forked from p30928647/excelize
5875 lines
167 KiB
Go
5875 lines
167 KiB
Go
// Copyright 2016 - 2021 The excelize Authors. All rights reserved. Use of
|
|
// this source code is governed by a BSD-style license that can be found in
|
|
// the LICENSE file.
|
|
//
|
|
// Package excelize providing a set of functions that allow you to write to
|
|
// and read from XLSX / XLSM / XLTM files. Supports reading and writing
|
|
// spreadsheet documents generated by Microsoft Exce™ 2007 and later. Supports
|
|
// complex components by high compatibility, and provided streaming API for
|
|
// generating or reading data from a worksheet with huge amounts of data. This
|
|
// library needs Go version 1.10 or later.
|
|
|
|
package excelize
|
|
|
|
import (
|
|
"bytes"
|
|
"container/list"
|
|
"errors"
|
|
"fmt"
|
|
"math"
|
|
"math/rand"
|
|
"net/url"
|
|
"reflect"
|
|
"regexp"
|
|
"sort"
|
|
"strconv"
|
|
"strings"
|
|
"time"
|
|
"unicode"
|
|
"unsafe"
|
|
|
|
"github.com/xuri/efp"
|
|
)
|
|
|
|
// Excel formula errors
|
|
const (
|
|
formulaErrorDIV = "#DIV/0!"
|
|
formulaErrorNAME = "#NAME?"
|
|
formulaErrorNA = "#N/A"
|
|
formulaErrorNUM = "#NUM!"
|
|
formulaErrorVALUE = "#VALUE!"
|
|
formulaErrorREF = "#REF!"
|
|
formulaErrorNULL = "#NULL"
|
|
formulaErrorSPILL = "#SPILL!"
|
|
formulaErrorCALC = "#CALC!"
|
|
formulaErrorGETTINGDATA = "#GETTING_DATA"
|
|
)
|
|
|
|
// Numeric precision correct numeric values as legacy Excel application
|
|
// https://en.wikipedia.org/wiki/Numeric_precision_in_Microsoft_Excel In the
|
|
// top figure the fraction 1/9000 in Excel is displayed. Although this number
|
|
// has a decimal representation that is an infinite string of ones, Excel
|
|
// displays only the leading 15 figures. In the second line, the number one
|
|
// is added to the fraction, and again Excel displays only 15 figures.
|
|
const numericPrecision = 1000000000000000
|
|
|
|
// cellRef defines the structure of a cell reference.
|
|
type cellRef struct {
|
|
Col int
|
|
Row int
|
|
Sheet string
|
|
}
|
|
|
|
// cellRef defines the structure of a cell range.
|
|
type cellRange struct {
|
|
From cellRef
|
|
To cellRef
|
|
}
|
|
|
|
// formula criteria condition enumeration.
|
|
const (
|
|
_ byte = iota
|
|
criteriaEq
|
|
criteriaLe
|
|
criteriaGe
|
|
criteriaL
|
|
criteriaG
|
|
criteriaBeg
|
|
criteriaEnd
|
|
criteriaErr
|
|
)
|
|
|
|
// formulaCriteria defined formula criteria parser result.
|
|
type formulaCriteria struct {
|
|
Type byte
|
|
Condition string
|
|
}
|
|
|
|
// ArgType is the type if formula argument type.
|
|
type ArgType byte
|
|
|
|
// Formula argument types enumeration.
|
|
const (
|
|
ArgUnknown ArgType = iota
|
|
ArgNumber
|
|
ArgString
|
|
ArgList
|
|
ArgMatrix
|
|
ArgError
|
|
ArgEmpty
|
|
)
|
|
|
|
// formulaArg is the argument of a formula or function.
|
|
type formulaArg struct {
|
|
SheetName string
|
|
Number float64
|
|
String string
|
|
List []formulaArg
|
|
Matrix [][]formulaArg
|
|
Boolean bool
|
|
Error string
|
|
Type ArgType
|
|
cellRefs, cellRanges *list.List
|
|
}
|
|
|
|
// Value returns a string data type of the formula argument.
|
|
func (fa formulaArg) Value() (value string) {
|
|
switch fa.Type {
|
|
case ArgNumber:
|
|
if fa.Boolean {
|
|
if fa.Number == 0 {
|
|
return "FALSE"
|
|
}
|
|
return "TRUE"
|
|
}
|
|
return fmt.Sprintf("%g", fa.Number)
|
|
case ArgString:
|
|
return fa.String
|
|
case ArgError:
|
|
return fa.Error
|
|
}
|
|
return
|
|
}
|
|
|
|
// ToNumber returns a formula argument with number data type.
|
|
func (fa formulaArg) ToNumber() formulaArg {
|
|
var n float64
|
|
var err error
|
|
switch fa.Type {
|
|
case ArgString:
|
|
n, err = strconv.ParseFloat(fa.String, 64)
|
|
if err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
case ArgNumber:
|
|
n = fa.Number
|
|
}
|
|
return newNumberFormulaArg(n)
|
|
}
|
|
|
|
// ToBool returns a formula argument with boolean data type.
|
|
func (fa formulaArg) ToBool() formulaArg {
|
|
var b bool
|
|
var err error
|
|
switch fa.Type {
|
|
case ArgString:
|
|
b, err = strconv.ParseBool(fa.String)
|
|
if err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
case ArgNumber:
|
|
if fa.Boolean && fa.Number == 1 {
|
|
b = true
|
|
}
|
|
}
|
|
return newBoolFormulaArg(b)
|
|
}
|
|
|
|
// ToList returns a formula argument with array data type.
|
|
func (fa formulaArg) ToList() []formulaArg {
|
|
switch fa.Type {
|
|
case ArgMatrix:
|
|
list := []formulaArg{}
|
|
for _, row := range fa.Matrix {
|
|
list = append(list, row...)
|
|
}
|
|
return list
|
|
case ArgList:
|
|
return fa.List
|
|
case ArgNumber, ArgString, ArgError, ArgUnknown:
|
|
return []formulaArg{fa}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// formulaFuncs is the type of the formula functions.
|
|
type formulaFuncs struct {
|
|
f *File
|
|
sheet, cell string
|
|
}
|
|
|
|
// tokenPriority defined basic arithmetic operator priority.
|
|
var tokenPriority = map[string]int{
|
|
"^": 5,
|
|
"*": 4,
|
|
"/": 4,
|
|
"+": 3,
|
|
"-": 3,
|
|
"=": 2,
|
|
"<>": 2,
|
|
"<": 2,
|
|
"<=": 2,
|
|
">": 2,
|
|
">=": 2,
|
|
"&": 1,
|
|
}
|
|
|
|
// CalcCellValue provides a function to get calculated cell value. This
|
|
// feature is currently in working processing. Array formula, table formula
|
|
// and some other formulas are not supported currently.
|
|
//
|
|
// Supported formula functions:
|
|
//
|
|
// ABS
|
|
// ACOS
|
|
// ACOSH
|
|
// ACOT
|
|
// ACOTH
|
|
// AND
|
|
// ARABIC
|
|
// ASIN
|
|
// ASINH
|
|
// ATAN
|
|
// ATAN2
|
|
// ATANH
|
|
// AVERAGE
|
|
// AVERAGEA
|
|
// BASE
|
|
// BIN2DEC
|
|
// BIN2HEX
|
|
// BIN2OCT
|
|
// BITAND
|
|
// BITLSHIFT
|
|
// BITOR
|
|
// BITRSHIFT
|
|
// BITXOR
|
|
// CEILING
|
|
// CEILING.MATH
|
|
// CEILING.PRECISE
|
|
// CHAR
|
|
// CHOOSE
|
|
// CLEAN
|
|
// CODE
|
|
// COLUMN
|
|
// COLUMNS
|
|
// COMBIN
|
|
// COMBINA
|
|
// CONCAT
|
|
// CONCATENATE
|
|
// COS
|
|
// COSH
|
|
// COT
|
|
// COTH
|
|
// COUNT
|
|
// COUNTA
|
|
// COUNTBLANK
|
|
// CSC
|
|
// CSCH
|
|
// DATE
|
|
// DEC2BIN
|
|
// DEC2HEX
|
|
// DEC2OCT
|
|
// DECIMAL
|
|
// DEGREES
|
|
// ENCODEURL
|
|
// EVEN
|
|
// EXACT
|
|
// EXP
|
|
// FACT
|
|
// FACTDOUBLE
|
|
// FALSE
|
|
// FIND
|
|
// FINDB
|
|
// FISHER
|
|
// FISHERINV
|
|
// FLOOR
|
|
// FLOOR.MATH
|
|
// FLOOR.PRECISE
|
|
// GAMMA
|
|
// GAMMALN
|
|
// GCD
|
|
// HEX2BIN
|
|
// HEX2DEC
|
|
// HEX2OCT
|
|
// HLOOKUP
|
|
// IF
|
|
// IFERROR
|
|
// INT
|
|
// ISBLANK
|
|
// ISERR
|
|
// ISERROR
|
|
// ISEVEN
|
|
// ISNA
|
|
// ISNONTEXT
|
|
// ISNUMBER
|
|
// ISODD
|
|
// ISTEXT
|
|
// ISO.CEILING
|
|
// KURT
|
|
// LARGE
|
|
// LCM
|
|
// LEFT
|
|
// LEFTB
|
|
// LEN
|
|
// LENB
|
|
// LN
|
|
// LOG
|
|
// LOG10
|
|
// LOOKUP
|
|
// LOWER
|
|
// MAX
|
|
// MDETERM
|
|
// MEDIAN
|
|
// MID
|
|
// MIDB
|
|
// MIN
|
|
// MINA
|
|
// MOD
|
|
// MROUND
|
|
// MULTINOMIAL
|
|
// MUNIT
|
|
// NA
|
|
// NOT
|
|
// NOW
|
|
// OCT2BIN
|
|
// OCT2DEC
|
|
// OCT2HEX
|
|
// ODD
|
|
// OR
|
|
// PERMUT
|
|
// PI
|
|
// POWER
|
|
// PRODUCT
|
|
// PROPER
|
|
// QUOTIENT
|
|
// RADIANS
|
|
// RAND
|
|
// RANDBETWEEN
|
|
// REPLACE
|
|
// REPLACEB
|
|
// REPT
|
|
// RIGHT
|
|
// RIGHTB
|
|
// ROMAN
|
|
// ROUND
|
|
// ROUNDDOWN
|
|
// ROUNDUP
|
|
// ROW
|
|
// ROWS
|
|
// SEC
|
|
// SECH
|
|
// SHEET
|
|
// SIGN
|
|
// SIN
|
|
// SINH
|
|
// SMALL
|
|
// SQRT
|
|
// SQRTPI
|
|
// STDEV
|
|
// STDEVA
|
|
// SUBSTITUTE
|
|
// SUM
|
|
// SUMIF
|
|
// SUMSQ
|
|
// TAN
|
|
// TANH
|
|
// TODAY
|
|
// TRIM
|
|
// TRUE
|
|
// TRUNC
|
|
// UNICHAR
|
|
// UNICODE
|
|
// UPPER
|
|
// VLOOKUP
|
|
//
|
|
func (f *File) CalcCellValue(sheet, cell string) (result string, err error) {
|
|
var (
|
|
formula string
|
|
token efp.Token
|
|
)
|
|
if formula, err = f.GetCellFormula(sheet, cell); err != nil {
|
|
return
|
|
}
|
|
ps := efp.ExcelParser()
|
|
tokens := ps.Parse(formula)
|
|
if tokens == nil {
|
|
return
|
|
}
|
|
if token, err = f.evalInfixExp(sheet, cell, tokens); err != nil {
|
|
return
|
|
}
|
|
result = token.TValue
|
|
isNum, precision := isNumeric(result)
|
|
if isNum && precision > 15 {
|
|
num, _ := roundPrecision(result)
|
|
result = strings.ToUpper(num)
|
|
}
|
|
return
|
|
}
|
|
|
|
// getPriority calculate arithmetic operator priority.
|
|
func getPriority(token efp.Token) (pri int) {
|
|
pri = tokenPriority[token.TValue]
|
|
if token.TValue == "-" && token.TType == efp.TokenTypeOperatorPrefix {
|
|
pri = 6
|
|
}
|
|
if isBeginParenthesesToken(token) { // (
|
|
pri = 0
|
|
}
|
|
return
|
|
}
|
|
|
|
// newNumberFormulaArg constructs a number formula argument.
