ) );
+ r = r_;
+ }
+ else
+ r = cos( r );
+}
+
+
+void Complex::Div( const Complex& z )
+{
+ if( z.r == 0 && z.i == 0 )
+ throw lang::IllegalArgumentException();
+
+ double a1 = r;
+ double a2 = z.r;
+ double b1 = i;
+ double b2 = z.i;
+
+ double f = 1.0 / ( a2 * a2 + b2 * b2 );
+
+ r = ( a1 * a2 + b1 * b2 ) * f;
+ i = ( a2 * b1 - a1 * b2 ) * f;
+
+ if( !c ) c = z.c;
+}
+
+
+void Complex::Exp()
+{
+ double fE = exp( r );
+ r = fE * cos( i );
+ i = fE * sin( i );
+}
+
+
+void Complex::Ln()
+{
+ if( r == 0.0 && i == 0.0 )
+ throw lang::IllegalArgumentException();
+
+ double fAbs = Abs();
+ bool bNegi = i < 0.0;
+
+ i = acos( r / fAbs );
+
+ if( bNegi )
+ i = -i;
+
+ r = log( fAbs );
+}
+
+
+void Complex::Log10()
+{
+ Ln();
+ Mult( 0.434294481903251828 ); // * log10( e )
+}
+
+
+void Complex::Log2()
+{
+ Ln();
+ Mult( 1.442695040888963407 ); // * log2( e )
+}
+
+
+void Complex::Tan()
+{
+ if ( i )
+ {
+ if( !::rtl::math::isValidArcArg( 2.0 * r ) )
+ throw lang::IllegalArgumentException();
+ double fScale =1.0 / ( cos( 2.0 * r ) + cosh( 2.0 * i ));
+ r = sin( 2.0 * r ) * fScale;
+ i = sinh( 2.0 * i ) * fScale;
+ }
+ else
+ {
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+ r = tan( r );
+ }
+}
+
+
+void Complex::Sec()
+{
+ if( i )
+ {
+ if( !::rtl::math::isValidArcArg( 2 * r ) )
+ throw lang::IllegalArgumentException();
+ double fScale = 1.0 / (cosh( 2.0 * i) + cos ( 2.0 * r));
+ double r_;
+ r_ = 2.0 * cos( r ) * cosh( i ) * fScale;
+ i = 2.0 * sin( r ) * sinh( i ) * fScale;
+ r = r_;
+ }
+ else
+ {
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+ r = 1.0 / cos( r );
+ }
+}
+
+
+void Complex::Csc()
+{
+ if( i )
+ {
+ if( !::rtl::math::isValidArcArg( 2 * r ) )
+ throw lang::IllegalArgumentException();
+ double fScale = 1.0 / (cosh( 2.0 * i) - cos ( 2.0 * r));
+ double r_;
+ r_ = 2.0 * sin( r ) * cosh( i ) * fScale;
+ i = -2.0 * cos( r ) * sinh( i ) * fScale;
+ r = r_;
+ }
+ else
+ {
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+ r = 1.0 / sin( r );
+ }
+}
+
+
+void Complex::Cot()
+{
+ if ( i )
+ {
+ if( !::rtl::math::isValidArcArg( 2.0 * r ) )
+ throw lang::IllegalArgumentException();
+ double fScale =1.0 / ( cosh( 2.0 * i ) - cos( 2.0 * r ) );
+ r = sin( 2.0 * r ) * fScale;
+ i = - ( sinh( 2.0 * i ) * fScale );
+ }
+ else
+ {
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+ r = 1.0 / tan( r );
+ }
+}
+
+
+void Complex::Sinh()
+{
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+
+ if( i )
+ {
+ double r_;
+ r_ = sinh( r ) * cos( i );
+ i = cosh( r ) * sin( i );
+ r = r_;
+ }
+ else
+ r = sinh( r );
+}
+
+
+void Complex::Cosh()
+{
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+
+ if( i )
+ {
+ double r_;
+ r_ = cosh( r ) * cos( i );
+ i = sinh( r ) * sin( i );
+ r = r_;
+ }
+ else
+ r = cosh( r );
+}
+
+
+void Complex::Sech()
+{
+ if ( i )
+ {
+ if( !::rtl::math::isValidArcArg( 2.0 * r ) )
+ throw lang::IllegalArgumentException();
+ double fScale =1.0 / ( cosh( 2.0 * r ) + cos( 2.0 * i ));
+ double r_;
+ r_ = 2.0 * cosh( r ) * cos( i ) * fScale;
+ i = - (2.0 * sinh( r ) * sin( i ) * fScale );
+ r = r_ ;
+ }
+ else
+ {
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+ r = 1.0 / cosh( r );
+ }
+}
+
+
+void Complex::Csch()
+{
+ if ( i )
+ {
+ if( !::rtl::math::isValidArcArg( 2.0 * r ) )
+ throw lang::IllegalArgumentException();
+ double fScale =1.0 / ( cosh( 2.0 * r ) - cos( 2.0 * i ));
+ double r_;
+ r_ = 2.0 * sinh( r ) * cos( i ) * fScale;
+ i = - ( 2.0 * cosh( r ) * sin( i ) * fScale );
+ r = r_ ;
+ }
+ else
+ {
+ if( !::rtl::math::isValidArcArg( r ) )
+ throw lang::IllegalArgumentException();
+ r = 1.0 / sinh( r );
+ }
+}
