You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1145 lines
32 KiB
1145 lines
32 KiB
/* This file is part of the KDE project
|
|
Copyright (C) 1998-2002 The KSpread Team
|
|
www.koffice.org/kspread
|
|
Copyright (C) 2005 Tomas Mecir <mecirt@gmail.com>
|
|
|
|
This library is free software; you can redistribute it and/or
|
|
modify it under the terms of the GNU Library General Public
|
|
License as published by the Free Software Foundation; either
|
|
version 2 of the License.
|
|
|
|
This library is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Library General Public License for more details.
|
|
|
|
You should have received a copy of the GNU Library General Public License
|
|
along with this library; see the file COPYING.LIB. If not, write to
|
|
the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
|
|
* Boston, MA 02110-1301, USA.
|
|
*/
|
|
|
|
// built-in math functions
|
|
|
|
#include <kdebug.h>
|
|
#include <klocale.h>
|
|
|
|
#include "functions.h"
|
|
#include "valuecalc.h"
|
|
#include "valueconverter.h"
|
|
|
|
// these two are needed for SUBTOTAL:
|
|
#include "kspread_cell.h"
|
|
#include "kspread_sheet.h"
|
|
|
|
// needed for RANDBINOM and so
|
|
#include <math.h>
|
|
|
|
using namespace KSpread;
|
|
|
|
// RANDBINOM and RANDNEGBINOM won't support arbitrary precision
|
|
|
|
// prototypes
|
|
Value func_abs (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_ceil (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_ceiling (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_count (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_counta (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_countblank (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_countif (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_cur (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_div (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_eps (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_even (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_exp (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_fact (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_factdouble (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_fib (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_floor (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_gcd (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_int (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_inv (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_kproduct (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_lcm (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_ln (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_log2 (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_log10 (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_logn (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_max (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_maxa (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_mdeterm (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_min (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_mina (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_mmult (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_mod (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_mround (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_mult (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_multinomial (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_odd (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_pow (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_quotient (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_product (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_rand (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_randbetween (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_randbernoulli (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_randbinom (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_randexp (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_randnegbinom (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_randnorm (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_randpoisson (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_rootn (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_round (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_rounddown (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_roundup (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_sign (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_sqrt (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_sqrtpi (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_subtotal (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_sum (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_suma (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_sumif (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_sumsq (valVector args, ValueCalc *calc, FuncExtra *);