|
|
func newNumberFormulaArg(n float64) formulaArg {
|
|
if math.IsNaN(n) {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
return formulaArg{Type: ArgNumber, Number: n}
|
|
}
|
|
|
|
// newStringFormulaArg constructs a string formula argument.
|
|
func newStringFormulaArg(s string) formulaArg {
|
|
return formulaArg{Type: ArgString, String: s}
|
|
}
|
|
|
|
// newMatrixFormulaArg constructs a matrix formula argument.
|
|
func newMatrixFormulaArg(m [][]formulaArg) formulaArg {
|
|
return formulaArg{Type: ArgMatrix, Matrix: m}
|
|
}
|
|
|
|
// newListFormulaArg create a list formula argument.
|
|
func newListFormulaArg(l []formulaArg) formulaArg {
|
|
return formulaArg{Type: ArgList, List: l}
|
|
}
|
|
|
|
// newBoolFormulaArg constructs a boolean formula argument.
|
|
func newBoolFormulaArg(b bool) formulaArg {
|
|
var n float64
|
|
if b {
|
|
n = 1
|
|
}
|
|
return formulaArg{Type: ArgNumber, Number: n, Boolean: true}
|
|
}
|
|
|
|
// newErrorFormulaArg create an error formula argument of a given type with a
|
|
// specified error message.
|
|
func newErrorFormulaArg(formulaError, msg string) formulaArg {
|
|
return formulaArg{Type: ArgError, String: formulaError, Error: msg}
|
|
}
|
|
|
|
// newEmptyFormulaArg create an empty formula argument.
|
|
func newEmptyFormulaArg() formulaArg {
|
|
return formulaArg{Type: ArgEmpty}
|
|
}
|
|
|
|
// evalInfixExp evaluate syntax analysis by given infix expression after
|
|
// lexical analysis. Evaluate an infix expression containing formulas by
|
|
// stacks:
|
|
//
|
|
// opd - Operand
|
|
// opt - Operator
|
|
// opf - Operation formula
|
|
// opfd - Operand of the operation formula
|
|
// opft - Operator of the operation formula
|
|
// args - Arguments list of the operation formula
|
|
//
|
|
// TODO: handle subtypes: Nothing, Text, Logical, Error, Concatenation, Intersection, Union
|
|
//
|
|
func (f *File) evalInfixExp(sheet, cell string, tokens []efp.Token) (efp.Token, error) {
|
|
var err error
|
|
opdStack, optStack, opfStack, opfdStack, opftStack, argsStack := NewStack(), NewStack(), NewStack(), NewStack(), NewStack(), NewStack()
|
|
for i := 0; i < len(tokens); i++ {
|
|
token := tokens[i]
|
|
|
|
// out of function stack
|
|
if opfStack.Len() == 0 {
|
|
if err = f.parseToken(sheet, token, opdStack, optStack); err != nil {
|
|
return efp.Token{}, err
|
|
}
|
|
}
|
|
|
|
// function start
|
|
if isFunctionStartToken(token) {
|
|
opfStack.Push(token)
|
|
argsStack.Push(list.New().Init())
|
|
continue
|
|
}
|
|
|
|
// in function stack, walk 2 token at once
|
|
if opfStack.Len() > 0 {
|
|
var nextToken efp.Token
|
|
if i+1 < len(tokens) {
|
|
nextToken = tokens[i+1]
|
|
}
|
|
|
|
// current token is args or range, skip next token, order required: parse reference first
|
|
if token.TSubType == efp.TokenSubTypeRange {
|
|
if !opftStack.Empty() {
|
|
// parse reference: must reference at here
|
|
result, err := f.parseReference(sheet, token.TValue)
|
|
if err != nil {
|
|
return efp.Token{TValue: formulaErrorNAME}, err
|
|
}
|
|
if result.Type != ArgString {
|
|
return efp.Token{}, errors.New(formulaErrorVALUE)
|
|
}
|
|
opfdStack.Push(efp.Token{
|
|
TType: efp.TokenTypeOperand,
|
|
TSubType: efp.TokenSubTypeNumber,
|
|
TValue: result.String,
|
|
})
|
|
continue
|
|
}
|
|
if nextToken.TType == efp.TokenTypeArgument || nextToken.TType == efp.TokenTypeFunction {
|
|
// parse reference: reference or range at here
|
|
result, err := f.parseReference(sheet, token.TValue)
|
|
if err != nil {
|
|
return efp.Token{TValue: formulaErrorNAME}, err
|
|
}
|
|
if result.Type == ArgUnknown {
|
|
return efp.Token{}, errors.New(formulaErrorVALUE)
|
|
}
|
|
argsStack.Peek().(*list.List).PushBack(result)
|
|
continue
|
|
}
|
|
}
|
|
|
|
// check current token is opft
|
|
if err = f.parseToken(sheet, token, opfdStack, opftStack); err != nil {
|
|
return efp.Token{}, err
|
|
}
|
|
|
|
// current token is arg
|
|
if token.TType == efp.TokenTypeArgument {
|
|
for !opftStack.Empty() {
|
|
// calculate trigger
|
|
topOpt := opftStack.Peek().(efp.Token)
|
|
if err := calculate(opfdStack, topOpt); err != nil {
|
|
argsStack.Peek().(*list.List).PushFront(newErrorFormulaArg(formulaErrorVALUE, err.Error()))
|
|
}
|
|
opftStack.Pop()
|
|
}
|
|
if !opfdStack.Empty() {
|
|
argsStack.Peek().(*list.List).PushBack(newStringFormulaArg(opfdStack.Pop().(efp.Token).TValue))
|
|
}
|
|
continue
|
|
}
|
|
|
|
// current token is logical
|
|
if token.TType == efp.OperatorsInfix && token.TSubType == efp.TokenSubTypeLogical {
|
|
}
|
|
if token.TType == efp.TokenTypeOperand && token.TSubType == efp.TokenSubTypeLogical {
|
|
argsStack.Peek().(*list.List).PushBack(newStringFormulaArg(token.TValue))
|
|
}
|
|
|
|
// current token is text
|
|
if token.TType == efp.TokenTypeOperand && token.TSubType == efp.TokenSubTypeText {
|
|
argsStack.Peek().(*list.List).PushBack(newStringFormulaArg(token.TValue))
|
|
}
|
|
if err = f.evalInfixExpFunc(sheet, cell, token, nextToken, opfStack, opdStack, opftStack, opfdStack, argsStack); err != nil {
|
|
return efp.Token{}, err
|
|
}
|
|
}
|
|
}
|
|
for optStack.Len() != 0 {
|
|
topOpt := optStack.Peek().(efp.Token)
|
|
if err = calculate(opdStack, topOpt); err != nil {
|
|
return efp.Token{}, err
|
|
}
|
|
optStack.Pop()
|
|
}
|
|
if opdStack.Len() == 0 {
|
|
return efp.Token{}, errors.New("formula not valid")
|
|
}
|
|
return opdStack.Peek().(efp.Token), err
|
|
}
|
|
|
|
// evalInfixExpFunc evaluate formula function in the infix expression.
|
|
func (f *File) evalInfixExpFunc(sheet, cell string, token, nextToken efp.Token, opfStack, opdStack, opftStack, opfdStack, argsStack *Stack) error {
|
|
if !isFunctionStopToken(token) {
|
|
return nil
|
|
}
|
|
// current token is function stop
|
|
for !opftStack.Empty() {
|
|
// calculate trigger
|
|
topOpt := opftStack.Peek().(efp.Token)
|
|
if err := calculate(opfdStack, topOpt); err != nil {
|
|
return err
|
|
}
|
|
opftStack.Pop()
|
|
}
|
|
|
|
// push opfd to args
|
|
if opfdStack.Len() > 0 {
|
|
argsStack.Peek().(*list.List).PushBack(newStringFormulaArg(opfdStack.Pop().(efp.Token).TValue))
|
|
}
|
|
// call formula function to evaluate
|
|
arg := callFuncByName(&formulaFuncs{f: f, sheet: sheet, cell: cell}, strings.NewReplacer(
|
|
"_xlfn", "", ".", "").Replace(opfStack.Peek().(efp.Token).TValue),
|
|
[]reflect.Value{reflect.ValueOf(argsStack.Peek().(*list.List))})
|
|
if arg.Type == ArgError && opfStack.Len() == 1 {
|
|
return errors.New(arg.Value())
|
|
}
|
|
argsStack.Pop()
|
|
opfStack.Pop()
|
|
if opfStack.Len() > 0 { // still in function stack
|
|
if nextToken.TType == efp.TokenTypeOperatorInfix {
|
|
// mathematics calculate in formula function
|
|
opfdStack.Push(efp.Token{TValue: arg.Value(), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
} else {
|
|
argsStack.Peek().(*list.List).PushBack(arg)
|
|
}
|
|
} else {
|
|
opdStack.Push(efp.Token{TValue: arg.Value(), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// calcPow evaluate exponentiation arithmetic operations.
|
|
func calcPow(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
result := math.Pow(lOpdVal, rOpdVal)
|
|
opdStack.Push(efp.Token{TValue: fmt.Sprintf("%g", result), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcEq evaluate equal arithmetic operations.
|
|
func calcEq(rOpd, lOpd string, opdStack *Stack) error {
|
|
opdStack.Push(efp.Token{TValue: strings.ToUpper(strconv.FormatBool(rOpd == lOpd)), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcNEq evaluate not equal arithmetic operations.
|
|
func calcNEq(rOpd, lOpd string, opdStack *Stack) error {
|
|
opdStack.Push(efp.Token{TValue: strings.ToUpper(strconv.FormatBool(rOpd != lOpd)), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcL evaluate less than arithmetic operations.
|
|
func calcL(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
opdStack.Push(efp.Token{TValue: strings.ToUpper(strconv.FormatBool(rOpdVal > lOpdVal)), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcLe evaluate less than or equal arithmetic operations.
|
|
func calcLe(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
opdStack.Push(efp.Token{TValue: strings.ToUpper(strconv.FormatBool(rOpdVal >= lOpdVal)), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcG evaluate greater than or equal arithmetic operations.
|
|
func calcG(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
opdStack.Push(efp.Token{TValue: strings.ToUpper(strconv.FormatBool(rOpdVal < lOpdVal)), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcGe evaluate greater than or equal arithmetic operations.
|
|
func calcGe(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
opdStack.Push(efp.Token{TValue: strings.ToUpper(strconv.FormatBool(rOpdVal <= lOpdVal)), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcSplice evaluate splice '&' operations.
|
|
func calcSplice(rOpd, lOpd string, opdStack *Stack) error {
|
|
opdStack.Push(efp.Token{TValue: lOpd + rOpd, TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcAdd evaluate addition arithmetic operations.
|
|
func calcAdd(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
result := lOpdVal + rOpdVal
|
|
opdStack.Push(efp.Token{TValue: fmt.Sprintf("%g", result), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcSubtract evaluate subtraction arithmetic operations.
|
|
func calcSubtract(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
result := lOpdVal - rOpdVal
|
|
opdStack.Push(efp.Token{TValue: fmt.Sprintf("%g", result), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcMultiply evaluate multiplication arithmetic operations.
|
|
func calcMultiply(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
result := lOpdVal * rOpdVal
|
|
opdStack.Push(efp.Token{TValue: fmt.Sprintf("%g", result), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calcDiv evaluate division arithmetic operations.
|
|
func calcDiv(rOpd, lOpd string, opdStack *Stack) error {
|
|
lOpdVal, err := strconv.ParseFloat(lOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
rOpdVal, err := strconv.ParseFloat(rOpd, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
result := lOpdVal / rOpdVal
|
|
if rOpdVal == 0 {
|
|
return errors.New(formulaErrorDIV)
|
|
}
|
|
opdStack.Push(efp.Token{TValue: fmt.Sprintf("%g", result), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
return nil
|
|
}
|
|
|
|
// calculate evaluate basic arithmetic operations.