+
+
+ComplexList::~ComplexList()
+{
+}
+
+
+void ComplexList::Append( const uno::Sequence< uno::Sequence< OUString > >& r )
+{
+ for( const uno::Sequence< OUString >& rList : r )
+ {
+ for( const OUString& rStr : rList )
+ {
+ if( !rStr.isEmpty() )
+ Append( Complex( rStr ) );
+ }
+ }
+}
+
+
+void ComplexList::Append( const uno::Sequence< uno::Any >& aMultPars )
+{
+ for( const uno::Any& r : aMultPars )
+ {
+ switch( r.getValueTypeClass() )
+ {
+ case uno::TypeClass_VOID: break;
+ case uno::TypeClass_STRING:
+ {
+ auto pStr = o3tl::forceAccess<OUString>(r);
+
+ if( !pStr->isEmpty() )
+ Append( Complex( *pStr ) );
+ }
+ break;
+ case uno::TypeClass_DOUBLE:
+ Append( Complex( *o3tl::forceAccess<double>(r), 0.0 ) );
+ break;
+ case uno::TypeClass_SEQUENCE:
+ {
+ uno::Sequence< uno::Sequence< uno::Any > > aValArr;
+ if( !(r >>= aValArr) )
+ throw lang::IllegalArgumentException();
+
+ for( const uno::Sequence< uno::Any >& rArr : std::as_const(aValArr) )
+ Append( rArr );
+ }
+ break;
+ default:
+ throw lang::IllegalArgumentException();
+ }
+ }
+}
+
+ConvertData::ConvertData(const char p[], double fC, ConvertDataClass e, bool bPrefSupport)
+ : fConst(fC)
+ , aName(p, strlen(p), RTL_TEXTENCODING_MS_1252)
+ , eClass(e)
+ , bPrefixSupport(bPrefSupport)
+{
+}
+
+ConvertData::~ConvertData()
+{
+}
+
+sal_Int16 ConvertData::GetMatchingLevel( const OUString& rRef ) const
+{
+ OUString aStr = rRef;
+ sal_Int32 nLen = rRef.getLength();
+ sal_Int32 nIndex = rRef.lastIndexOf( '^' );
+ if( nIndex > 0 && nIndex == ( nLen - 2 ) )
+ aStr = aStr.copy( 0, nLen - 2 ) + OUStringChar( aStr[ nLen - 1 ] );
+ if( aName == aStr )
+ return 0;
+ else
+ {
+ const sal_Unicode* p = aStr.getStr();
+
+ nLen = aStr.getLength();
+ bool bPref = bPrefixSupport;
+ bool bOneChar = (bPref && nLen > 1 && (aName == p + 1));
+ if (bOneChar || (bPref && nLen > 2 && (aName == p + 2) &&
+ *p == 'd' && *(p+1) == 'a'))
+ {
+ sal_Int16 n;
+ switch( *p )
+ {
+ case 'y': n = -24; break; // yocto
+ case 'z': n = -21; break; // zepto
+ case 'a': n = -18; break;
+ case 'f': n = -15; break;
+ case 'p': n = -12; break;
+ case 'n': n = -9; break;
+ case 'u': n = -6; break;
+ case 'm': n = -3; break;
+ case 'c': n = -2; break;
+ case 'd':
+ {
+ if ( bOneChar )
+ n = -1; // deci
+ else
+ n = 1; // deca
+ }
+ break;
+ case 'e': n = 1; break;
+ case 'h': n = 2; break;
+ case 'k': n = 3; break;
+ case 'M': n = 6; break;
+ case 'G': n = 9; break;
+ case 'T': n = 12; break;
+ case 'P': n = 15; break;
+ case 'E': n = 18; break;
+ case 'Z': n = 21; break; // zetta
+ case 'Y': n = 24; break; // yotta
+ default:
+ n = INV_MATCHLEV;
+ }
+
+// We could weed some nonsense out, ODFF doesn't say so though.
+#if 0
+ if (n < 0 && Class() == CDC_Information)
+ n = INV_MATCHLEV; // milli-bits doesn't make sense
+#endif
+
+//! <HACK> "cm3" is not 10^-2 m^3 but 10^-6 m^3 !!! ------------------
+ if( n != INV_MATCHLEV )
+ {
+ sal_Unicode cLast = p[ aStr.getLength() - 1 ];
+ if( cLast == '2' )
+ n *= 2;
+ else if( cLast == '3' )
+ n *= 3;
+ }
+//! </HACK> -------------------------------------------------------------------
+
+ return n;
+ }
+ else if ( nLen > 2 && ( aName == p + 2 ) && ( Class() == CDC_Information ) )
+ {
+ const sal_Unicode* pStr = aStr.getStr();
+ if ( *(pStr + 1) != 'i')
+ return INV_MATCHLEV;
+ sal_Int16 n;
+ switch( *pStr )
+ {
+ case 'k': n = 10; break;
+ case 'M': n = 20; break;
+ case 'G': n = 30; break;
+ case 'T': n = 40; break;
+ case 'P': n = 50; break;
+ case 'E': n = 60; break;
+ case 'Z': n = 70; break;
+ case 'Y': n = 80; break;
+ default:
+ n = INV_MATCHLEV;
+ }
+ return n;
+ }
+ else
+ return INV_MATCHLEV;
+ }