|
|
Value func_trunc (valVector args, ValueCalc *calc, FuncExtra *);
|
|
|
|
|
|
// Value func_multipleOP (valVector args, ValueCalc *calc, FuncExtra *);
|
|
|
|
// registers all math functions
|
|
void RegisterMathFunctions()
|
|
{
|
|
FunctionRepository* repo = FunctionRepository::self();
|
|
Function *f;
|
|
|
|
/*
|
|
f = new Function ("MULTIPLEOPERATIONS", func_multipleOP);
|
|
repo->add (f);
|
|
*/
|
|
|
|
// functions that don't take array parameters
|
|
f = new Function ("ABS", func_abs);
|
|
repo->add (f);
|
|
f = new Function ("CEIL", func_ceil);
|
|
repo->add (f);
|
|
f = new Function ("CEILING", func_ceiling);
|
|
f->setParamCount (1, 2);
|
|
repo->add (f);
|
|
f = new Function ("CUR", func_cur);
|
|
repo->add (f);
|
|
f = new Function ("EPS", func_eps);
|
|
f->setParamCount (0);
|
|
repo->add (f);
|
|
f = new Function ("EVEN", func_even);
|
|
repo->add (f);
|
|
f = new Function ("EXP", func_exp);
|
|
repo->add (f);
|
|
f = new Function ("FACT", func_fact);
|
|
repo->add (f);
|
|
f = new Function ("FACTDOUBLE", func_factdouble);
|
|
repo->add (f);
|
|
f = new Function ("FIB", func_fib); // KSpread-specific, like Quattro-Pro's FIB
|
|
repo->add (f);
|
|
f = new Function ("FLOOR", func_floor);
|
|
repo->add (f);
|
|
f = new Function ("INT", func_int);
|
|
repo->add (f);
|
|
f = new Function ("INV", func_inv);
|
|
repo->add (f);
|
|
f = new Function ("LN", func_ln);
|
|
repo->add (f);
|
|
f = new Function ("LOG", func_log10);
|
|
repo->add (f);
|
|
f = new Function ("LOG2", func_log2);
|
|
repo->add (f);
|
|
f = new Function ("LOG10", func_log10); // same as LOG
|
|
repo->add (f);
|
|
f = new Function ("LOGN", func_logn);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("MOD", func_mod);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("MROUND", func_mround);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("MULTINOMIAL", func_multinomial);
|
|
f->setParamCount (1, -1);
|
|
repo->add (f);
|
|
f = new Function ("ODD", func_odd);
|
|
repo->add (f);
|
|
f = new Function ("POW", func_pow);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("POWER", func_pow);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("QUOTIENT", func_quotient);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("RAND", func_rand);
|
|
f->setParamCount (0);
|
|
repo->add (f);
|
|
f = new Function ("RANDBERNOULLI", func_randbernoulli);
|
|
repo->add (f);
|
|
f = new Function ("RANDBETWEEN", func_randbetween);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("RANDBINOM", func_randbinom);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("RANDEXP", func_randexp);
|
|
repo->add (f);
|
|
f = new Function ("RANDNEGBINOM", func_randnegbinom);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("RANDNORM", func_randnorm);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("RANDPOISSON", func_randpoisson);
|
|
repo->add (f);
|
|
f = new Function ("ROOTN", func_rootn);
|
|
f->setParamCount (2);
|
|
repo->add (f);
|
|
f = new Function ("ROUND", func_round);
|
|
f->setParamCount (1, 2);
|
|
repo->add (f);
|
|
f = new Function ("ROUNDDOWN", func_rounddown);
|
|
f->setParamCount (1, 2);
|
|
repo->add (f);
|
|
f = new Function ("ROUNDUP", func_roundup);
|
|
f->setParamCount (1, 2);
|
|
repo->add (f);
|
|
f = new Function ("SIGN", func_sign);
|
|
repo->add (f);
|
|
f = new Function ("SQRT", func_sqrt);
|
|
repo->add (f);
|
|
f = new Function ("SQRTPI", func_sqrtpi);
|
|
repo->add (f);
|
|
f = new Function ("TRUNC", func_trunc);
|
|
f->setParamCount (1, 2);
|
|
repo->add (f);
|
|
|
|
// functions that operate over arrays
|
|
f = new Function ("COUNT", func_count);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("COUNTA", func_counta);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("COUNTBLANK", func_countblank);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("COUNTIF", func_countif);
|
|
f->setParamCount (2);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("DIV", func_div);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("G_PRODUCT", func_kproduct); // Gnumeric compatibility