|
|
func calculate(opdStack *Stack, opt efp.Token) error {
|
|
if opt.TValue == "-" && opt.TType == efp.TokenTypeOperatorPrefix {
|
|
if opdStack.Len() < 1 {
|
|
return errors.New("formula not valid")
|
|
}
|
|
opd := opdStack.Pop().(efp.Token)
|
|
opdVal, err := strconv.ParseFloat(opd.TValue, 64)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
result := 0 - opdVal
|
|
opdStack.Push(efp.Token{TValue: fmt.Sprintf("%g", result), TType: efp.TokenTypeOperand, TSubType: efp.TokenSubTypeNumber})
|
|
}
|
|
tokenCalcFunc := map[string]func(rOpd, lOpd string, opdStack *Stack) error{
|
|
"^": calcPow,
|
|
"*": calcMultiply,
|
|
"/": calcDiv,
|
|
"+": calcAdd,
|
|
"=": calcEq,
|
|
"<>": calcNEq,
|
|
"<": calcL,
|
|
"<=": calcLe,
|
|
">": calcG,
|
|
">=": calcGe,
|
|
"&": calcSplice,
|
|
}
|
|
if opt.TValue == "-" && opt.TType == efp.TokenTypeOperatorInfix {
|
|
if opdStack.Len() < 2 {
|
|
return errors.New("formula not valid")
|
|
}
|
|
rOpd := opdStack.Pop().(efp.Token)
|
|
lOpd := opdStack.Pop().(efp.Token)
|
|
if err := calcSubtract(rOpd.TValue, lOpd.TValue, opdStack); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
fn, ok := tokenCalcFunc[opt.TValue]
|
|
if ok {
|
|
if opdStack.Len() < 2 {
|
|
return errors.New("formula not valid")
|
|
}
|
|
rOpd := opdStack.Pop().(efp.Token)
|
|
lOpd := opdStack.Pop().(efp.Token)
|
|
if err := fn(rOpd.TValue, lOpd.TValue, opdStack); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// parseOperatorPrefixToken parse operator prefix token.
|
|
func (f *File) parseOperatorPrefixToken(optStack, opdStack *Stack, token efp.Token) (err error) {
|
|
if optStack.Len() == 0 {
|
|
optStack.Push(token)
|
|
} else {
|
|
tokenPriority := getPriority(token)
|
|
topOpt := optStack.Peek().(efp.Token)
|
|
topOptPriority := getPriority(topOpt)
|
|
if tokenPriority > topOptPriority {
|
|
optStack.Push(token)
|
|
} else {
|
|
for tokenPriority <= topOptPriority {
|
|
optStack.Pop()
|
|
if err = calculate(opdStack, topOpt); err != nil {
|
|
return
|
|
}
|
|
if optStack.Len() > 0 {
|
|
topOpt = optStack.Peek().(efp.Token)
|
|
topOptPriority = getPriority(topOpt)
|
|
continue
|
|
}
|
|
break
|
|
}
|
|
optStack.Push(token)
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// isFunctionStartToken determine if the token is function stop.
|
|
func isFunctionStartToken(token efp.Token) bool {
|
|
return token.TType == efp.TokenTypeFunction && token.TSubType == efp.TokenSubTypeStart
|
|
}
|
|
|
|
// isFunctionStopToken determine if the token is function stop.
|
|
func isFunctionStopToken(token efp.Token) bool {
|
|
return token.TType == efp.TokenTypeFunction && token.TSubType == efp.TokenSubTypeStop
|
|
}
|
|
|
|
// isBeginParenthesesToken determine if the token is begin parentheses: (.
|
|
func isBeginParenthesesToken(token efp.Token) bool {
|
|
return token.TType == efp.TokenTypeSubexpression && token.TSubType == efp.TokenSubTypeStart
|
|
}
|
|
|
|
// isEndParenthesesToken determine if the token is end parentheses: ).
|
|
func isEndParenthesesToken(token efp.Token) bool {
|
|
return token.TType == efp.TokenTypeSubexpression && token.TSubType == efp.TokenSubTypeStop
|
|
}
|
|
|
|
// isOperatorPrefixToken determine if the token is parse operator prefix
|
|
// token.
|
|
func isOperatorPrefixToken(token efp.Token) bool {
|
|
_, ok := tokenPriority[token.TValue]
|
|
if (token.TValue == "-" && token.TType == efp.TokenTypeOperatorPrefix) || (ok && token.TType == efp.TokenTypeOperatorInfix) {
|
|
return true
|
|
}
|
|
return false
|
|
}
|
|
|
|
// getDefinedNameRefTo convert defined name to reference range.
|
|
func (f *File) getDefinedNameRefTo(definedNameName string, currentSheet string) (refTo string) {
|
|
for _, definedName := range f.GetDefinedName() {
|
|
if definedName.Name == definedNameName {
|
|
refTo = definedName.RefersTo
|
|
// worksheet scope takes precedence over scope workbook when both definedNames exist
|
|
if definedName.Scope == currentSheet {
|
|
break
|
|
}
|
|
}
|
|
}
|
|
return refTo
|
|
}
|
|
|
|
// parseToken parse basic arithmetic operator priority and evaluate based on
|
|
// operators and operands.
|
|
func (f *File) parseToken(sheet string, token efp.Token, opdStack, optStack *Stack) error {
|
|
// parse reference: must reference at here
|
|
if token.TSubType == efp.TokenSubTypeRange {
|
|
refTo := f.getDefinedNameRefTo(token.TValue, sheet)
|
|
if refTo != "" {
|
|
token.TValue = refTo
|
|
}
|
|
result, err := f.parseReference(sheet, token.TValue)
|
|
if err != nil {
|
|
return errors.New(formulaErrorNAME)
|
|
}
|
|
if result.Type != ArgString {
|
|
return errors.New(formulaErrorVALUE)
|
|
}
|
|
token.TValue = result.String
|
|
token.TType = efp.TokenTypeOperand
|
|
token.TSubType = efp.TokenSubTypeNumber
|
|
}
|
|
if isOperatorPrefixToken(token) {
|
|
if err := f.parseOperatorPrefixToken(optStack, opdStack, token); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
if isBeginParenthesesToken(token) { // (
|
|
optStack.Push(token)
|
|
}
|
|
if isEndParenthesesToken(token) { // )
|
|
for !isBeginParenthesesToken(optStack.Peek().(efp.Token)) { // != (
|
|
topOpt := optStack.Peek().(efp.Token)
|
|
if err := calculate(opdStack, topOpt); err != nil {
|
|
return err
|
|
}
|
|
optStack.Pop()
|
|
}
|
|
optStack.Pop()
|
|
}
|
|
// opd
|
|
if token.TType == efp.TokenTypeOperand && token.TSubType == efp.TokenSubTypeNumber {
|
|
opdStack.Push(token)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// parseReference parse reference and extract values by given reference
|
|
// characters and default sheet name.
|
|
func (f *File) parseReference(sheet, reference string) (arg formulaArg, err error) {
|
|
reference = strings.Replace(reference, "$", "", -1)
|
|
refs, cellRanges, cellRefs := list.New(), list.New(), list.New()
|
|
for _, ref := range strings.Split(reference, ":") {
|
|
tokens := strings.Split(ref, "!")
|
|
cr := cellRef{}
|
|
if len(tokens) == 2 { // have a worksheet name
|
|
cr.Sheet = tokens[0]
|
|
// cast to cell coordinates
|
|
if cr.Col, cr.Row, err = CellNameToCoordinates(tokens[1]); err != nil {
|
|
// cast to column
|
|
if cr.Col, err = ColumnNameToNumber(tokens[1]); err != nil {
|
|
// cast to row
|
|
if cr.Row, err = strconv.Atoi(tokens[1]); err != nil {
|
|
err = newInvalidColumnNameError(tokens[1])
|
|
return
|
|
}
|
|
cr.Col = TotalColumns
|
|
}
|
|
}
|
|
if refs.Len() > 0 {
|
|
e := refs.Back()
|
|
cellRefs.PushBack(e.Value.(cellRef))
|
|
refs.Remove(e)
|
|
}
|
|
refs.PushBack(cr)
|
|
continue
|
|
}
|
|
// cast to cell coordinates
|
|
if cr.Col, cr.Row, err = CellNameToCoordinates(tokens[0]); err != nil {
|
|
// cast to column
|
|
if cr.Col, err = ColumnNameToNumber(tokens[0]); err != nil {
|
|
// cast to row
|
|
if cr.Row, err = strconv.Atoi(tokens[0]); err != nil {
|
|
err = newInvalidColumnNameError(tokens[0])
|
|
return
|
|
}
|
|
cr.Col = TotalColumns
|
|
}
|
|
cellRanges.PushBack(cellRange{
|
|
From: cellRef{Sheet: sheet, Col: cr.Col, Row: 1},
|
|
To: cellRef{Sheet: sheet, Col: cr.Col, Row: TotalRows},
|
|
})
|
|
cellRefs.Init()
|
|
arg, err = f.rangeResolver(cellRefs, cellRanges)
|
|
return
|
|
}
|
|
e := refs.Back()
|
|
if e == nil {
|
|
cr.Sheet = sheet
|
|
refs.PushBack(cr)
|
|
continue
|
|
}
|
|
cellRanges.PushBack(cellRange{
|
|
From: e.Value.(cellRef),
|
|
To: cr,
|
|
})
|
|
refs.Remove(e)
|
|
}
|
|
if refs.Len() > 0 {
|
|
e := refs.Back()
|
|
cellRefs.PushBack(e.Value.(cellRef))
|
|
refs.Remove(e)
|
|
}
|
|
arg, err = f.rangeResolver(cellRefs, cellRanges)
|
|
return
|
|
}
|
|
|
|
// prepareValueRange prepare value range.
|
|
func prepareValueRange(cr cellRange, valueRange []int) {
|
|
if cr.From.Row < valueRange[0] || valueRange[0] == 0 {
|
|
valueRange[0] = cr.From.Row
|
|
}
|
|
if cr.From.Col < valueRange[2] || valueRange[2] == 0 {
|
|
valueRange[2] = cr.From.Col
|
|
}
|
|
if cr.To.Row > valueRange[1] || valueRange[1] == 0 {
|
|
valueRange[1] = cr.To.Row
|
|
}
|
|
if cr.To.Col > valueRange[3] || valueRange[3] == 0 {
|
|
valueRange[3] = cr.To.Col
|
|
}
|
|
}
|
|
|
|
// prepareValueRef prepare value reference.
|
|
func prepareValueRef(cr cellRef, valueRange []int) {
|
|
if cr.Row < valueRange[0] || valueRange[0] == 0 {
|
|
valueRange[0] = cr.Row
|
|
}
|
|
if cr.Col < valueRange[2] || valueRange[2] == 0 {
|
|
valueRange[2] = cr.Col
|
|
}
|
|
if cr.Row > valueRange[1] || valueRange[1] == 0 {
|
|
valueRange[1] = cr.Row
|
|
}
|
|
if cr.Col > valueRange[3] || valueRange[3] == 0 {
|
|
valueRange[3] = cr.Col
|
|
}
|
|
}
|
|
|
|
// rangeResolver extract value as string from given reference and range list.
|
|
// This function will not ignore the empty cell. For example, A1:A2:A2:B3 will
|
|
// be reference A1:B3.
|
|
func (f *File) rangeResolver(cellRefs, cellRanges *list.List) (arg formulaArg, err error) {
|
|
arg.cellRefs, arg.cellRanges = cellRefs, cellRanges
|
|
// value range order: from row, to row, from column, to column
|
|
valueRange := []int{0, 0, 0, 0}
|
|
var sheet string
|
|
// prepare value range
|
|
for temp := cellRanges.Front(); temp != nil; temp = temp.Next() {
|
|
cr := temp.Value.(cellRange)
|
|
if cr.From.Sheet != cr.To.Sheet {
|
|
err = errors.New(formulaErrorVALUE)
|
|
}
|
|
rng := []int{cr.From.Col, cr.From.Row, cr.To.Col, cr.To.Row}
|
|
_ = sortCoordinates(rng)
|
|
cr.From.Col, cr.From.Row, cr.To.Col, cr.To.Row = rng[0], rng[1], rng[2], rng[3]
|
|
prepareValueRange(cr, valueRange)
|
|
if cr.From.Sheet != "" {
|
|
sheet = cr.From.Sheet
|
|
}
|
|
}
|
|
for temp := cellRefs.Front(); temp != nil; temp = temp.Next() {
|
|
cr := temp.Value.(cellRef)
|
|
if cr.Sheet != "" {
|
|
sheet = cr.Sheet
|
|
}
|
|
prepareValueRef(cr, valueRange)
|
|
}
|
|
// extract value from ranges
|
|
if cellRanges.Len() > 0 {
|
|
arg.Type = ArgMatrix
|
|
for row := valueRange[0]; row <= valueRange[1]; row++ {
|
|
var matrixRow = []formulaArg{}
|
|
for col := valueRange[2]; col <= valueRange[3]; col++ {
|
|
var cell, value string
|
|
if cell, err = CoordinatesToCellName(col, row); err != nil {
|
|
return
|
|
}
|
|
if value, err = f.GetCellValue(sheet, cell); err != nil {
|
|
return
|
|
}
|
|
matrixRow = append(matrixRow, formulaArg{
|
|
String: value,
|
|
Type: ArgString,
|
|
})
|
|
}
|
|
arg.Matrix = append(arg.Matrix, matrixRow)
|
|
}
|
|
return
|
|
}
|
|
// extract value from references
|
|
for temp := cellRefs.Front(); temp != nil; temp = temp.Next() {
|
|
cr := temp.Value.(cellRef)
|
|
var cell string
|
|
if cell, err = CoordinatesToCellName(cr.Col, cr.Row); err != nil {
|
|
return
|
|
}
|
|
if arg.String, err = f.GetCellValue(cr.Sheet, cell); err != nil {
|
|
return
|
|
}
|
|
arg.Type = ArgString
|
|
}
|
|
return
|
|
}
|
|
|
|
// callFuncByName calls the no error or only error return function with
|
|
// reflect by given receiver, name and parameters.