+}
+
+
+double ConvertData::Convert(
+ double f, const ConvertData& r, sal_Int16 nLevFrom, sal_Int16 nLevTo ) const
+{
+ if( Class() != r.Class() )
+ throw lang::IllegalArgumentException();
+
+ bool bBinFromLev = ( nLevFrom > 0 && ( nLevFrom % 10 ) == 0 );
+ bool bBinToLev = ( nLevTo > 0 && ( nLevTo % 10 ) == 0 );
+
+ if ( Class() == CDC_Information && ( bBinFromLev || bBinToLev ) )
+ {
+ if ( bBinFromLev && bBinToLev )
+ {
+ nLevFrom = sal::static_int_cast<sal_Int16>( nLevFrom - nLevTo );
+ f *= r.fConst / fConst;
+ if( nLevFrom )
+ f *= pow( 2.0, nLevFrom );
+ }
+ else if ( bBinFromLev )
+ f *= ( r.fConst / fConst ) * ( pow( 2.0, nLevFrom ) / pow( 10.0, nLevTo ) );
+ else
+ f *= ( r.fConst / fConst ) * ( pow( 10.0, nLevFrom ) / pow( 2.0, nLevTo ) );
+ return f;
+ }
+
+ nLevFrom = sal::static_int_cast<sal_Int16>( nLevFrom - nLevTo ); // effective level
+
+ f *= r.fConst / fConst;
+
+ if( nLevFrom )
+ f = ::rtl::math::pow10Exp( f, nLevFrom );
+
+ return f;
+}
+
+
+double ConvertData::ConvertFromBase( double f, sal_Int16 n ) const
+{
+ return ::rtl::math::pow10Exp( f * fConst, -n );
+}
+
+ConvertDataLinear::~ConvertDataLinear()
+{
+}
+
+double ConvertDataLinear::Convert(
+ double f, const ConvertData& r, sal_Int16 nLevFrom, sal_Int16 nLevTo ) const
+{
+ if( Class() != r.Class() )
+ throw lang::IllegalArgumentException();
+ return r.ConvertFromBase( ConvertToBase( f, nLevFrom ), nLevTo );
+}
+
+
+double ConvertDataLinear::ConvertToBase( double f, sal_Int16 n ) const
+{
+ if( n )
+ f = ::rtl::math::pow10Exp( f, n );
+
+ f /= fConst;
+ f -= fOffs;
+
+ return f;
+}
+
+
+double ConvertDataLinear::ConvertFromBase( double f, sal_Int16 n ) const
+{
+ f += fOffs;
+ f *= fConst;
+
+ if( n )
+ f = ::rtl::math::pow10Exp( f, -n );
+
+ return f;
+}
+
+
+ConvertDataList::ConvertDataList()
+{
+#define NEWD(str,unit,cl) maVector.emplace_back(new ConvertData(str,unit,cl))
+#define NEWDP(str,unit,cl) maVector.emplace_back(new ConvertData(str,unit,cl,true))
+#define NEWL(str,unit,offs,cl) maVector.emplace_back(new ConvertDataLinear(str,unit,offs,cl))
+#define NEWLP(str,unit,offs,cl) maVector.emplace_back(new ConvertDataLinear(str,unit,offs,cl,true))
+
+ // *** are extra and not standard Excel Analysis Addin!
+
+ // MASS: 1 Gram is...
+ NEWDP( "g", 1.0000000000000000E00, CDC_Mass ); // Gram
+ NEWD( "sg", 6.8522050005347800E-05, CDC_Mass ); // Pieces
+ NEWD( "lbm", 2.2046229146913400E-03, CDC_Mass ); // Pound (commercial weight)
+ NEWDP( "u", 6.0221370000000000E23, CDC_Mass ); // U (atomic mass)
+ NEWD( "ozm", 3.5273971800362700E-02, CDC_Mass ); // Ounce (commercial weight)
+ NEWD( "stone", 1.574730e-04, CDC_Mass ); // *** Stone
+ NEWD( "ton", 1.102311e-06, CDC_Mass ); // *** Ton
+ NEWD( "grain", 1.543236E01, CDC_Mass ); // *** Grain
+ NEWD( "pweight", 7.054792E-01, CDC_Mass ); // *** Pennyweight
+ NEWD( "hweight", 1.968413E-05, CDC_Mass ); // *** Hundredweight
+ NEWD( "shweight", 2.204623E-05, CDC_Mass ); // *** Shorthundredweight
+ NEWD( "brton", 9.842065E-07, CDC_Mass ); // *** Gross Registered Ton
+ NEWD( "cwt", 2.2046226218487758E-05, CDC_Mass ); // U.S. (short) hundredweight
+ NEWD( "shweight", 2.2046226218487758E-05, CDC_Mass ); // U.S. (short) hundredweight also
+ NEWD( "uk_cwt", 1.9684130552221213E-05, CDC_Mass ); // Imperial hundredweight
+ NEWD( "lcwt", 1.9684130552221213E-05, CDC_Mass ); // Imperial hundredweight also
+ NEWD( "hweight", 1.9684130552221213E-05, CDC_Mass ); // Imperial hundredweight also
+ NEWD( "uk_ton", 9.8420652761106063E-07, CDC_Mass ); // Imperial ton
+ NEWD( "LTON", 9.8420652761106063E-07, CDC_Mass ); // Imperial ton also
+
+ // LENGTH: 1 Meter is...