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("GCD", func_gcd);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("KPRODUCT", func_kproduct);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("LCM", func_lcm);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("MAX", func_max);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("MAXA", func_maxa);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("MDETERM", func_mdeterm);
|
|
f->setParamCount (1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("MIN", func_min);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("MINA", func_mina);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("MMULT", func_mmult);
|
|
f->setParamCount (2);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("MULTIPLY", func_product); // same as PRODUCT
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("PRODUCT", func_product);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("SUM", func_sum);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("SUMA", func_suma);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("SUBTOTAL", func_subtotal);
|
|
f->setParamCount (2);
|
|
f->setAcceptArray ();
|
|
f->setNeedsExtra (true);
|
|
repo->add (f);
|
|
f = new Function ("SUMIF", func_sumif);
|
|
f->setParamCount (2, 3);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
f = new Function ("SUMSQ", func_sumsq);
|
|
f->setParamCount (1, -1);
|
|
f->setAcceptArray ();
|
|
repo->add (f);
|
|
}
|
|
|
|
// Function: SQRT
|
|
Value func_sqrt (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->sqrt (args[0]);
|
|
}
|
|
|
|
// Function: SQRTPI
|
|
Value func_sqrtpi (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
// sqrt (val * PI)
|
|
return calc->sqrt (calc->mul (args[0], calc->pi()));
|
|
}
|
|
|
|
// Function: ROOTN
|
|
Value func_rootn (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->pow (args[0], calc->div (1, args[1]));
|
|
}
|
|
|
|
// Function: CUR
|
|
Value func_cur (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->pow (args[0], 1.0/3.0);
|
|
}
|
|
|
|
// Function: ABS
|
|
Value func_abs (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->abs (args[0]);
|
|
}
|
|
|
|
// Function: exp
|
|
Value func_exp (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->exp (args[0]);
|
|
}
|
|
|
|
// Function: ceil
|
|
Value func_ceil (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->roundUp (args[0], 0);
|
|
}
|
|
|
|
// Function: ceiling
|
|
Value func_ceiling (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value number = args[0];
|
|
Value res;
|
|
if (args.count() == 2)
|
|
res = args[1];
|
|
else
|
|
res = calc->gequal (number, 0.0) ? 1.0 : -1.0;
|
|
|
|
if (calc->isZero(res))
|
|
return Value::errorDIV0();
|
|
|
|
Value d = calc->div (number, res);
|
|
if (calc->greater (0, d))
|
|
return Value::errorVALUE();
|
|
|
|
Value rud = calc->roundDown (d);
|
|
if (calc->approxEqual (rud, d))
|
|
d = calc->mul (rud, res);
|
|
else
|
|
d = calc->mul (calc->roundUp (d), res);
|
|
|
|
return d;
|
|
}
|
|
|
|
// Function: floor
|
|
Value func_floor (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->roundDown (args[0], 0);
|
|
}
|
|
|
|
// Function: ln
|
|
Value func_ln (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->ln (args[0]);
|
|
}
|
|
|
|
// Function: LOGn
|
|
Value func_logn (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->log (args[0], args[1]);
|
|
}
|
|
|
|
// Function: LOG2
|
|
Value func_log2 (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->log (args[0], 2.0);
|
|
}
|
|
|
|
// Function: LOG10
|
|
Value func_log10 (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->log (args[0]);
|
|
}
|
|
|
|
// Function: sum
|
|
Value func_sum (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->sum (args, false);
|
|
}
|
|
|
|
// Function: suma
|
|
Value func_suma (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->sum (args, true);
|
|
}
|
|
|
|
Value func_sumif (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value checkRange = args[0];
|
|
TQString condition = calc->conv()->asString (args[1]).asString();
|
|
Value sumRange = checkRange;
|
|
if (args.count() == 3)
|
|
sumRange = args[2];
|
|
|
|
Condition cond;
|
|
calc->getCond (cond, condition);
|
|
|
|
return calc->sumIf (sumRange, checkRange, cond);
|
|
}
|
|
|
|
// Function: product
|
|
Value func_product (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->product (args, 0.0);