|
|
func callFuncByName(receiver interface{}, name string, params []reflect.Value) (arg formulaArg) {
|
|
function := reflect.ValueOf(receiver).MethodByName(name)
|
|
if function.IsValid() {
|
|
rt := function.Call(params)
|
|
if len(rt) == 0 {
|
|
return
|
|
}
|
|
arg = rt[0].Interface().(formulaArg)
|
|
return
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("not support %s function", name))
|
|
}
|
|
|
|
// formulaCriteriaParser parse formula criteria.
|
|
func formulaCriteriaParser(exp string) (fc *formulaCriteria) {
|
|
fc = &formulaCriteria{}
|
|
if exp == "" {
|
|
return
|
|
}
|
|
if match := regexp.MustCompile(`^([0-9]+)$`).FindStringSubmatch(exp); len(match) > 1 {
|
|
fc.Type, fc.Condition = criteriaEq, match[1]
|
|
return
|
|
}
|
|
if match := regexp.MustCompile(`^=(.*)$`).FindStringSubmatch(exp); len(match) > 1 {
|
|
fc.Type, fc.Condition = criteriaEq, match[1]
|
|
return
|
|
}
|
|
if match := regexp.MustCompile(`^<=(.*)$`).FindStringSubmatch(exp); len(match) > 1 {
|
|
fc.Type, fc.Condition = criteriaLe, match[1]
|
|
return
|
|
}
|
|
if match := regexp.MustCompile(`^>=(.*)$`).FindStringSubmatch(exp); len(match) > 1 {
|
|
fc.Type, fc.Condition = criteriaGe, match[1]
|
|
return
|
|
}
|
|
if match := regexp.MustCompile(`^<(.*)$`).FindStringSubmatch(exp); len(match) > 1 {
|
|
fc.Type, fc.Condition = criteriaL, match[1]
|
|
return
|
|
}
|
|
if match := regexp.MustCompile(`^>(.*)$`).FindStringSubmatch(exp); len(match) > 1 {
|
|
fc.Type, fc.Condition = criteriaG, match[1]
|
|
return
|
|
}
|
|
if strings.Contains(exp, "*") {
|
|
if strings.HasPrefix(exp, "*") {
|
|
fc.Type, fc.Condition = criteriaEnd, strings.TrimPrefix(exp, "*")
|
|
}
|
|
if strings.HasSuffix(exp, "*") {
|
|
fc.Type, fc.Condition = criteriaBeg, strings.TrimSuffix(exp, "*")
|
|
}
|
|
return
|
|
}
|
|
fc.Type, fc.Condition = criteriaEq, exp
|
|
return
|
|
}
|
|
|
|
// formulaCriteriaEval evaluate formula criteria expression.
|
|
func formulaCriteriaEval(val string, criteria *formulaCriteria) (result bool, err error) {
|
|
var value, expected float64
|
|
var e error
|
|
var prepareValue = func(val, cond string) (value float64, expected float64, err error) {
|
|
if value, err = strconv.ParseFloat(val, 64); err != nil {
|
|
return
|
|
}
|
|
if expected, err = strconv.ParseFloat(criteria.Condition, 64); err != nil {
|
|
return
|
|
}
|
|
return
|
|
}
|
|
switch criteria.Type {
|
|
case criteriaEq:
|
|
return val == criteria.Condition, err
|
|
case criteriaLe:
|
|
value, expected, e = prepareValue(val, criteria.Condition)
|
|
return value <= expected && e == nil, err
|
|
case criteriaGe:
|
|
value, expected, e = prepareValue(val, criteria.Condition)
|
|
return value >= expected && e == nil, err
|
|
case criteriaL:
|
|
value, expected, e = prepareValue(val, criteria.Condition)
|
|
return value < expected && e == nil, err
|
|
case criteriaG:
|
|
value, expected, e = prepareValue(val, criteria.Condition)
|
|
return value > expected && e == nil, err
|
|
case criteriaBeg:
|
|
return strings.HasPrefix(val, criteria.Condition), err
|
|
case criteriaEnd:
|
|
return strings.HasSuffix(val, criteria.Condition), err
|
|
}
|
|
return
|
|
}
|
|
|
|
// Engineering Functions
|
|
|
|
// BIN2DEC function converts a Binary (a base-2 number) into a decimal number.
|
|
// The syntax of the function is:
|
|
//
|
|
// BIN2DEC(number)
|
|
//
|
|
func (fn *formulaFuncs) BIN2DEC(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "BIN2DEC requires 1 numeric argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
number := token.ToNumber()
|
|
if number.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, number.Error)
|
|
}
|
|
return fn.bin2dec(token.Value())
|
|
}
|
|
|
|
// BIN2HEX function converts a Binary (Base 2) number into a Hexadecimal
|
|
// (Base 16) number. The syntax of the function is:
|
|
//
|
|
// BIN2HEX(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) BIN2HEX(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "BIN2HEX requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "BIN2HEX allows at most 2 arguments")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
number := token.ToNumber()
|
|
if number.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, number.Error)
|
|
}
|
|
decimal, newList := fn.bin2dec(token.Value()), list.New()
|
|
if decimal.Type != ArgNumber {
|
|
return decimal
|
|
}
|
|
newList.PushBack(decimal)
|
|
if argsList.Len() == 2 {
|
|
newList.PushBack(argsList.Back().Value.(formulaArg))
|
|
}
|
|
return fn.dec2x("BIN2HEX", newList)
|
|
}
|
|
|
|
// BIN2OCT function converts a Binary (Base 2) number into an Octal (Base 8)
|
|
// number. The syntax of the function is:
|
|
//
|
|
// BIN2OCT(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) BIN2OCT(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "BIN2OCT requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "BIN2OCT allows at most 2 arguments")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
number := token.ToNumber()
|
|
if number.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, number.Error)
|
|
}
|
|
decimal, newList := fn.bin2dec(token.Value()), list.New()
|
|
if decimal.Type != ArgNumber {
|
|
return decimal
|
|
}
|
|
newList.PushBack(decimal)
|
|
if argsList.Len() == 2 {
|
|
newList.PushBack(argsList.Back().Value.(formulaArg))
|
|
}
|
|
return fn.dec2x("BIN2OCT", newList)
|
|
}
|
|
|
|
// bin2dec is an implementation of the formula function BIN2DEC.
|
|
func (fn *formulaFuncs) bin2dec(number string) formulaArg {
|
|
decimal, length := 0.0, len(number)
|
|
for i := length; i > 0; i-- {
|
|
s := string(number[length-i])
|
|
if 10 == i && s == "1" {
|
|
decimal += math.Pow(-2.0, float64(i-1))
|
|
continue
|
|
}
|
|
if s == "1" {
|
|
decimal += math.Pow(2.0, float64(i-1))
|
|
continue
|
|
}
|
|
if s != "0" {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
}
|
|
return newNumberFormulaArg(decimal)
|
|
}
|
|
|
|
// BITAND function returns the bitwise 'AND' for two supplied integers. The
|
|
// syntax of the function is:
|
|
//
|
|
// BITAND(number1,number2)
|
|
//
|
|
func (fn *formulaFuncs) BITAND(argsList *list.List) formulaArg {
|
|
return fn.bitwise("BITAND", argsList)
|
|
}
|
|
|
|
// BITLSHIFT function returns a supplied integer, shifted left by a specified
|
|
// number of bits. The syntax of the function is:
|
|
//
|
|
// BITLSHIFT(number1,shift_amount)
|
|
//
|
|
func (fn *formulaFuncs) BITLSHIFT(argsList *list.List) formulaArg {
|
|
return fn.bitwise("BITLSHIFT", argsList)
|
|
}
|
|
|
|
// BITOR function returns the bitwise 'OR' for two supplied integers. The
|
|
// syntax of the function is:
|
|
//
|
|
// BITOR(number1,number2)
|
|
//
|
|
func (fn *formulaFuncs) BITOR(argsList *list.List) formulaArg {
|
|
return fn.bitwise("BITOR", argsList)
|
|
}
|
|
|
|
// BITRSHIFT function returns a supplied integer, shifted right by a specified
|
|
// number of bits. The syntax of the function is:
|
|
//
|
|
// BITRSHIFT(number1,shift_amount)
|
|
//
|
|
func (fn *formulaFuncs) BITRSHIFT(argsList *list.List) formulaArg {
|
|
return fn.bitwise("BITRSHIFT", argsList)
|
|
}
|
|
|
|
// BITXOR function returns the bitwise 'XOR' (exclusive 'OR') for two supplied
|
|
// integers. The syntax of the function is:
|
|
//
|
|
// BITXOR(number1,number2)
|
|
//
|
|
func (fn *formulaFuncs) BITXOR(argsList *list.List) formulaArg {
|
|
return fn.bitwise("BITXOR", argsList)
|
|
}
|
|
|
|
// bitwise is an implementation of the formula function BITAND, BITLSHIFT,
|
|
// BITOR, BITRSHIFT and BITXOR.
|
|
func (fn *formulaFuncs) bitwise(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires 2 numeric arguments", name))
|
|
}
|
|
num1, num2 := argsList.Front().Value.(formulaArg).ToNumber(), argsList.Back().Value.(formulaArg).ToNumber()
|
|
if num1.Type != ArgNumber || num2.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
max := math.Pow(2, 48) - 1
|
|
if num1.Number < 0 || num1.Number > max || num2.Number < 0 || num2.Number > max {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
bitwiseFuncMap := map[string]func(a, b int) int{
|
|
"BITAND": func(a, b int) int { return a & b },
|
|
"BITLSHIFT": func(a, b int) int { return a << uint(b) },
|
|
"BITOR": func(a, b int) int { return a | b },
|
|
"BITRSHIFT": func(a, b int) int { return a >> uint(b) },
|
|
"BITXOR": func(a, b int) int { return a ^ b },
|
|
}
|
|
bitwiseFunc := bitwiseFuncMap[name]
|
|
return newNumberFormulaArg(float64(bitwiseFunc(int(num1.Number), int(num2.Number))))
|
|
}
|
|
|
|
// DEC2BIN function converts a decimal number into a Binary (Base 2) number.
|
|
// The syntax of the function is:
|
|
//
|
|
// DEC2BIN(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) DEC2BIN(argsList *list.List) formulaArg {
|
|
return fn.dec2x("DEC2BIN", argsList)
|
|
}
|
|
|
|
// DEC2HEX function converts a decimal number into a Hexadecimal (Base 16)
|
|
// number. The syntax of the function is:
|
|
//
|
|
// DEC2HEX(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) DEC2HEX(argsList *list.List) formulaArg {
|
|
return fn.dec2x("DEC2HEX", argsList)
|
|
}
|
|
|
|
// DEC2OCT function converts a decimal number into an Octal (Base 8) number.
|
|
// The syntax of the function is:
|
|
//
|
|
// DEC2OCT(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) DEC2OCT(argsList *list.List) formulaArg {
|
|
return fn.dec2x("DEC2OCT", argsList)
|
|
}
|
|
|
|
// dec2x is an implementation of the formula function DEC2BIN, DEC2HEX and
|
|
// DEC2OCT.