+ NEWDP( "m", 1.0000000000000000E00, CDC_Length ); // Meter
+ NEWD( "mi", 6.2137119223733397E-04, CDC_Length ); // Britsh Mile 6,21371192237333969617434184363e-4
+ NEWD( "Nmi", 5.3995680345572354E-04, CDC_Length ); // Nautical Mile 5,39956803455723542116630669546e-4
+ NEWD( "in", 3.9370078740157480E01, CDC_Length ); // Inch 39,37007874015748031496062992126
+ NEWD( "ft", 3.2808398950131234E00, CDC_Length ); // Foot 3,2808398950131233595800524934383
+ NEWD( "yd", 1.0936132983377078E00, CDC_Length ); // Yard 1,0936132983377077865266841644794
+ NEWDP( "ang", 1.0000000000000000E10, CDC_Length ); // Angstroem
+ NEWD( "Pica", 2.8346456692913386E03, CDC_Length ); // Pica Point (1/72 Inch) 2834,6456692913385826771653543307
+ NEWD( "picapt", 2.8346456692913386E03, CDC_Length ); // Pica Point (1/72 Inch) 2834,6456692913385826771653543307
+ NEWD( "pica", 2.36220472441E02, CDC_Length ); // pica (1/6 Inch)
+ NEWD( "ell", 8.748906E-01, CDC_Length ); // *** Ell
+ NEWDP( "parsec", 3.240779E-17, CDC_Length ); // *** Parsec
+ NEWDP( "pc", 3.240779E-17, CDC_Length ); // *** Parsec also
+ NEWDP( "lightyear", 1.0570234557732930E-16, CDC_Length ); // *** Light Year
+ NEWDP( "ly", 1.0570234557732930E-16, CDC_Length ); // *** Light Year also
+ NEWD( "survey_mi", 6.2136994949494949E-04, CDC_Length ); // U.S. survey mile
+
+ // TIME: 1 Second is...
+ NEWD( "yr", 3.1688087814028950E-08, CDC_Time ); // Year
+ NEWD( "day", 1.1574074074074074E-05, CDC_Time ); // Day
+ NEWD( "d", 1.1574074074074074E-05, CDC_Time ); // Day also
+ NEWD( "hr", 2.7777777777777778E-04, CDC_Time ); // Hour
+ NEWD( "mn", 1.6666666666666667E-02, CDC_Time ); // Minute
+ NEWD( "min", 1.6666666666666667E-02, CDC_Time ); // Minute also
+ NEWDP( "sec", 1.0000000000000000E00, CDC_Time ); // Second
+ NEWDP( "s", 1.0000000000000000E00, CDC_Time ); // Second also
+
+ // PRESSURE: 1 Pascal is...
+ NEWDP( "Pa", 1.0000000000000000E00, CDC_Pressure ); // Pascal
+ NEWDP( "atm", 9.8692329999819300E-06, CDC_Pressure ); // Atmosphere
+ NEWDP( "at", 9.8692329999819300E-06, CDC_Pressure ); // Atmosphere also
+ NEWDP( "mmHg", 7.5006170799862700E-03, CDC_Pressure ); // mm Hg (Mercury)
+ NEWD( "Torr", 7.5006380000000000E-03, CDC_Pressure ); // *** Torr
+ NEWD( "psi", 1.4503770000000000E-04, CDC_Pressure ); // *** Psi
+
+ // FORCE: 1 Newton is...
+ NEWDP( "N", 1.0000000000000000E00, CDC_Force ); // Newton
+ NEWDP( "dyn", 1.0000000000000000E05, CDC_Force ); // Dyn
+ NEWDP( "dy", 1.0000000000000000E05, CDC_Force ); // Dyn also
+ NEWD( "lbf", 2.24808923655339E-01, CDC_Force ); // Pound-Force
+ NEWDP( "pond", 1.019716E02, CDC_Force ); // *** Pond
+
+ // ENERGY: 1 Joule is...
+ NEWDP( "J", 1.0000000000000000E00, CDC_Energy ); // Joule
+ NEWDP( "e", 1.0000000000000000E07, CDC_Energy ); // Erg -> https://en.wikipedia.org/wiki/Erg
+ NEWDP( "c", 2.3900624947346700E-01, CDC_Energy ); // Thermodynamical Calorie
+ NEWDP( "cal", 2.3884619064201700E-01, CDC_Energy ); // Calorie
+ NEWDP( "eV", 6.2414570000000000E18, CDC_Energy ); // Electronvolt
+ NEWDP( "ev", 6.2414570000000000E18, CDC_Energy ); // Electronvolt also
+ NEWD( "HPh", 3.7250611111111111E-07, CDC_Energy ); // Horsepower Hours
+ NEWD( "hh", 3.7250611111111111E-07, CDC_Energy ); // Horsepower Hours also
+ NEWDP( "Wh", 2.7777777777777778E-04, CDC_Energy ); // Watt Hours
+ NEWDP( "wh", 2.7777777777777778E-04, CDC_Energy ); // Watt Hours also
+ NEWD( "flb", 2.37304222192651E01, CDC_Energy ); // Foot Pound
+ NEWD( "BTU", 9.4781506734901500E-04, CDC_Energy ); // British Thermal Unit
+ NEWD( "btu", 9.4781506734901500E-04, CDC_Energy ); // British Thermal Unit also
+
+ // POWER: 1 Watt is...
+ NEWDP( "W", 1.0000000000000000E00, CDC_Power ); // Watt
+ NEWDP( "w", 1.0000000000000000E00, CDC_Power ); // Watt also
+ NEWD( "HP", 1.341022E-03, CDC_Power ); // Horsepower
+ NEWD( "h", 1.341022E-03, CDC_Power ); // Horsepower also
+ NEWD( "PS", 1.359622E-03, CDC_Power ); // *** German Pferdestaerke
+
+ // MAGNETISM: 1 Tesla is...
+ NEWDP( "T", 1.0000000000000000E00, CDC_Magnetism ); // Tesla
+ NEWDP( "ga", 1.0000000000000000E04, CDC_Magnetism ); // Gauss
+
+ // TEMPERATURE: 1 Kelvin is...