|
|
}
|
|
|
|
// Function: kproduct
|
|
Value func_kproduct (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->product (args, 1.0);
|
|
}
|
|
|
|
// Function: DIV
|
|
Value func_div (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value val = args[0];
|
|
for (unsigned int i = 1; i < args.count(); ++i)
|
|
{
|
|
val = calc->div (val, args[i]);
|
|
if (val.isError())
|
|
return val;
|
|
}
|
|
return val;
|
|
}
|
|
|
|
// Function: SUMSQ
|
|
Value func_sumsq (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value res;
|
|
calc->arrayWalk (args, res, calc->awFunc ("sumsq"), 0);
|
|
return res;
|
|
}
|
|
|
|
// Function: MAX
|
|
Value func_max (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value m = calc->max (args, false);
|
|
return m.isEmpty() ? Value(0.0) : m;
|
|
}
|
|
|
|
// Function: MAXA
|
|
Value func_maxa (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value m = calc->max (args);
|
|
return m.isEmpty() ? Value(0.0) : m;
|
|
}
|
|
|
|
// Function: MIN
|
|
Value func_min (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value m = calc->min (args, false);
|
|
return m.isEmpty() ? Value(0.0) : m;
|
|
}
|
|
|
|
// Function: MINA
|
|
Value func_mina (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value m = calc->min (args);
|
|
return m.isEmpty() ? Value(0.0) : m;
|
|
}
|
|
|
|
// Function: INT
|
|
Value func_int (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->conv()->asInteger (args[0]);
|
|
}
|
|
|
|
// Function: QUOTIENT
|
|
Value func_quotient (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
if (calc->isZero (args[1]))
|
|
return Value::errorDIV0();
|
|
return calc->conv()->asInteger (calc->div (args[0], args[1]));
|
|
}
|
|
|
|
|
|
// Function: eps
|
|
Value func_eps (valVector, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->eps ();
|
|
}
|
|
|
|
Value func_randexp (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
// -1 * d * log (random)
|
|
return calc->mul (calc->mul (args[0], -1), calc->random());
|
|
}
|
|
|
|
Value func_randbinom (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
// this function will not support arbitrary precision
|
|
|
|
double d = calc->conv()->asFloat (args[0]).asFloat();
|
|
int tr = calc->conv()->asInteger (args[1]).asInteger();
|
|
|
|
if ( d < 0 || d > 1 )
|
|
return Value::errorVALUE();
|
|
|
|
if ( tr < 0 )
|
|
return Value::errorVALUE();
|
|
|
|
// taken from gnumeric
|
|
double x = pow(1 - d, tr);
|
|
double r = (double) rand() / ( RAND_MAX + 1.0 );
|
|
double t = x;
|
|
int i = 0;
|
|
|
|
while (r > t)
|
|
{
|
|
x *= (((tr - i) * d) / ((1 + i) * (1 - d)));
|
|
i++;
|
|
t += x;
|
|
}
|
|
|
|
return Value (i);
|
|
}
|
|
|
|
Value func_randnegbinom (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
// this function will not support arbitrary precision
|
|
|
|
double d = calc->conv()->asFloat (args[0]).asFloat();
|
|
int f = calc->conv()->asInteger (args[1]).asInteger();
|
|
|
|
if ( d < 0 || d > 1 )
|
|
return Value::errorVALUE();
|
|
|
|
if ( f < 0 )
|
|
return Value::errorVALUE();
|
|
|
|
|
|
// taken from Gnumeric
|
|
double x = pow(d, f);
|
|
double r = (double) rand() / ( RAND_MAX + 1.0 );
|
|
double t = x;
|
|
int i = 0;
|
|
|
|
while (r > t)
|
|
{
|
|
x *= ( ( ( f + i ) * ( 1 - d ) ) / (1 + i) ) ;
|
|
i++;
|
|
t += x;
|
|
}
|
|
|
|
return Value (i);
|
|
}
|
|
|
|
Value func_randbernoulli (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value rnd = calc->random ();
|
|
return Value (calc->greater (rnd, args[0]) ? 1.0 : 0.0);
|
|
}
|
|
|
|
Value func_randnorm (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value mu = args[0];
|
|
Value sigma = args[1];
|
|
|
|
//using polar form of the Box-Muller transformation
|
|
//refer to http://www.taygeta.com/random/gaussian.html for more info
|
|
|
|
Value x1, x2, w;
|
|
do {
|
|
// x1,x2 = 2 * random() - 1
|
|
x1 = calc->random (2.0);
|
|
x2 = calc->random (2.0);
|
|
x1 = calc->sub (x1, 1);
|
|
x1 = calc->sub (x2, 1);
|
|
w = calc->add (calc->sqr(x1), calc->sqr (x2));
|
|
} while (calc->gequal (w, 1.0)); // w >= 1.0
|
|
|
|
//sqrt ((-2.0 * log (w)) / w) :
|
|
w = calc->sqrt (calc->div (calc->mul (-2.0, calc->ln (w)), w));
|
|
Value res = calc->mul (x1, w);
|
|
|
|
res = calc->add (calc->mul (res, sigma), mu); // res*sigma + mu
|
|
return res;
|
|
}
|
|
|
|
Value func_randpoisson (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
if (calc->lower (args[0], 0))
|
|
return Value::errorVALUE();
|
|
|
|
// taken from Gnumeric...