|
|
func (fn *formulaFuncs) dec2x(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires at least 1 argument", name))
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s allows at most 2 arguments", name))
|
|
}
|
|
decimal := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if decimal.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, decimal.Error)
|
|
}
|
|
maxLimitMap := map[string]float64{
|
|
"DEC2BIN": 511,
|
|
"HEX2BIN": 511,
|
|
"OCT2BIN": 511,
|
|
"BIN2HEX": 549755813887,
|
|
"DEC2HEX": 549755813887,
|
|
"OCT2HEX": 549755813887,
|
|
"BIN2OCT": 536870911,
|
|
"DEC2OCT": 536870911,
|
|
"HEX2OCT": 536870911,
|
|
}
|
|
minLimitMap := map[string]float64{
|
|
"DEC2BIN": -512,
|
|
"HEX2BIN": -512,
|
|
"OCT2BIN": -512,
|
|
"BIN2HEX": -549755813888,
|
|
"DEC2HEX": -549755813888,
|
|
"OCT2HEX": -549755813888,
|
|
"BIN2OCT": -536870912,
|
|
"DEC2OCT": -536870912,
|
|
"HEX2OCT": -536870912,
|
|
}
|
|
baseMap := map[string]int{
|
|
"DEC2BIN": 2,
|
|
"HEX2BIN": 2,
|
|
"OCT2BIN": 2,
|
|
"BIN2HEX": 16,
|
|
"DEC2HEX": 16,
|
|
"OCT2HEX": 16,
|
|
"BIN2OCT": 8,
|
|
"DEC2OCT": 8,
|
|
"HEX2OCT": 8,
|
|
}
|
|
maxLimit, minLimit := maxLimitMap[name], minLimitMap[name]
|
|
base := baseMap[name]
|
|
if decimal.Number < minLimit || decimal.Number > maxLimit {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
n := int64(decimal.Number)
|
|
binary := strconv.FormatUint(*(*uint64)(unsafe.Pointer(&n)), base)
|
|
if argsList.Len() == 2 {
|
|
places := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if places.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, places.Error)
|
|
}
|
|
binaryPlaces := len(binary)
|
|
if places.Number < 0 || places.Number > 10 || binaryPlaces > int(places.Number) {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(fmt.Sprintf("%s%s", strings.Repeat("0", int(places.Number)-binaryPlaces), binary)))
|
|
}
|
|
if decimal.Number < 0 && len(binary) > 10 {
|
|
return newStringFormulaArg(strings.ToUpper(binary[len(binary)-10:]))
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(binary))
|
|
}
|
|
|
|
// HEX2BIN function converts a Hexadecimal (Base 16) number into a Binary
|
|
// (Base 2) number. The syntax of the function is:
|
|
//
|
|
// HEX2BIN(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) HEX2BIN(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HEX2BIN requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HEX2BIN allows at most 2 arguments")
|
|
}
|
|
decimal, newList := fn.hex2dec(argsList.Front().Value.(formulaArg).Value()), list.New()
|
|
if decimal.Type != ArgNumber {
|
|
return decimal
|
|
}
|
|
newList.PushBack(decimal)
|
|
if argsList.Len() == 2 {
|
|
newList.PushBack(argsList.Back().Value.(formulaArg))
|
|
}
|
|
return fn.dec2x("HEX2BIN", newList)
|
|
}
|
|
|
|
// HEX2DEC function converts a hexadecimal (a base-16 number) into a decimal
|
|
// number. The syntax of the function is:
|
|
//
|
|
// HEX2DEC(number)
|
|
//
|
|
func (fn *formulaFuncs) HEX2DEC(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HEX2DEC requires 1 numeric argument")
|
|
}
|
|
return fn.hex2dec(argsList.Front().Value.(formulaArg).Value())
|
|
}
|
|
|
|
// HEX2OCT function converts a Hexadecimal (Base 16) number into an Octal
|
|
// (Base 8) number. The syntax of the function is:
|
|
//
|
|
// HEX2OCT(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) HEX2OCT(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HEX2OCT requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HEX2OCT allows at most 2 arguments")
|
|
}
|
|
decimal, newList := fn.hex2dec(argsList.Front().Value.(formulaArg).Value()), list.New()
|
|
if decimal.Type != ArgNumber {
|
|
return decimal
|
|
}
|
|
newList.PushBack(decimal)
|
|
if argsList.Len() == 2 {
|
|
newList.PushBack(argsList.Back().Value.(formulaArg))
|
|
}
|
|
return fn.dec2x("HEX2OCT", newList)
|
|
}
|
|
|
|
// hex2dec is an implementation of the formula function HEX2DEC.
|
|
func (fn *formulaFuncs) hex2dec(number string) formulaArg {
|
|
decimal, length := 0.0, len(number)
|
|
for i := length; i > 0; i-- {
|
|
num, err := strconv.ParseInt(string(number[length-i]), 16, 64)
|
|
if err != nil {
|
|
return newErrorFormulaArg(formulaErrorNUM, err.Error())
|
|
}
|
|
if 10 == i && string(number[length-i]) == "F" {
|
|
decimal += math.Pow(-16.0, float64(i-1))
|
|
continue
|
|
}
|
|
decimal += float64(num) * math.Pow(16.0, float64(i-1))
|
|
}
|
|
return newNumberFormulaArg(decimal)
|
|
}
|
|
|
|
// OCT2BIN function converts an Octal (Base 8) number into a Binary (Base 2)
|
|
// number. The syntax of the function is:
|
|
//
|
|
// OCT2BIN(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) OCT2BIN(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "OCT2BIN requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "OCT2BIN allows at most 2 arguments")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
number := token.ToNumber()
|
|
if number.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, number.Error)
|
|
}
|
|
decimal, newList := fn.oct2dec(token.Value()), list.New()
|
|
newList.PushBack(decimal)
|
|
if argsList.Len() == 2 {
|
|
newList.PushBack(argsList.Back().Value.(formulaArg))
|
|
}
|
|
return fn.dec2x("OCT2BIN", newList)
|
|
}
|
|
|
|
// OCT2DEC function converts an Octal (a base-8 number) into a decimal number.
|
|
// The syntax of the function is:
|
|
//
|
|
// OCT2DEC(number)
|
|
//
|
|
func (fn *formulaFuncs) OCT2DEC(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "OCT2DEC requires 1 numeric argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
number := token.ToNumber()
|
|
if number.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, number.Error)
|
|
}
|
|
return fn.oct2dec(token.Value())
|
|
}
|
|
|
|
// OCT2HEX function converts an Octal (Base 8) number into a Hexadecimal
|
|
// (Base 16) number. The syntax of the function is:
|
|
//
|
|
// OCT2HEX(number,[places])
|
|
//
|
|
func (fn *formulaFuncs) OCT2HEX(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "OCT2HEX requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "OCT2HEX allows at most 2 arguments")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
number := token.ToNumber()
|
|
if number.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, number.Error)
|
|
}
|
|
decimal, newList := fn.oct2dec(token.Value()), list.New()
|
|
newList.PushBack(decimal)
|
|
if argsList.Len() == 2 {
|
|
newList.PushBack(argsList.Back().Value.(formulaArg))
|
|
}
|
|
return fn.dec2x("OCT2HEX", newList)
|
|
}
|
|
|
|
// oct2dec is an implementation of the formula function OCT2DEC.
|
|
func (fn *formulaFuncs) oct2dec(number string) formulaArg {
|
|
decimal, length := 0.0, len(number)
|
|
for i := length; i > 0; i-- {
|
|
num, _ := strconv.Atoi(string(number[length-i]))
|
|
if 10 == i && string(number[length-i]) == "7" {
|
|
decimal += math.Pow(-8.0, float64(i-1))
|
|
continue
|
|
}
|
|
decimal += float64(num) * math.Pow(8.0, float64(i-1))
|
|
}
|
|
return newNumberFormulaArg(decimal)
|
|
}
|
|
|
|
// Math and Trigonometric Functions
|
|
|
|
// ABS function returns the absolute value of any supplied number. The syntax
|
|
// of the function is:
|
|
//
|
|
// ABS(number)
|
|
//
|
|
func (fn *formulaFuncs) ABS(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ABS requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Abs(arg.Number))
|
|
}
|
|
|
|
// ACOS function calculates the arccosine (i.e. the inverse cosine) of a given
|
|
// number, and returns an angle, in radians, between 0 and π. The syntax of
|
|
// the function is:
|
|
//
|
|
// ACOS(number)
|
|
//
|
|
func (fn *formulaFuncs) ACOS(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ACOS requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Acos(arg.Number))
|
|
}
|
|
|
|
// ACOSH function calculates the inverse hyperbolic cosine of a supplied number.
|
|
// of the function is:
|
|
//
|
|
// ACOSH(number)
|
|
//
|
|
func (fn *formulaFuncs) ACOSH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ACOSH requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Acosh(arg.Number))
|
|
}
|
|
|
|
// ACOT function calculates the arccotangent (i.e. the inverse cotangent) of a
|
|
// given number, and returns an angle, in radians, between 0 and π. The syntax
|
|
// of the function is:
|
|
//
|
|
// ACOT(number)
|
|
//
|
|
func (fn *formulaFuncs) ACOT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ACOT requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Pi/2 - math.Atan(arg.Number))
|
|
}
|
|
|
|
// ACOTH function calculates the hyperbolic arccotangent (coth) of a supplied
|
|
// value. The syntax of the function is:
|
|
//
|
|
// ACOTH(number)
|
|
//
|
|
func (fn *formulaFuncs) ACOTH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ACOTH requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Atanh(1 / arg.Number))
|
|
}
|
|
|
|
// ARABIC function converts a Roman numeral into an Arabic numeral. The syntax
|
|
// of the function is:
|
|
//
|
|
// ARABIC(text)
|
|
//
|
|
func (fn *formulaFuncs) ARABIC(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ARABIC requires 1 numeric argument")
|
|
}
|
|
text := argsList.Front().Value.(formulaArg).Value()
|
|
if len(text) > 255 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
text = strings.ToUpper(text)
|
|
number, actualStart, index, isNegative := 0, 0, len(text)-1, false
|
|
startIndex, subtractNumber, currentPartValue, currentCharValue, prevCharValue := 0, 0, 0, 0, -1
|
|
for index >= 0 && text[index] == ' ' {
|
|
index--
|
|
}
|
|
for actualStart <= index && text[actualStart] == ' ' {
|
|
actualStart++
|
|
}
|
|
if actualStart <= index && text[actualStart] == '-' {
|
|
isNegative = true
|
|
actualStart++
|
|
}
|
|
charMap := map[rune]int{'I': 1, 'V': 5, 'X': 10, 'L': 50, 'C': 100, 'D': 500, 'M': 1000}
|
|
for index >= actualStart {
|
|
startIndex = index
|
|
startChar := text[startIndex]
|
|
index--
|
|
for index >= actualStart && (text[index]|' ') == startChar {
|
|
index--
|
|
}
|
|
currentCharValue = charMap[rune(startChar)]
|
|
currentPartValue = (startIndex - index) * currentCharValue
|
|
if currentCharValue >= prevCharValue {
|
|
number += currentPartValue - subtractNumber
|
|
prevCharValue = currentCharValue
|
|
subtractNumber = 0
|
|
continue
|
|
}
|
|
subtractNumber += currentPartValue
|
|
}
|
|
if subtractNumber != 0 {
|
|
number -= subtractNumber
|
|
}
|
|
if isNegative {
|
|
number = -number
|
|
}
|
|
return newNumberFormulaArg(float64(number))
|
|
}
|
|
|
|
// ASIN function calculates the arcsine (i.e. the inverse sine) of a given
|
|
// number, and returns an angle, in radians, between -π/2 and π/2. The syntax
|
|
// of the function is:
|
|
//
|
|
// ASIN(number)
|
|
//
|
|
func (fn *formulaFuncs) ASIN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ASIN requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Asin(arg.Number))
|
|
}
|
|
|
|
// ASINH function calculates the inverse hyperbolic sine of a supplied number.