+ NEWL( "C", 1.0000000000000000E00, -2.7315000000000000E02, CDC_Temperature ); // Celsius
+ NEWL( "cel", 1.0000000000000000E00, -2.7315000000000000E02, CDC_Temperature ); // Celsius also
+ NEWL( "F", 1.8000000000000000E00, -2.5537222222222222E02, CDC_Temperature ); // Fahrenheit
+ NEWL( "fah", 1.8000000000000000E00, -2.5537222222222222E02, CDC_Temperature ); // Fahrenheit also
+ NEWLP( "K", 1.0000000000000000E00, +0.0000000000000000E00, CDC_Temperature ); // Kelvin
+ NEWLP( "kel", 1.0000000000000000E00, +0.0000000000000000E00, CDC_Temperature ); // Kelvin also
+ NEWL( "Reau", 8.0000000000000000E-01, -2.7315000000000000E02, CDC_Temperature ); // *** Reaumur
+ NEWL( "Rank", 1.8000000000000000E00, +0.0000000000000000E00, CDC_Temperature ); // *** Rankine
+
+ // VOLUME: 1 Liter is...
+ NEWD( "tsp", 2.0288413621105798E02, CDC_Volume ); // US teaspoon 1/768 gallon
+ NEWD( "tbs", 6.7628045403685994E01, CDC_Volume ); // US tablespoon 1/256 gallon
+ NEWD( "oz", 3.3814022701842997E01, CDC_Volume ); // Ounce Liquid 1/128 gallon
+ NEWD( "cup", 4.2267528377303746E00, CDC_Volume ); // Cup 1/16 gallon
+ NEWD( "pt", 2.1133764188651873E00, CDC_Volume ); // US Pint 1/8 gallon
+ NEWD( "us_pt", 2.1133764188651873E00, CDC_Volume ); // US Pint also
+ NEWD( "uk_pt", 1.7597539863927023E00, CDC_Volume ); // UK Pint 1/8 imperial gallon
+ NEWD( "qt", 1.0566882094325937E00, CDC_Volume ); // Quart 1/4 gallon
+ NEWD( "gal", 2.6417205235814842E-01, CDC_Volume ); // Gallon 1/3.785411784
+ NEWDP( "l", 1.0000000000000000E00, CDC_Volume ); // Liter
+ NEWDP( "L", 1.0000000000000000E00, CDC_Volume ); // Liter also
+ NEWDP( "lt", 1.0000000000000000E00, CDC_Volume ); // Liter also
+ NEWDP( "m3", 1.0000000000000000E-03, CDC_Volume ); // *** Cubic Meter
+ NEWD( "mi3", 2.3991275857892772E-13, CDC_Volume ); // *** Cubic Britsh Mile
+ NEWD( "Nmi3", 1.5742621468581148E-13, CDC_Volume ); // *** Cubic Nautical Mile
+ NEWD( "in3", 6.1023744094732284E01, CDC_Volume ); // *** Cubic Inch
+ NEWD( "ft3", 3.5314666721488590E-02, CDC_Volume ); // *** Cubic Foot
+ NEWD( "yd3", 1.3079506193143922E-03, CDC_Volume ); // *** Cubic Yard
+ NEWDP( "ang3", 1.0000000000000000E27, CDC_Volume ); // *** Cubic Angstroem
+ NEWD( "Pica3", 2.2776990435870636E07, CDC_Volume ); // *** Cubic Pica Point (1/72 inch)
+ NEWD( "picapt3", 2.2776990435870636E07, CDC_Volume ); // *** Cubic Pica Point (1/72 inch)
+ NEWD( "pica3", 1.31811287245E04, CDC_Volume ); // *** Cubic Pica (1/6 inch)
+ NEWD( "barrel", 6.2898107704321051E-03, CDC_Volume ); // *** Barrel (=42gal)
+ NEWD( "bushel", 2.837759E-02, CDC_Volume ); // *** Bushel
+ NEWD( "regton", 3.531467E-04, CDC_Volume ); // *** Register ton
+ NEWD( "GRT", 3.531467E-04, CDC_Volume ); // *** Register ton also
+ NEWD( "Schooner", 2.3529411764705882E00, CDC_Volume ); // *** austr. Schooner
+ NEWD( "Middy", 3.5087719298245614E00, CDC_Volume ); // *** austr. Middy
+ NEWD( "Glass", 5.0000000000000000E00, CDC_Volume ); // *** austr. Glass
+ NEWD( "Sixpack", 0.5, CDC_Volume ); // ***
+ NEWD( "Humpen", 2.0, CDC_Volume ); // ***
+ NEWD( "ly3", 1.1810108125623799E-51, CDC_Volume ); // *** Cubic light-year
+ NEWD( "MTON", 1.4125866688595436E00, CDC_Volume ); // *** Measurement ton
+ NEWD( "tspm", 2.0000000000000000E02, CDC_Volume ); // *** Modern teaspoon
+ NEWD( "uk_gal", 2.1996924829908779E-01, CDC_Volume ); // U.K. / Imperial gallon 1/4.54609
+ NEWD( "uk_qt", 8.7987699319635115E-01, CDC_Volume ); // U.K. / Imperial quart 1/4 imperial gallon
+
+ // 1 Square Meter is...