|
|
Value x = calc->exp (calc->mul (-1, args[0])); // e^(-A)
|
|
Value r = calc->random ();
|
|
Value t = x;
|
|
int i = 0;
|
|
|
|
while (calc->greater (r, t)) { // r > t
|
|
x = calc->mul (x, calc->div (args[0], i + 1)); // x *= (A/(i+1))
|
|
t = calc->add (t, x); //t += x
|
|
i++;
|
|
}
|
|
|
|
return Value (i);
|
|
}
|
|
|
|
// Function: rand
|
|
Value func_rand (valVector, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->random ();
|
|
}
|
|
|
|
// Function: RANDBETWEEN
|
|
Value func_randbetween (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value v1 = args[0];
|
|
Value v2 = args[1];
|
|
if (calc->greater (v2, v1)) {
|
|
v1 = args[1];
|
|
v2 = args[0];
|
|
}
|
|
return calc->add (v1, calc->random (calc->sub (v2, v1)));
|
|
}
|
|
|
|
// Function: POW
|
|
Value func_pow (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->pow (args[0], args[1]);
|
|
}
|
|
|
|
// Function: MOD
|
|
Value func_mod (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->mod (args[0], args[1]);
|
|
}
|
|
|
|
// Function: fact
|
|
Value func_fact (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->fact (args[0]);
|
|
}
|
|
|
|
// Function: FACTDOUBLE
|
|
Value func_factdouble (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->factDouble (args[0]);
|
|
}
|
|
|
|
// Function: MULTINOMIAL
|
|
Value func_multinomial (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
// (a+b+c)! / a!b!c! (any number of params possible)
|
|
Value num = 0, den = 1;
|
|
for (unsigned int i = 0; i < args.count(); ++i) {
|
|
num = calc->add (num, args[i]);
|
|
den = calc->mul (den, calc->fact (args[i]));
|
|
}
|
|
num = calc->fact (num);
|
|
return calc->div (num, den);
|
|
}
|
|
|
|
// Function: sign
|
|
Value func_sign (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return Value (calc->sign (args[0]));
|
|
}
|
|
|
|
// Function: INV
|
|
Value func_inv (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->mul (args[0], -1);
|
|
}
|
|
|
|
Value func_mround (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value d = args[0];
|
|
Value m = args[1];
|
|
|
|
// signs must be the same
|
|
if ((calc->greater (d, 0) && calc->lower (m, 0))
|
|
|| (calc->lower (d, 0) && calc->greater (m, 0)))
|
|
return Value::errorVALUE();
|
|
|
|
int sign = 1;
|
|
|
|
if (calc->lower (d, 0))
|
|
{
|
|
sign = -1;
|
|
d = calc->mul (d, -1);
|
|
m = calc->mul (m, -1);
|
|
}
|
|
|
|
// from gnumeric:
|
|
Value mod = calc->mod (d, m);
|
|
Value div = calc->sub (d, mod);
|
|
|
|
Value result = div;
|
|
if (calc->greater (mod, calc->div (m, 2))) // mod > m/2
|
|
result = calc->add (result, m); // result += m
|
|
result = calc->mul (result, sign); // add the sign
|
|
|
|
return result;
|
|
}
|
|
|
|
// Function: ROUNDDOWN
|
|
Value func_rounddown (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
if (args.count() == 2)
|
|
return calc->roundDown (args[0], args[1]);
|
|
return calc->roundDown (args[0], 0);
|
|
}
|
|
|
|
// Function: ROUNDUP
|
|
Value func_roundup (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
if (args.count() == 2)
|
|
return calc->roundUp (args[0], args[1]);
|
|
return calc->roundUp (args[0], 0);
|
|
}
|
|
|
|
// Function: ROUND
|
|
Value func_round (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
if (args.count() == 2)
|
|
return calc->round (args[0], args[1]);
|
|
return calc->round (args[0], 0);
|
|
}
|
|
|
|
// Function: EVEN
|
|
Value func_even (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
const Value value = calc->roundUp (args[0], 0);
|
|
return calc->isZero( calc->mod(value, 2) ) ? value : calc->add(value, 1);
|
|
}
|
|
|
|
// Function: ODD
|
|
Value func_odd (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
const Value value = calc->roundUp (args[0], 0);
|
|
return calc->isZero( calc->mod(value, 2) ) ? calc->add(value, 1) : value;
|
|
}
|
|
|
|
Value func_trunc (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
if (args.count() == 1)
|
|
return calc->roundDown (args[0]);
|
|
return calc->roundDown (args[0], args[1]);
|
|
}
|
|
|
|
// Function: COUNT
|
|
Value func_count (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->count (args, false);
|
|
}
|
|
|
|
// Function: COUNTA
|
|
Value func_counta (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
return calc->count (args);
|
|
}
|
|
|
|
// Function: COUNTBLANK
|
|
Value func_countblank (valVector args, ValueCalc *, FuncExtra *)
|
|
{
|
|
int cnt = 0;
|
|
for (unsigned int i = 0; i < args.count(); ++i)
|
|
if (args[i].isArray()) {
|
|
int rows = args[i].rows();
|
|
int cols = args[i].columns();
|
|
for (int r = 0; r < rows; ++r)
|
|
for (int c = 0; c < cols; ++c)
|
|
if (args[i].element (c, r).isEmpty())
|
|
cnt++;
|
|
} else
|
|
if (args[i].isEmpty())
|
|
cnt++;
|
|
return Value (cnt);
|
|
}
|
|
|
|
// Function: COUNTIF
|
|
Value func_countif (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value range = args[0];
|
|
TQString condition = calc->conv()->asString (args[1]).asString();
|
|
|
|
Condition cond;
|
|
calc->getCond (cond, condition);
|
|
|
|
return calc->countIf (range, cond);
|
|
}
|
|
|
|
// Function: FIB
|
|
Value func_fib (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
/*
|
|
Lucas' formula for the nth Fibonacci number F(n) is given by
|
|
|
|
((1+sqrt(5))/2)^n - ((1-sqrt(5))/2)^n
|
|
F(n) = ------------------------------------- .