|
|
// The syntax of the function is:
|
|
//
|
|
// ASINH(number)
|
|
//
|
|
func (fn *formulaFuncs) ASINH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ASINH requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Asinh(arg.Number))
|
|
}
|
|
|
|
// ATAN function calculates the arctangent (i.e. the inverse tangent) of a
|
|
// given number, and returns an angle, in radians, between -π/2 and +π/2. The
|
|
// syntax of the function is:
|
|
//
|
|
// ATAN(number)
|
|
//
|
|
func (fn *formulaFuncs) ATAN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ATAN requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Atan(arg.Number))
|
|
}
|
|
|
|
// ATANH function calculates the inverse hyperbolic tangent of a supplied
|
|
// number. The syntax of the function is:
|
|
//
|
|
// ATANH(number)
|
|
//
|
|
func (fn *formulaFuncs) ATANH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ATANH requires 1 numeric argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type == ArgError {
|
|
return arg
|
|
}
|
|
return newNumberFormulaArg(math.Atanh(arg.Number))
|
|
}
|
|
|
|
// ATAN2 function calculates the arctangent (i.e. the inverse tangent) of a
|
|
// given set of x and y coordinates, and returns an angle, in radians, between
|
|
// -π/2 and +π/2. The syntax of the function is:
|
|
//
|
|
// ATAN2(x_num,y_num)
|
|
//
|
|
func (fn *formulaFuncs) ATAN2(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ATAN2 requires 2 numeric arguments")
|
|
}
|
|
x := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if x.Type == ArgError {
|
|
return x
|
|
}
|
|
y := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if y.Type == ArgError {
|
|
return y
|
|
}
|
|
return newNumberFormulaArg(math.Atan2(x.Number, y.Number))
|
|
}
|
|
|
|
// BASE function converts a number into a supplied base (radix), and returns a
|
|
// text representation of the calculated value. The syntax of the function is:
|
|
//
|
|
// BASE(number,radix,[min_length])
|
|
//
|
|
func (fn *formulaFuncs) BASE(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "BASE requires at least 2 arguments")
|
|
}
|
|
if argsList.Len() > 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "BASE allows at most 3 arguments")
|
|
}
|
|
var minLength int
|
|
var err error
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
radix := argsList.Front().Next().Value.(formulaArg).ToNumber()
|
|
if radix.Type == ArgError {
|
|
return radix
|
|
}
|
|
if int(radix.Number) < 2 || int(radix.Number) > 36 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "radix must be an integer >= 2 and <= 36")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
if minLength, err = strconv.Atoi(argsList.Back().Value.(formulaArg).String); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
}
|
|
result := strconv.FormatInt(int64(number.Number), int(radix.Number))
|
|
if len(result) < minLength {
|
|
result = strings.Repeat("0", minLength-len(result)) + result
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(result))
|
|
}
|
|
|
|
// CEILING function rounds a supplied number away from zero, to the nearest
|
|
// multiple of a given number. The syntax of the function is:
|
|
//
|
|
// CEILING(number,significance)
|
|
//
|
|
func (fn *formulaFuncs) CEILING(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CEILING requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CEILING allows at most 2 arguments")
|
|
}
|
|
number, significance, res := 0.0, 1.0, 0.0
|
|
n := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if n.Type == ArgError {
|
|
return n
|
|
}
|
|
number = n.Number
|
|
if number < 0 {
|
|
significance = -1
|
|
}
|
|
if argsList.Len() > 1 {
|
|
s := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if s.Type == ArgError {
|
|
return s
|
|
}
|
|
significance = s.Number
|
|
}
|
|
if significance < 0 && number > 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "negative sig to CEILING invalid")
|
|
}
|
|
if argsList.Len() == 1 {
|
|
return newNumberFormulaArg(math.Ceil(number))
|
|
}
|
|
number, res = math.Modf(number / significance)
|
|
if res > 0 {
|
|
number++
|
|
}
|
|
return newNumberFormulaArg(number * significance)
|
|
}
|
|
|
|
// CEILINGMATH function rounds a supplied number up to a supplied multiple of
|
|
// significance. The syntax of the function is:
|
|
//
|
|
// CEILING.MATH(number,[significance],[mode])
|
|
//
|
|
func (fn *formulaFuncs) CEILINGMATH(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CEILING.MATH requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CEILING.MATH allows at most 3 arguments")
|
|
}
|
|
number, significance, mode := 0.0, 1.0, 1.0
|
|
n := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if n.Type == ArgError {
|
|
return n
|
|
}
|
|
number = n.Number
|
|
if number < 0 {
|
|
significance = -1
|
|
}
|
|
if argsList.Len() > 1 {
|
|
s := argsList.Front().Next().Value.(formulaArg).ToNumber()
|
|
if s.Type == ArgError {
|
|
return s
|
|
}
|
|
significance = s.Number
|
|
}
|
|
if argsList.Len() == 1 {
|
|
return newNumberFormulaArg(math.Ceil(number))
|
|
}
|
|
if argsList.Len() > 2 {
|
|
m := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if m.Type == ArgError {
|
|
return m
|
|
}
|
|
mode = m.Number
|
|
}
|
|
val, res := math.Modf(number / significance)
|
|
if res != 0 {
|
|
if number > 0 {
|
|
val++
|
|
} else if mode < 0 {
|
|
val--
|
|
}
|
|
}
|
|
return newNumberFormulaArg(val * significance)
|
|
}
|
|
|
|
// CEILINGPRECISE function rounds a supplied number up (regardless of the
|
|
// number's sign), to the nearest multiple of a given number. The syntax of
|
|
// the function is:
|
|
//
|
|
// CEILING.PRECISE(number,[significance])
|
|
//
|
|
func (fn *formulaFuncs) CEILINGPRECISE(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CEILING.PRECISE requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CEILING.PRECISE allows at most 2 arguments")
|
|
}
|
|
number, significance := 0.0, 1.0
|
|
n := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if n.Type == ArgError {
|
|
return n
|
|
}
|
|
number = n.Number
|
|
if number < 0 {
|
|
significance = -1
|
|
}
|
|
if argsList.Len() == 1 {
|
|
return newNumberFormulaArg(math.Ceil(number))
|
|
}
|
|
if argsList.Len() > 1 {
|
|
s := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if s.Type == ArgError {
|
|
return s
|
|
}
|
|
significance = s.Number
|
|
significance = math.Abs(significance)
|
|
if significance == 0 {
|
|
return newNumberFormulaArg(significance)
|
|
}
|
|
}
|
|
val, res := math.Modf(number / significance)
|
|
if res != 0 {
|
|
if number > 0 {
|
|
val++
|
|
}
|
|
}
|
|
return newNumberFormulaArg(val * significance)
|
|
}
|
|
|
|
// COMBIN function calculates the number of combinations (in any order) of a
|
|
// given number objects from a set. The syntax of the function is:
|
|
//
|
|
// COMBIN(number,number_chosen)
|
|
//
|
|
func (fn *formulaFuncs) COMBIN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COMBIN requires 2 argument")
|
|
}
|
|
number, chosen, val := 0.0, 0.0, 1.0
|
|
n := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if n.Type == ArgError {
|
|
return n
|
|
}
|
|
number = n.Number
|
|
c := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if c.Type == ArgError {
|
|
return c
|
|
}
|
|
chosen = c.Number
|
|
number, chosen = math.Trunc(number), math.Trunc(chosen)
|
|
if chosen > number {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COMBIN requires number >= number_chosen")
|
|
}
|
|
if chosen == number || chosen == 0 {
|
|
return newNumberFormulaArg(1)
|
|
}
|
|
for c := float64(1); c <= chosen; c++ {
|
|
val *= (number + 1 - c) / c
|
|
}
|
|
return newNumberFormulaArg(math.Ceil(val))
|
|
}
|
|
|
|
// COMBINA function calculates the number of combinations, with repetitions,
|
|
// of a given number objects from a set. The syntax of the function is:
|
|
//
|
|
// COMBINA(number,number_chosen)
|
|
//
|
|
func (fn *formulaFuncs) COMBINA(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COMBINA requires 2 argument")
|
|
}
|
|
var number, chosen float64
|
|
n := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if n.Type == ArgError {
|
|
return n
|
|
}
|
|
number = n.Number
|
|
c := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if c.Type == ArgError {
|
|
return c
|
|
}
|
|
chosen = c.Number
|
|
number, chosen = math.Trunc(number), math.Trunc(chosen)
|
|
if number < chosen {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COMBINA requires number > number_chosen")
|
|
}
|
|
if number == 0 {
|
|
return newNumberFormulaArg(number)
|
|
}
|
|
args := list.New()
|
|
args.PushBack(formulaArg{
|
|
String: fmt.Sprintf("%g", number+chosen-1),
|
|
Type: ArgString,
|
|
})
|
|
args.PushBack(formulaArg{
|
|
String: fmt.Sprintf("%g", number-1),
|
|
Type: ArgString,
|
|
})
|
|
return fn.COMBIN(args)
|
|
}
|
|
|
|
// COS function calculates the cosine of a given angle. The syntax of the
|
|
// function is:
|
|
//
|
|
// COS(number)
|
|
//
|
|
func (fn *formulaFuncs) COS(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COS requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
return newNumberFormulaArg(math.Cos(val.Number))
|
|
}
|
|
|
|
// COSH function calculates the hyperbolic cosine (cosh) of a supplied number.
|
|
// The syntax of the function is:
|
|
//
|
|
// COSH(number)
|
|
//
|
|
func (fn *formulaFuncs) COSH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COSH requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
return newNumberFormulaArg(math.Cosh(val.Number))
|
|
}
|
|
|
|
// COT function calculates the cotangent of a given angle. The syntax of the
|
|
// function is:
|
|
//
|
|
// COT(number)
|
|
//
|
|
func (fn *formulaFuncs) COT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COT requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
if val.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg(1 / math.Tan(val.Number))
|
|
}
|
|
|
|
// COTH function calculates the hyperbolic cotangent (coth) of a supplied
|
|
// angle. The syntax of the function is:
|
|
//
|
|
// COTH(number)
|
|
//
|
|
func (fn *formulaFuncs) COTH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COTH requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
if val.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg((math.Exp(val.Number) + math.Exp(-val.Number)) / (math.Exp(val.Number) - math.Exp(-val.Number)))
|
|
}
|
|
|
|
// CSC function calculates the cosecant of a given angle. The syntax of the
|
|
// function is:
|
|
//
|
|
// CSC(number)
|
|
//
|
|
func (fn *formulaFuncs) CSC(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CSC requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
if val.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg(1 / math.Sin(val.Number))
|
|
}
|
|
|
|
// CSCH function calculates the hyperbolic cosecant (csch) of a supplied
|
|
// angle. The syntax of the function is:
|
|
//
|
|
// CSCH(number)
|
|
//
|
|
func (fn *formulaFuncs) CSCH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CSCH requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
if val.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg(1 / math.Sinh(val.Number))
|
|
}
|
|
|
|
// DECIMAL function converts a text representation of a number in a specified
|
|
// base, into a decimal value. The syntax of the function is:
|
|
//
|
|
// DECIMAL(text,radix)
|
|
//
|
|
func (fn *formulaFuncs) DECIMAL(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "DECIMAL requires 2 numeric arguments")
|
|
}
|
|
var text = argsList.Front().Value.(formulaArg).String
|
|
var radix int
|
|
var err error
|
|
radix, err = strconv.Atoi(argsList.Back().Value.(formulaArg).String)
|
|
if err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
if len(text) > 2 && (strings.HasPrefix(text, "0x") || strings.HasPrefix(text, "0X")) {
|
|
text = text[2:]
|
|
}
|
|
val, err := strconv.ParseInt(text, radix, 64)
|
|
if err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
return newNumberFormulaArg(float64(val))
|
|
}
|
|
|
|
// DEGREES function converts radians into degrees. The syntax of the function
|
|
// is:
|
|
//
|
|
// DEGREES(angle)
|
|
//
|
|
func (fn *formulaFuncs) DEGREES(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "DEGREES requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
if val.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg(180.0 / math.Pi * val.Number)
|
|
}
|
|
|
|
// EVEN function rounds a supplied number away from zero (i.e. rounds a
|
|
// positive number up and a negative number down), to the next even number.
|
|
// The syntax of the function is:
|
|
//
|
|
// EVEN(number)
|
|
//
|
|
func (fn *formulaFuncs) EVEN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "EVEN requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
sign := math.Signbit(number.Number)
|
|
m, frac := math.Modf(number.Number / 2)
|
|
val := m * 2
|
|
if frac != 0 {
|
|
if !sign {
|
|
val += 2
|
|
} else {
|
|
val -= 2
|
|
}
|
|
}
|
|
return newNumberFormulaArg(val)
|
|
}
|
|
|
|
// EXP function calculates the value of the mathematical constant e, raised to
|
|
// the power of a given number. The syntax of the function is:
|
|
//
|
|
// EXP(number)
|
|
//
|
|
func (fn *formulaFuncs) EXP(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "EXP requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(fmt.Sprintf("%g", math.Exp(number.Number))))
|
|
}
|
|
|
|
// fact returns the factorial of a supplied number.