+ NEWDP( "m2", 1.0000000000000000E00, CDC_Area ); // *** Square Meter
+ NEWD( "mi2", 3.8610215854244585E-07, CDC_Area ); // *** Square Britsh Mile
+ NEWD( "Nmi2", 2.9155334959812286E-07, CDC_Area ); // *** Square Nautical Mile
+ NEWD( "in2", 1.5500031000062000E03, CDC_Area ); // *** Square Inch
+ NEWD( "ft2", 1.0763910416709722E01, CDC_Area ); // *** Square Foot
+ NEWD( "yd2", 1.1959900463010803E00, CDC_Area ); // *** Square Yard
+ NEWDP( "ang2", 1.0000000000000000E20, CDC_Area ); // *** Square Angstroem
+ NEWD( "Pica2", 8.0352160704321409E06, CDC_Area ); // *** Square Pica Point (1/72 inch)
+ NEWD( "picapt2", 8.0352160704321409E06, CDC_Area ); // *** Square Pica Point (1/72 inch)
+ NEWD( "pica2", 5.58001116002232E04, CDC_Area ); // *** Square Pica (1/6 inch)
+ NEWD( "Morgen", 4.0000000000000000E-04, CDC_Area ); // *** Morgen
+ NEWDP( "ar", 1.000000E-02, CDC_Area ); // *** Ar
+ NEWD( "acre", 2.471053815E-04, CDC_Area ); // *** Acre
+ NEWD( "uk_acre", 2.4710538146716534E-04, CDC_Area ); // *** International acre
+ NEWD( "us_acre", 2.4710439304662790E-04, CDC_Area ); // *** U.S. survey/statute acre
+ NEWD( "ly2", 1.1172985860549147E-32, CDC_Area ); // *** Square Light-year
+ NEWD( "ha", 1.000000E-04, CDC_Area ); // *** Hectare
+
+ // SPEED: 1 Meter per Second is...
+ NEWDP( "m/s", 1.0000000000000000E00, CDC_Speed ); // *** Meters per Second
+ NEWDP( "m/sec", 1.0000000000000000E00, CDC_Speed ); // *** Meters per Second also
+ NEWDP( "m/h", 3.6000000000000000E03, CDC_Speed ); // *** Meters per Hour
+ NEWDP( "m/hr", 3.6000000000000000E03, CDC_Speed ); // *** Meters per Hour also
+ NEWD( "mph", 2.2369362920544023E00, CDC_Speed ); // *** Britsh Miles per Hour
+ NEWD( "kn", 1.9438444924406048E00, CDC_Speed ); // *** Knot = Nautical Miles per Hour
+ NEWD( "admkn", 1.9438446603753486E00, CDC_Speed ); // *** Admiralty Knot
+ NEWD( "ludicrous speed", 2.0494886343432328E-14, CDC_Speed ); // ***
+ NEWD( "ridiculous speed", 4.0156958471424288E-06, CDC_Speed); // ***
+
+ // INFORMATION: 1 Bit is...
+ NEWDP( "bit", 1.00E00, CDC_Information); // *** Bit
+ NEWDP( "byte", 1.25E-01, CDC_Information); // *** Byte
+}
+
+
+ConvertDataList::~ConvertDataList()
+{
+}
+
+
+double ConvertDataList::Convert( double fVal, const OUString& rFrom, const OUString& rTo )
+{
+ ConvertData* pFrom = nullptr;
+ ConvertData* pTo = nullptr;
+ bool bSearchFrom = true;
+ bool bSearchTo = true;
+ sal_Int16 nLevelFrom = 0;
+ sal_Int16 nLevelTo = 0;
+
+ for( const auto& rItem : maVector )
+ {
+ ConvertData* p = rItem.get();
+ if( bSearchFrom )
+ {
+ sal_Int16 n = p->GetMatchingLevel( rFrom );
+ if( n != INV_MATCHLEV )
+ {
+ if( n )
+ { // only first match for partial equality rulz a little bit more
+ pFrom = p;
+ nLevelFrom = n;
+ }
+ else
+ { // ... but exact match rulz most
+ pFrom = p;
+ bSearchFrom = false;
+ nLevelFrom = n;
+ }
+ }
+ }
+
+ if( bSearchTo )
+ {
+ sal_Int16 n = p->GetMatchingLevel( rTo );
+ if( n != INV_MATCHLEV )
+ {
+ if( n )
+ { // only first match for partial equality rulz a little bit more
+ pTo = p;
+ nLevelTo = n;
+ }
+ else
+ { // ... but exact match rulz most
+ pTo = p;
+ bSearchTo = false;
+ nLevelTo = n;
+ }
+ }
+ }
+
+ if( !bSearchFrom && !bSearchTo )
+ break;
+ }
+
+ if( !pFrom || !pTo )
+ throw lang::IllegalArgumentException();
+
+ return pFrom->Convert( fVal, *pTo, nLevelFrom, nLevelTo );
+}
+
+
+ScaDate::ScaDate() :
+ nOrigDay( 1 ),
+ nDay( 1 ),
+ nMonth( 1 ),
+ nYear( 1900 ),
+ bLastDayMode( true ),
+ bLastDay( false ),
+ b30Days( false ),
+ bUSMode( false )
+{
+}
+
+ScaDate::ScaDate( sal_Int32 nNullDate, sal_Int32 nDate, sal_Int32 nBase )
+{
+ DaysToDate( nNullDate + nDate, nOrigDay, nMonth, nYear );
+ bLastDayMode = (nBase != 5);
+ bLastDay = (nOrigDay >= ::DaysInMonth( nMonth, nYear ));
+ b30Days = (nBase == 0) || (nBase == 4);
+ bUSMode = (nBase == 0);
+ setDay();
+}
+
+ScaDate::ScaDate( const ScaDate& rCopy ) :
+ nOrigDay( rCopy.nOrigDay ),
+ nDay( rCopy.nDay ),