|
|
sqrt(5)
|
|
|
|
*/
|
|
Value n = args[0];
|
|
Value s = calc->sqrt (5.0);
|
|
// u1 = ((1+sqrt(5))/2)^n
|
|
Value u1 = calc->pow (calc->div (calc->add (1, s), 2), n);
|
|
// u2 = ((1-sqrt(5))/2)^n
|
|
Value u2 = calc->pow (calc->div (calc->sub (1, s), 2), n);
|
|
|
|
Value result = calc->div (calc->sub (u1, u2), s);
|
|
return result;
|
|
}
|
|
|
|
static Value func_gcd_helper(const Value &val, ValueCalc *calc)
|
|
{
|
|
Value res = 0;
|
|
if (!val.isArray ())
|
|
return val;
|
|
for (unsigned int row = 0; row < val.rows(); ++row)
|
|
for (unsigned int col = 0; col < val.columns(); ++col)
|
|
{
|
|
Value v = val.element (col, row);
|
|
if (v.isArray ())
|
|
v = func_gcd_helper (v, calc);
|
|
res = calc->gcd (res, v);
|
|
}
|
|
return res;
|
|
}
|
|
|
|
// Function: GCD
|
|
Value func_gcd (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value result = 0;
|
|
for (unsigned int i = 0; i < args.count(); ++i)
|
|
if (args[i].isArray())
|
|
result = calc->gcd (result, func_gcd_helper (args[i], calc));
|
|
else
|
|
result = calc->gcd (result, args[i]);
|
|
return result;
|
|
}
|
|
|
|
static Value func_lcm_helper(const Value &val, ValueCalc *calc)
|
|
{
|
|
Value res = 0;
|
|
if (!val.isArray ())
|
|
return val;
|
|
for (unsigned int row = 0; row < val.rows(); ++row)
|
|
for (unsigned int col = 0; col < val.columns(); ++col)
|
|
{
|
|
Value v = val.element (col, row);
|
|
if (v.isArray ())
|
|
v = func_lcm_helper (v, calc);
|
|
res = calc->lcm (res, v);
|
|
}
|
|
return res;
|
|
}
|
|
|
|
// Function: lcm
|
|
Value func_lcm (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value result = 0;
|
|
for (unsigned int i = 0; i < args.count(); ++i)
|
|
if (args[i].isArray())
|
|
result = calc->lcm (result, func_lcm_helper (args[i], calc));
|
|
else
|
|
result = calc->lcm (result, args[i]);
|
|
return result;
|
|
}
|
|
|
|
Value determinant (ValueCalc *calc, Value matrix)
|
|
{
|
|
// this is a --- SLOOOW --- recursive function
|
|
// using this for something bigger than 10x10 or so = suicide :P
|
|
// but I'm too lazy to adjust gnumeric's code - remains as a TODO then
|
|
// as a note, gnumeric uses LUP decomposition to compute this
|
|
|
|
// take first row, generate smaller matrices, recursion, multiply
|
|
Value res = 0.0;
|
|
int n = matrix.columns();
|
|
if (n == 1) return matrix.element (0, 0);
|
|
if (n == 2) return calc->sub (
|
|
calc->mul (matrix.element (1,1), matrix.element (0,0)),
|
|
calc->mul (matrix.element (1,0), matrix.element (0,1)));
|
|
|
|
// n >= 3
|
|
for (int i = 0; i < n; ++i) {
|
|
Value smaller (n-1, n-1);
|
|
int col = 0;
|
|
for (int c = 0; c < n; ++c)
|
|
if (c != i) {
|
|
// copy column c to column col in new matrix
|
|
for (int r = 1; r < n; r++)
|
|
smaller.setElement (col, r-1, matrix.element (c, r));
|
|
col++;
|
|
}
|
|
Value minor = determinant (calc, smaller);
|
|
if (i % 2 == 1) minor = calc->mul (minor, -1);
|
|
res = calc->add (res, calc->mul (minor, matrix.element (i, 0)));
|
|
}
|
|
return res;
|
|
}
|
|
|
|
// Function: mdeterm
|
|
Value func_mdeterm (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value m = args[0];
|
|
unsigned r = m.rows ();
|
|
unsigned c = m.columns ();
|
|
if (r != c) // must be a square matrix
|
|
return Value::errorVALUE();