|
|
func fact(number float64) float64 {
|
|
val := float64(1)
|
|
for i := float64(2); i <= number; i++ {
|
|
val *= i
|
|
}
|
|
return val
|
|
}
|
|
|
|
// FACT function returns the factorial of a supplied number. The syntax of the
|
|
// function is:
|
|
//
|
|
// FACT(number)
|
|
//
|
|
func (fn *formulaFuncs) FACT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FACT requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if number.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
return newNumberFormulaArg(fact(number.Number))
|
|
}
|
|
|
|
// FACTDOUBLE function returns the double factorial of a supplied number. The
|
|
// syntax of the function is:
|
|
//
|
|
// FACTDOUBLE(number)
|
|
//
|
|
func (fn *formulaFuncs) FACTDOUBLE(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FACTDOUBLE requires 1 numeric argument")
|
|
}
|
|
val := 1.0
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if number.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
for i := math.Trunc(number.Number); i > 1; i -= 2 {
|
|
val *= i
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(fmt.Sprintf("%g", val)))
|
|
}
|
|
|
|
// FLOOR function rounds a supplied number towards zero to the nearest
|
|
// multiple of a specified significance. The syntax of the function is:
|
|
//
|
|
// FLOOR(number,significance)
|
|
//
|
|
func (fn *formulaFuncs) FLOOR(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FLOOR requires 2 numeric arguments")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
significance := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if significance.Type == ArgError {
|
|
return significance
|
|
}
|
|
if significance.Number < 0 && number.Number >= 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, "invalid arguments to FLOOR")
|
|
}
|
|
val := number.Number
|
|
val, res := math.Modf(val / significance.Number)
|
|
if res != 0 {
|
|
if number.Number < 0 && res < 0 {
|
|
val--
|
|
}
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(fmt.Sprintf("%g", val*significance.Number)))
|
|
}
|
|
|
|
// FLOORMATH function rounds a supplied number down to a supplied multiple of
|
|
// significance. The syntax of the function is:
|
|
//
|
|
// FLOOR.MATH(number,[significance],[mode])
|
|
//
|
|
func (fn *formulaFuncs) FLOORMATH(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FLOOR.MATH requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FLOOR.MATH allows at most 3 arguments")
|
|
}
|
|
significance, mode := 1.0, 1.0
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if number.Number < 0 {
|
|
significance = -1
|
|
}
|
|
if argsList.Len() > 1 {
|
|
s := argsList.Front().Next().Value.(formulaArg).ToNumber()
|
|
if s.Type == ArgError {
|
|
return s
|
|
}
|
|
significance = s.Number
|
|
}
|
|
if argsList.Len() == 1 {
|
|
return newNumberFormulaArg(math.Floor(number.Number))
|
|
}
|
|
if argsList.Len() > 2 {
|
|
m := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if m.Type == ArgError {
|
|
return m
|
|
}
|
|
mode = m.Number
|
|
}
|
|
val, res := math.Modf(number.Number / significance)
|
|
if res != 0 && number.Number < 0 && mode > 0 {
|
|
val--
|
|
}
|
|
return newNumberFormulaArg(val * significance)
|
|
}
|
|
|
|
// FLOORPRECISE function rounds a supplied number down to a supplied multiple
|
|
// of significance. The syntax of the function is:
|
|
//
|
|
// FLOOR.PRECISE(number,[significance])
|
|
//
|
|
func (fn *formulaFuncs) FLOORPRECISE(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FLOOR.PRECISE requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FLOOR.PRECISE allows at most 2 arguments")
|
|
}
|
|
var significance float64
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if number.Number < 0 {
|
|
significance = -1
|
|
}
|
|
if argsList.Len() == 1 {
|
|
return newNumberFormulaArg(math.Floor(number.Number))
|
|
}
|
|
if argsList.Len() > 1 {
|
|
s := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if s.Type == ArgError {
|
|
return s
|
|
}
|
|
significance = s.Number
|
|
significance = math.Abs(significance)
|
|
if significance == 0 {
|
|
return newNumberFormulaArg(significance)
|
|
}
|
|
}
|
|
val, res := math.Modf(number.Number / significance)
|
|
if res != 0 {
|
|
if number.Number < 0 {
|
|
val--
|
|
}
|
|
}
|
|
return newNumberFormulaArg(val * significance)
|
|
}
|
|
|
|
// gcd returns the greatest common divisor of two supplied integers.
|
|
func gcd(x, y float64) float64 {
|
|
x, y = math.Trunc(x), math.Trunc(y)
|
|
if x == 0 {
|
|
return y
|
|
}
|
|
if y == 0 {
|
|
return x
|
|
}
|
|
for x != y {
|
|
if x > y {
|
|
x = x - y
|
|
} else {
|
|
y = y - x
|
|
}
|
|
}
|
|
return x
|
|
}
|
|
|
|
// GCD function returns the greatest common divisor of two or more supplied
|
|
// integers. The syntax of the function is:
|
|
//
|
|
// GCD(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) GCD(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "GCD requires at least 1 argument")
|
|
}
|
|
var (
|
|
val float64
|
|
nums = []float64{}
|
|
)
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
num := token.ToNumber()
|
|
if num.Type == ArgError {
|
|
return num
|
|
}
|
|
val = num.Number
|
|
case ArgNumber:
|
|
val = token.Number
|
|
}
|
|
nums = append(nums, val)
|
|
}
|
|
if nums[0] < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "GCD only accepts positive arguments")
|
|
}
|
|
if len(nums) == 1 {
|
|
return newNumberFormulaArg(nums[0])
|
|
}
|
|
cd := nums[0]
|
|
for i := 1; i < len(nums); i++ {
|
|
if nums[i] < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "GCD only accepts positive arguments")
|
|
}
|
|
cd = gcd(cd, nums[i])
|
|
}
|
|
return newNumberFormulaArg(cd)
|
|
}
|
|
|
|
// INT function truncates a supplied number down to the closest integer. The
|
|
// syntax of the function is:
|
|
//
|
|
// INT(number)
|
|
//
|
|
func (fn *formulaFuncs) INT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "INT requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
val, frac := math.Modf(number.Number)
|
|
if frac < 0 {
|
|
val--
|
|
}
|
|
return newNumberFormulaArg(val)
|
|
}
|
|
|
|
// ISOCEILING function rounds a supplied number up (regardless of the number's
|
|
// sign), to the nearest multiple of a supplied significance. The syntax of
|
|
// the function is:
|
|
//
|
|
// ISO.CEILING(number,[significance])
|
|
//
|
|
func (fn *formulaFuncs) ISOCEILING(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISO.CEILING requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISO.CEILING allows at most 2 arguments")
|
|
}
|
|
var significance float64
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if number.Number < 0 {
|
|
significance = -1
|
|
}
|
|
if argsList.Len() == 1 {
|
|
return newNumberFormulaArg(math.Ceil(number.Number))
|
|
}
|
|
if argsList.Len() > 1 {
|
|
s := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if s.Type == ArgError {
|
|
return s
|
|
}
|
|
significance = s.Number
|
|
significance = math.Abs(significance)
|
|
if significance == 0 {
|
|
return newNumberFormulaArg(significance)
|
|
}
|
|
}
|
|
val, res := math.Modf(number.Number / significance)
|
|
if res != 0 {
|
|
if number.Number > 0 {
|
|
val++
|
|
}
|
|
}
|
|
return newNumberFormulaArg(val * significance)
|
|
}
|
|
|
|
// lcm returns the least common multiple of two supplied integers.
|
|
func lcm(a, b float64) float64 {
|
|
a = math.Trunc(a)
|
|
b = math.Trunc(b)
|
|
if a == 0 && b == 0 {
|
|
return 0
|
|
}
|
|
return a * b / gcd(a, b)
|
|
}
|
|
|
|
// LCM function returns the least common multiple of two or more supplied
|
|
// integers. The syntax of the function is:
|
|
//
|
|
// LCM(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) LCM(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LCM requires at least 1 argument")
|
|
}
|
|
var (
|
|
val float64
|
|
nums = []float64{}
|
|
err error
|
|
)
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
if token.String == "" {
|
|
continue
|
|
}
|
|
if val, err = strconv.ParseFloat(token.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
case ArgNumber:
|
|
val = token.Number
|
|
}
|
|
nums = append(nums, val)
|
|
}
|
|
if nums[0] < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LCM only accepts positive arguments")
|
|
}
|
|
if len(nums) == 1 {
|
|
return newNumberFormulaArg(nums[0])
|
|
}
|
|
cm := nums[0]
|
|
for i := 1; i < len(nums); i++ {
|
|
if nums[i] < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LCM only accepts positive arguments")
|
|
}
|
|
cm = lcm(cm, nums[i])
|
|
}
|
|
return newNumberFormulaArg(cm)
|
|
}
|
|
|
|
// LN function calculates the natural logarithm of a given number. The syntax
|
|
// of the function is:
|
|
//
|
|
// LN(number)
|
|
//
|
|
func (fn *formulaFuncs) LN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LN requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Log(number.Number))
|
|
}
|
|
|
|
// LOG function calculates the logarithm of a given number, to a supplied
|
|
// base. The syntax of the function is:
|
|
//
|
|
// LOG(number,[base])
|
|
//
|
|
func (fn *formulaFuncs) LOG(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LOG requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LOG allows at most 2 arguments")
|
|
}
|
|
base := 10.0
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if argsList.Len() > 1 {
|
|
b := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if b.Type == ArgError {
|
|
return b
|
|
}
|
|
base = b.Number
|
|
}
|
|
if number.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorDIV)
|
|
}
|
|
if base == 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorDIV)
|
|
}
|
|
if base == 1 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg(math.Log(number.Number) / math.Log(base))
|
|
}
|
|
|
|
// LOG10 function calculates the base 10 logarithm of a given number. The
|
|
// syntax of the function is:
|
|
//
|
|
// LOG10(number)
|
|
//
|
|
func (fn *formulaFuncs) LOG10(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LOG10 requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Log10(number.Number))
|
|
}
|
|
|
|
// minor function implement a minor of a matrix A is the determinant of some
|
|
// smaller square matrix.
|
|
func minor(sqMtx [][]float64, idx int) [][]float64 {
|
|
ret := [][]float64{}
|
|
for i := range sqMtx {
|
|
if i == 0 {
|
|
continue
|
|
}
|
|
row := []float64{}
|
|
for j := range sqMtx {
|
|
if j == idx {
|
|
continue
|
|
}
|
|
row = append(row, sqMtx[i][j])
|
|
}
|
|
ret = append(ret, row)
|
|
}
|
|
return ret
|
|
}
|
|
|
|
// det determinant of the 2x2 matrix.