+ nMonth( rCopy.nMonth ),
+ nYear( rCopy.nYear ),
+ bLastDayMode( rCopy.bLastDayMode ),
+ bLastDay( rCopy.bLastDay ),
+ b30Days( rCopy.b30Days ),
+ bUSMode( rCopy.bUSMode )
+{
+}
+
+ScaDate& ScaDate::operator=( const ScaDate& rCopy )
+{
+ if( this != &rCopy )
+ {
+ nOrigDay = rCopy.nOrigDay;
+ nDay = rCopy.nDay;
+ nMonth = rCopy.nMonth;
+ nYear = rCopy.nYear;
+ bLastDayMode = rCopy.bLastDayMode;
+ bLastDay = rCopy.bLastDay;
+ b30Days = rCopy.b30Days;
+ bUSMode = rCopy.bUSMode;
+ }
+ return *this;
+}
+
+void ScaDate::setDay()
+{
+ if( b30Days )
+ {
+ // 30-days-mode: set nDay to 30 if original was last day in month
+ nDay = std::min( nOrigDay, static_cast< sal_uInt16 >( 30 ) );
+ if( bLastDay || (nDay >= ::DaysInMonth( nMonth, nYear )) )
+ nDay = 30;
+ }
+ else
+ {
+ // set nDay to last day in this month if original was last day
+ sal_uInt16 nLastDay = ::DaysInMonth( nMonth, nYear );
+ nDay = bLastDay ? nLastDay : std::min( nOrigDay, nLastDay );
+ }
+}
+
+sal_Int32 ScaDate::getDaysInMonthRange( sal_uInt16 nFrom, sal_uInt16 nTo ) const
+{
+ if( nFrom > nTo )
+ return 0;
+
+ sal_Int32 nRet = 0;
+ if( b30Days )
+ nRet = (nTo - nFrom + 1) * 30;
+ else
+ {
+ for( sal_uInt16 nMonthIx = nFrom; nMonthIx <= nTo; ++nMonthIx )
+ nRet += getDaysInMonth( nMonthIx );
+ }
+ return nRet;
+}
+
+sal_Int32 ScaDate::getDaysInYearRange( sal_uInt16 nFrom, sal_uInt16 nTo ) const
+{
+ if( nFrom > nTo )
+ return 0;
+
+ return b30Days ? ((nTo - nFrom + 1) * 360) : ::GetDaysInYears( nFrom, nTo );
+}
+
+void ScaDate::doAddYears( sal_Int32 nYearCount )
+{
+ sal_Int32 nNewYear = nYearCount + nYear;
+ if( (nNewYear < 0) || (nNewYear > 0x7FFF) )
+ throw lang::IllegalArgumentException();
+ nYear = static_cast< sal_uInt16 >( nNewYear );
+}
+
+void ScaDate::addMonths( sal_Int32 nMonthCount )
+{
+ sal_Int32 nNewMonth = nMonthCount + nMonth;
+ if( nNewMonth > 12 )
+ {
+ --nNewMonth;
+ doAddYears( nNewMonth / 12 );
+ nMonth = static_cast< sal_uInt16 >( nNewMonth % 12 ) + 1;
+ }
+ else if( nNewMonth < 1 )
+ {
+ doAddYears( nNewMonth / 12 - 1 );
+ nMonth = static_cast< sal_uInt16 >( nNewMonth % 12 + 12 );
+ }
+ else
+ nMonth = static_cast< sal_uInt16 >( nNewMonth );
+ setDay();
+}
+
+sal_Int32 ScaDate::getDate( sal_Int32 nNullDate ) const
+{
+ sal_uInt16 nLastDay = ::DaysInMonth( nMonth, nYear );
+ sal_uInt16 nRealDay = (bLastDayMode && bLastDay) ? nLastDay : std::min( nLastDay, nOrigDay );
+ return ::DateToDays( nRealDay, nMonth, nYear ) - nNullDate;
+}
+
+sal_Int32 ScaDate::getDiff( const ScaDate& rFrom, const ScaDate& rTo )
+{
+ if( rFrom > rTo )
+ return getDiff( rTo, rFrom );
+
+ sal_Int32 nDiff = 0;
+ ScaDate aFrom( rFrom );
+ ScaDate aTo( rTo );
+
+ if( rTo.b30Days )
+ {
+ // corrections for base 0 (US NASD)
+ if( rTo.bUSMode )
+ {
+ if( ((rFrom.nMonth == 2) || (rFrom.nDay < 30)) && (aTo.nOrigDay == 31) )
+ aTo.nDay = 31;
+ else if( (aTo.nMonth == 2) && aTo.bLastDay )
+ aTo.nDay = ::DaysInMonth( 2, aTo.nYear );
+ }
+ // corrections for base 4 (Europe)
+ else
+ {
+ if( (aFrom.nMonth == 2) && (aFrom.nDay == 30) )
+ aFrom.nDay = ::DaysInMonth( 2, aFrom.nYear );
+ if( (aTo.nMonth == 2) && (aTo.nDay == 30) )
+ aTo.nDay = ::DaysInMonth( 2, aTo.nYear );
+ }
+ }
+
+ if( (aFrom.nYear < aTo.nYear) || ((aFrom.nYear == aTo.nYear) && (aFrom.nMonth < aTo.nMonth)) )
+ {
+ // move aFrom to 1st day of next month
+ nDiff = aFrom.getDaysInMonth() - aFrom.nDay + 1;
+ aFrom.nOrigDay = aFrom.nDay = 1;
+ aFrom.bLastDay = false;
+ aFrom.addMonths( 1 );
+
+ if( aFrom.nYear < aTo.nYear )
+ {
+ // move aFrom to 1st day of next year
+ nDiff += aFrom.getDaysInMonthRange( aFrom.nMonth, 12 );
+ aFrom.addMonths( 13 - aFrom.nMonth );
+
+ // move aFrom to 1st day of this year
+ nDiff += aFrom.getDaysInYearRange( aFrom.nYear, aTo.nYear - 1 );
+ aFrom.addYears( aTo.nYear - aFrom.nYear );
+ }
+
+ // move aFrom to 1st day of this month
+ nDiff += aFrom.getDaysInMonthRange( aFrom.nMonth, aTo.nMonth - 1 );