|
|
|
|
return determinant (calc, args[0]);
|
|
}
|
|
|
|
// Function: mmult
|
|
Value func_mmult (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
Value m1 = args[0];
|
|
Value m2 = args[1];
|
|
unsigned r1 = m1.rows ();
|
|
unsigned c1 = m1.columns ();
|
|
unsigned r2 = m2.rows ();
|
|
unsigned c2 = m2.columns ();
|
|
if (c1 != r2) // row/column counts must match
|
|
return Value::errorVALUE();
|
|
|
|
// create the resulting matrix
|
|
Value res (c2, r1);
|
|
|
|
// perform the multiplication - O(n^3) algorithm
|
|
for (uint row = 0; row < r1; ++row)
|
|
for (uint col = 0; col < c2; ++col) {
|
|
Value val = 0.0;
|
|
for (uint pos = 0; pos < c1; ++pos)
|
|
val = calc->add (val,
|
|
calc->mul (m1.element (pos, row), m2.element (col, pos)));
|
|
res.setElement (col, row, val);
|
|
}
|
|
return res;
|
|
}
|
|
|
|
// Function: SUBTOTAL
|
|
// This function requires access to the Sheet and so on, because
|
|
// it needs to check whether cells contain the SUBTOTAL formula or not ...
|
|
// Cells containing a SUBTOTAL formula must be ignored.
|
|
Value func_subtotal (valVector args, ValueCalc *calc, FuncExtra *e)
|
|
{
|
|
int function = calc->conv()->asInteger (args[0]).asInteger();
|
|
Value range = args[1];
|
|
int r1 = -1, c1 = -1, r2 = -1, c2 = -1;
|
|
if (e) {
|
|
r1 = e->ranges[1].row1;
|
|
c1 = e->ranges[1].col1;
|
|
r2 = e->ranges[1].row2;
|
|
c2 = e->ranges[1].col2;
|
|
}
|
|
|
|
// if we have a range, run through it, and put an empty value to the place
|
|
// of all occurences of the SUBTOTAL function
|
|
Value empty;
|
|
if ((r1 > 0) && (c1 > 0) && (r2 > 0) && (c2 > 0)) {
|
|
for (int r = r1; r <= r2; ++r)
|
|
for (int c = c1; c <= c2; ++c) {
|
|
Cell *cell = e->sheet->cellAt (c, r);
|
|
if (cell->isDefault())
|
|
continue;
|
|
if (cell->isFormula() && cell->text().find ("SUBTOTAL", 0, false) != -1)
|
|
// cell contains the word SUBTOTAL - replace value with empty
|
|
range.setElement (c-c1, r-r1, empty);
|
|
}
|
|
}
|
|
|
|
// Good. Now we can execute the necessary function on the range.
|
|
Value res;
|
|
Function *f;
|
|
valVector a;
|
|
switch (function) {
|
|
case 1: // Average
|
|
res = calc->avg (range, false);
|
|
break;
|
|
case 2: // Count
|
|
res = calc->count (range, false);
|
|
break;
|
|
case 3: // CountA
|
|
res = calc->count (range);
|
|
break;
|
|
case 4: // MAX
|
|
res = calc->max (range, false);
|
|
break;
|
|
case 5: // Min
|
|
res = calc->min (range, false);
|
|
break;
|
|
case 6: // Product
|
|
res = calc->product (range, 0.0, false);
|
|
break;
|
|
case 7: // StDev
|
|
res = calc->stddev (range, false);
|
|
break;
|
|
case 8: // StDevP
|
|
res = calc->stddevP (range, false);
|
|
break;
|
|
case 9: // Sum
|
|
res = calc->sum (range, false);
|
|
break;
|
|
case 10: // Var
|
|
f = FunctionRepository::self()->function ("VAR");
|
|
if (!f) return Value::errorVALUE();
|
|
a.reserve (1);
|
|
a[0] = range;
|
|
res = f->exec (a, calc, 0);
|
|
break;
|
|
case 11: // VarP
|
|
f = FunctionRepository::self()->function ("VARP");
|
|
if (!f) return Value::errorVALUE();
|
|
a.reserve (1);
|
|
a[0] = range;
|
|
res = f->exec (a, calc, 0);
|
|
break;
|
|
default:
|
|
return Value::errorVALUE();
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*
|
|
Commented out.
|
|
Absolutely no idea what this thing is supposed to do.
|
|
To anyone who would enable this code: it still uses koscript calls - you need
|
|
to convert it to the new style prior to uncommenting.
|
|
|
|
// Function: MULTIPLEOPERATIONS
|
|
Value func_multipleOP (valVector args, ValueCalc *calc, FuncExtra *)
|
|
{
|
|
if (gCell)
|
|
{
|
|
context.setValue( new KSValue( ((Interpreter *) context.interpreter() )->cell()->value().asFloat() ) );
|
|
return true;
|
|
}
|
|
|
|
gCell = ((Interpreter *) context.interpreter() )->cell();
|
|
|
|
TQValueList<KSValue::Ptr>& args = context.value()->listValue();
|
|
TQValueList<KSValue::Ptr>& extra = context.extraData()->listValue();
|
|
|
|
if ( !KSUtil::checkArgumentsCount( context, 5, "MULTIPLEOPERATIONS", true ) )
|
|
{
|
|
gCell = 0;
|
|
return false;
|
|
}
|
|
|
|
// 0: cell must contain formula with double/int result
|
|
// 0, 1, 2, 3, 4: must contain integer/double
|
|
for (int i = 0; i < 5; ++i)
|
|
{
|
|
if ( !KSUtil::checkType( context, args[i], KSValue::DoubleType, true ) )
|
|
{
|
|
gCell = 0;
|
|
return false;
|
|
}
|
|
}
|
|
|
|
// ((Interpreter *) context.interpreter() )->document()->emitBeginOperation();
|
|
|
|
double oldCol = args[1]->doubleValue();
|
|
double oldRow = args[3]->doubleValue();
|
|
kdDebug() << "Old values: Col: " << oldCol << ", Row: " << oldRow << endl;
|
|
|
|
Cell * cell;
|
|
Sheet * sheet = ((Interpreter *) context.interpreter() )->sheet();
|
|
|
|
Point point( extra[1]->stringValue() );
|
|
Point point2( extra[3]->stringValue() );
|
|
Point point3( extra[0]->stringValue() );
|
|
|
|
if ( ( args[1]->doubleValue() != args[2]->doubleValue() )
|
|
|| ( args[3]->doubleValue() != args[4]->doubleValue() ) )
|
|
{
|
|
cell = sheet->cellAt( point.pos.x(), point.pos.y() );
|
|
cell->setValue( args[2]->doubleValue() );
|
|
kdDebug() << "Setting value " << args[2]->doubleValue() << " on cell " << point.pos.x()
|
|
<< ", " << point.pos.y() << endl;
|
|
|
|
cell = sheet->cellAt( point2.pos.x(), point.pos.y() );
|
|
cell->setValue( args[4]->doubleValue() );
|
|
kdDebug() << "Setting value " << args[4]->doubleValue() << " on cell " << point2.pos.x()
|
|
<< ", " << point2.pos.y() << endl;
|
|
}
|
|
|
|
Cell * cell1 = sheet->cellAt( point3.pos.x(), point3.pos.y() );
|
|
cell1->calc( false );
|
|
|
|
double d = cell1->value().asFloat();
|
|
kdDebug() << "Cell: " << point3.pos.x() << "; " << point3.pos.y() << " with value "
|
|
<< d << endl;
|
|
|
|
kdDebug() << "Resetting old values" << endl;
|
|
|
|
cell = sheet->cellAt( point.pos.x(), point.pos.y() );
|
|
cell->setValue( oldCol );
|
|
|
|
cell = sheet->cellAt( point2.pos.x(), point2.pos.y() );
|
|
cell->setValue( oldRow );
|
|
|
|
cell1->calc( false );
|
|
|
|
// ((Interpreter *) context.interpreter() )->document()->emitEndOperation();
|
|
|
|
context.setValue( new KSValue( (double) d ) );
|
|
|
|
gCell = 0;
|
|
return true;
|
|
}
|
|
|
|
*/
|