+ aFrom.addMonths( aTo.nMonth - aFrom.nMonth );
+ }
+ // finally add remaining days in this month
+ nDiff += aTo.nDay - aFrom.nDay;
+ return std::max<sal_Int32>(nDiff, 0);
+}
+
+bool ScaDate::operator<( const ScaDate& rCmp ) const
+{
+ if( nYear != rCmp.nYear )
+ return nYear < rCmp.nYear;
+ if( nMonth != rCmp.nMonth )
+ return nMonth < rCmp.nMonth;
+ if( nDay != rCmp.nDay )
+ return nDay < rCmp.nDay;
+ if( bLastDay || rCmp.bLastDay )
+ return !bLastDay && rCmp.bLastDay;
+ return nOrigDay < rCmp.nOrigDay;
+}
+
+
+ScaAnyConverter::ScaAnyConverter( const uno::Reference< uno::XComponentContext >& xContext )
+ : nDefaultFormat(0)
+ , bHasValidFormat(false)
+{
+ xFormatter = util::NumberFormatter::create(xContext);
+}
+
+ScaAnyConverter::~ScaAnyConverter()
+{
+}
+
+void ScaAnyConverter::init( const uno::Reference< beans::XPropertySet >& xPropSet )
+{
+ // try to get default number format
+ bHasValidFormat = false;
+ if( !xFormatter.is() )
+ return;
+
+ // get XFormatsSupplier from outer XPropertySet
+ uno::Reference< util::XNumberFormatsSupplier > xFormatsSupp( xPropSet, uno::UNO_QUERY );
+ if( !xFormatsSupp.is() )
+ return;
+
+ // get XNumberFormatTypes from XNumberFormatsSupplier to get standard index
+ uno::Reference< util::XNumberFormats > xFormats( xFormatsSupp->getNumberFormats() );
+ uno::Reference< util::XNumberFormatTypes > xFormatTypes( xFormats, uno::UNO_QUERY );
+ if( xFormatTypes.is() )
+ {
+ lang::Locale eLocale;
+ nDefaultFormat = xFormatTypes->getStandardIndex( eLocale );
+ xFormatter->attachNumberFormatsSupplier( xFormatsSupp );
+ bHasValidFormat = true;
+ }
+}
+
+double ScaAnyConverter::convertToDouble( const OUString& rString ) const
+{
+ double fValue = 0.0;
+ if( bHasValidFormat )
+ {
+ try
+ {
+ fValue = xFormatter->convertStringToNumber( nDefaultFormat, rString );
+ }
+ catch( uno::Exception& )
+ {
+ throw lang::IllegalArgumentException();
+ }
+ }
+ else
+ {
+ rtl_math_ConversionStatus eStatus;
+ sal_Int32 nEnd;
+ fValue = ::rtl::math::stringToDouble( rString, '.', ',', &eStatus, &nEnd );
+ if( (eStatus != rtl_math_ConversionStatus_Ok) || (nEnd < rString.getLength()) )
+ throw lang::IllegalArgumentException();
+ }
+ return fValue;
+}
+
+bool ScaAnyConverter::getDouble(
+ double& rfResult,
+ const uno::Any& rAny ) const
+{
+ rfResult = 0.0;
+ bool bContainsVal = true;
+ switch( rAny.getValueTypeClass() )
+ {
+ case uno::TypeClass_VOID:
+ bContainsVal = false;
+ break;
+ case uno::TypeClass_DOUBLE:
+ rAny >>= rfResult;
+ break;
+ case uno::TypeClass_STRING:
+ {
+ auto pString = o3tl::forceAccess< OUString >( rAny );
+ if( !pString->isEmpty() )
+ rfResult = convertToDouble( *pString );
+ else
+ bContainsVal = false;
+ }
+ break;
+ default:
+ throw lang::IllegalArgumentException();
+ }
+ return bContainsVal;
+}
+
+bool ScaAnyConverter::getDouble(
+ double& rfResult,
+ const uno::Reference< beans::XPropertySet >& xPropSet,
+ const uno::Any& rAny )
+{
+ init( xPropSet );
+ return getDouble( rfResult, rAny );
+}
+
+double ScaAnyConverter::getDouble(
+ const uno::Reference< beans::XPropertySet >& xPropSet,
+ const uno::Any& rAny,
+ double fDefault )
+{
+ double fResult;
+ if( !getDouble( fResult, xPropSet, rAny ) )
+ fResult = fDefault;
+ return fResult;
+}
+
+bool ScaAnyConverter::getInt32(
+ sal_Int32& rnResult,
+ const uno::Reference< beans::XPropertySet >& xPropSet,
+ const uno::Any& rAny )
+{
+ double fResult;
+ bool bContainsVal = getDouble( fResult, xPropSet, rAny );
+ if( (fResult <= -2147483649.0) || (fResult >= 2147483648.0) )
+ throw lang::IllegalArgumentException();
+
+ rnResult = static_cast< sal_Int32 >( fResult );
+ return bContainsVal;
+}
+
+sal_Int32 ScaAnyConverter::getInt32(
+ const uno::Reference< beans::XPropertySet >& xPropSet,
+ const uno::Any& rAny,
+ sal_Int32 nDefault )
+{
+ sal_Int32 nResult;
+ if( !getInt32( nResult, xPropSet, rAny ) )
+ nResult = nDefault;
+ return nResult;
+}
+
+}
+
+/* vim:set shiftwidth=4 softtabstop=4 expandtab: */