|
|
func det(sqMtx [][]float64) float64 {
|
|
if len(sqMtx) == 2 {
|
|
m00 := sqMtx[0][0]
|
|
m01 := sqMtx[0][1]
|
|
m10 := sqMtx[1][0]
|
|
m11 := sqMtx[1][1]
|
|
return m00*m11 - m10*m01
|
|
}
|
|
var res, sgn float64 = 0, 1
|
|
for j := range sqMtx {
|
|
res += sgn * sqMtx[0][j] * det(minor(sqMtx, j))
|
|
sgn *= -1
|
|
}
|
|
return res
|
|
}
|
|
|
|
// MDETERM calculates the determinant of a square matrix. The
|
|
// syntax of the function is:
|
|
//
|
|
// MDETERM(array)
|
|
//
|
|
func (fn *formulaFuncs) MDETERM(argsList *list.List) (result formulaArg) {
|
|
var (
|
|
num float64
|
|
numMtx = [][]float64{}
|
|
err error
|
|
strMtx [][]formulaArg
|
|
)
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MDETERM requires at least 1 argument")
|
|
}
|
|
strMtx = argsList.Front().Value.(formulaArg).Matrix
|
|
var rows = len(strMtx)
|
|
for _, row := range argsList.Front().Value.(formulaArg).Matrix {
|
|
if len(row) != rows {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
numRow := []float64{}
|
|
for _, ele := range row {
|
|
if num, err = strconv.ParseFloat(ele.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
numRow = append(numRow, num)
|
|
}
|
|
numMtx = append(numMtx, numRow)
|
|
}
|
|
return newNumberFormulaArg(det(numMtx))
|
|
}
|
|
|
|
// MOD function returns the remainder of a division between two supplied
|
|
// numbers. The syntax of the function is:
|
|
//
|
|
// MOD(number,divisor)
|
|
//
|
|
func (fn *formulaFuncs) MOD(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MOD requires 2 numeric arguments")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
divisor := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if divisor.Type == ArgError {
|
|
return divisor
|
|
}
|
|
if divisor.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, "MOD divide by zero")
|
|
}
|
|
trunc, rem := math.Modf(number.Number / divisor.Number)
|
|
if rem < 0 {
|
|
trunc--
|
|
}
|
|
return newNumberFormulaArg(number.Number - divisor.Number*trunc)
|
|
}
|
|
|
|
// MROUND function rounds a supplied number up or down to the nearest multiple
|
|
// of a given number. The syntax of the function is:
|
|
//
|
|
// MROUND(number,multiple)
|
|
//
|
|
func (fn *formulaFuncs) MROUND(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MROUND requires 2 numeric arguments")
|
|
}
|
|
n := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if n.Type == ArgError {
|
|
return n
|
|
}
|
|
multiple := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if multiple.Type == ArgError {
|
|
return multiple
|
|
}
|
|
if multiple.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
if multiple.Number < 0 && n.Number > 0 ||
|
|
multiple.Number > 0 && n.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
number, res := math.Modf(n.Number / multiple.Number)
|
|
if math.Trunc(res+0.5) > 0 {
|
|
number++
|
|
}
|
|
return newNumberFormulaArg(number * multiple.Number)
|
|
}
|
|
|
|
// MULTINOMIAL function calculates the ratio of the factorial of a sum of
|
|
// supplied values to the product of factorials of those values. The syntax of
|
|
// the function is:
|
|
//
|
|
// MULTINOMIAL(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) MULTINOMIAL(argsList *list.List) formulaArg {
|
|
val, num, denom := 0.0, 0.0, 1.0
|
|
var err error
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
if token.String == "" {
|
|
continue
|
|
}
|
|
if val, err = strconv.ParseFloat(token.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
case ArgNumber:
|
|
val = token.Number
|
|
}
|
|
num += val
|
|
denom *= fact(val)
|
|
}
|
|
return newNumberFormulaArg(fact(num) / denom)
|
|
}
|
|
|
|
// MUNIT function returns the unit matrix for a specified dimension. The
|
|
// syntax of the function is:
|
|
//
|
|
// MUNIT(dimension)
|
|
//
|
|
func (fn *formulaFuncs) MUNIT(argsList *list.List) (result formulaArg) {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MUNIT requires 1 numeric argument")
|
|
}
|
|
dimension := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if dimension.Type == ArgError || dimension.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, dimension.Error)
|
|
}
|
|
matrix := make([][]formulaArg, 0, int(dimension.Number))
|
|
for i := 0; i < int(dimension.Number); i++ {
|
|
row := make([]formulaArg, int(dimension.Number))
|
|
for j := 0; j < int(dimension.Number); j++ {
|
|
if i == j {
|
|
row[j] = newNumberFormulaArg(1.0)
|
|
} else {
|
|
row[j] = newNumberFormulaArg(0.0)
|
|
}
|
|
}
|
|
matrix = append(matrix, row)
|
|
}
|
|
return newMatrixFormulaArg(matrix)
|
|
}
|
|
|
|
// ODD function ounds a supplied number away from zero (i.e. rounds a positive
|
|
// number up and a negative number down), to the next odd number. The syntax
|
|
// of the function is:
|
|
//
|
|
// ODD(number)
|
|
//
|
|
func (fn *formulaFuncs) ODD(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ODD requires 1 numeric argument")
|
|
}
|
|
number := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if number.Number == 0 {
|
|
return newNumberFormulaArg(1)
|
|
}
|
|
sign := math.Signbit(number.Number)
|
|
m, frac := math.Modf((number.Number - 1) / 2)
|
|
val := m*2 + 1
|
|
if frac != 0 {
|
|
if !sign {
|
|
val += 2
|
|
} else {
|
|
val -= 2
|
|
}
|
|
}
|
|
return newNumberFormulaArg(val)
|
|
}
|
|
|
|
// PI function returns the value of the mathematical constant π (pi), accurate
|
|
// to 15 digits (14 decimal places). The syntax of the function is:
|
|
//
|
|
// PI()
|
|
//
|
|
func (fn *formulaFuncs) PI(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "PI accepts no arguments")
|
|
}
|
|
return newNumberFormulaArg(math.Pi)
|
|
}
|
|
|
|
// POWER function calculates a given number, raised to a supplied power.
|
|
// The syntax of the function is:
|
|
//
|
|
// POWER(number,power)
|
|
//
|
|
func (fn *formulaFuncs) POWER(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "POWER requires 2 numeric arguments")
|
|
}
|
|
x := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if x.Type == ArgError {
|
|
return x
|
|
}
|
|
y := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if y.Type == ArgError {
|
|
return y
|
|
}
|
|
if x.Number == 0 && y.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
if x.Number == 0 && y.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg(math.Pow(x.Number, y.Number))
|
|
}
|
|
|
|
// PRODUCT function returns the product (multiplication) of a supplied set of
|
|
// numerical values. The syntax of the function is:
|
|
//
|
|
// PRODUCT(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) PRODUCT(argsList *list.List) formulaArg {
|
|
val, product := 0.0, 1.0
|
|
var err error
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgUnknown:
|
|
continue
|
|
case ArgString:
|
|
if token.String == "" {
|
|
continue
|
|
}
|
|
if val, err = strconv.ParseFloat(token.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
product = product * val
|
|
case ArgNumber:
|
|
product = product * token.Number
|
|
case ArgMatrix:
|
|
for _, row := range token.Matrix {
|
|
for _, value := range row {
|
|
if value.String == "" {
|
|
continue
|
|
}
|
|
if val, err = strconv.ParseFloat(value.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
product = product * val
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return newNumberFormulaArg(product)
|
|
}
|
|
|
|
// QUOTIENT function returns the integer portion of a division between two
|
|
// supplied numbers. The syntax of the function is:
|
|
//
|
|
// QUOTIENT(numerator,denominator)
|
|
//
|
|
func (fn *formulaFuncs) QUOTIENT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "QUOTIENT requires 2 numeric arguments")
|
|
}
|
|
x := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if x.Type == ArgError {
|
|
return x
|
|
}
|
|
y := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if y.Type == ArgError {
|
|
return y
|
|
}
|
|
if y.Number == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
return newNumberFormulaArg(math.Trunc(x.Number / y.Number))
|
|
}
|
|
|
|
// RADIANS function converts radians into degrees. The syntax of the function is:
|
|
//
|
|
// RADIANS(angle)
|
|
//
|
|
func (fn *formulaFuncs) RADIANS(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "RADIANS requires 1 numeric argument")
|
|
}
|
|
angle := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if angle.Type == ArgError {
|
|
return angle
|
|
}
|
|
return newNumberFormulaArg(math.Pi / 180.0 * angle.Number)
|
|
}
|
|
|
|
// RAND function generates a random real number between 0 and 1. The syntax of
|
|
// the function is:
|
|
//
|
|
// RAND()
|
|
//
|
|
func (fn *formulaFuncs) RAND(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "RAND accepts no arguments")
|
|
}
|
|
return newNumberFormulaArg(rand.New(rand.NewSource(time.Now().UnixNano())).Float64())
|
|
}
|
|
|
|
// RANDBETWEEN function generates a random integer between two supplied
|
|
// integers. The syntax of the function is:
|
|
//
|
|
// RANDBETWEEN(bottom,top)
|
|
//
|
|
func (fn *formulaFuncs) RANDBETWEEN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "RANDBETWEEN requires 2 numeric arguments")
|
|
}
|
|
bottom := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if bottom.Type == ArgError {
|
|
return bottom
|
|
}
|
|
top := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if top.Type == ArgError {
|
|
return top
|
|
}
|
|
if top.Number < bottom.Number {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
num := rand.New(rand.NewSource(time.Now().UnixNano())).Int63n(int64(top.Number - bottom.Number + 1))
|
|
return newNumberFormulaArg(float64(num + int64(bottom.Number)))
|
|
}
|
|
|
|
// romanNumerals defined a numeral system that originated in ancient Rome and
|
|
// remained the usual way of writing numbers throughout Europe well into the
|
|
// Late Middle Ages.
|
|
type romanNumerals struct {
|
|
n float64
|
|
s string
|
|
}
|
|
|
|
var romanTable = [][]romanNumerals{
|
|
{
|
|
{1000, "M"}, {900, "CM"}, {500, "D"}, {400, "CD"}, {100, "C"}, {90, "XC"},
|
|
{50, "L"}, {40, "XL"}, {10, "X"}, {9, "IX"}, {5, "V"}, {4, "IV"}, {1, "I"},
|
|
},
|
|
{
|
|
{1000, "M"}, {950, "LM"}, {900, "CM"}, {500, "D"}, {450, "LD"}, {400, "CD"},
|
|
{100, "C"}, {95, "VC"}, {90, "XC"}, {50, "L"}, {45, "VL"}, {40, "XL"},
|
|
{10, "X"}, {9, "IX"}, {5, "V"}, {4, "IV"}, {1, "I"},
|
|
},
|
|
{
|
|
{1000, "M"}, {990, "XM"}, {950, "LM"}, {900, "CM"}, {500, "D"}, {490, "XD"},
|
|
{450, "LD"}, {400, "CD"}, {100, "C"}, {99, "IC"}, {90, "XC"}, {50, "L"},
|
|
{45, "VL"}, {40, "XL"}, {10, "X"}, {9, "IX"}, {5, "V"}, {4, "IV"}, {1, "I"},
|
|
},
|
|
{
|
|
{1000, "M"}, {995, "VM"}, {990, "XM"}, {950, "LM"}, {900, "CM"}, {500, "D"},
|
|
{495, "VD"}, {490, "XD"}, {450, "LD"}, {400, "CD"}, {100, "C"}, {99, "IC"},
|
|
{90, "XC"}, {50, "L"}, {45, "VL"}, {40, "XL"}, {10, "X"}, {9, "IX"},
|
|
{5, "V"}, {4, "IV"}, {1, "I"},
|
|
},
|
|
{
|
|
{1000, "M"}, {999, "IM"}, {995, "VM"}, {990, "XM"}, {950, "LM"}, {900, "CM"},
|
|
{500, "D"}, {499, "ID"}, {495, "VD"}, {490, "XD"}, {450, "LD"}, {400, "CD"},
|
|
{100, "C"}, {99, "IC"}, {90, "XC"}, {50, "L"}, {45, "VL"}, {40, "XL"},
|
|
{10, "X"}, {9, "IX"}, {5, "V"}, {4, "IV"}, {1, "I"},
|
|
},
|
|
}
|
|
|
|
// ROMAN function converts an arabic number to Roman. I.e. for a supplied
|
|
// integer, the function returns a text string depicting the roman numeral
|
|
// form of the number. The syntax of the function is:
|
|
//
|
|
// ROMAN(number,[form])
|
|
//
|
|
func (fn *formulaFuncs) ROMAN(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ROMAN requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ROMAN allows at most 2 arguments")
|
|
}
|
|
var form int
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if argsList.Len() > 1 {
|
|
f := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if f.Type == ArgError {
|
|
return f
|
|
}
|
|
form = int(f.Number)
|
|
if form < 0 {
|
|
form = 0
|
|
} else if form > 4 {
|
|
form = 4
|
|
}
|
|
}
|
|
decimalTable := romanTable[0]
|
|
switch form {
|
|
case 1:
|
|
decimalTable = romanTable[1]
|
|
case 2:
|
|
decimalTable = romanTable[2]
|
|
case 3:
|
|
decimalTable = romanTable[3]
|
|
case 4:
|
|
decimalTable = romanTable[4]
|
|
}
|
|
val := math.Trunc(number.Number)
|
|
buf := bytes.Buffer{}
|
|
for _, r := range decimalTable {
|
|
for val >= r.n {
|
|
buf.WriteString(r.s)
|
|
val -= r.n
|
|
}
|
|
}
|
|
return newStringFormulaArg(buf.String())
|
|
}
|
|
|
|
type roundMode byte
|
|
|
|
const (
|
|
closest roundMode = iota
|
|
down
|
|
up
|
|
)
|
|
|
|
// round rounds a supplied number up or down.
|
|
func (fn *formulaFuncs) round(number, digits float64, mode roundMode) float64 {
|
|
var significance float64
|
|
if digits > 0 {
|
|
significance = math.Pow(1/10.0, digits)
|
|
} else {
|
|
significance = math.Pow(10.0, -digits)
|
|
}
|
|
val, res := math.Modf(number / significance)
|
|
switch mode {
|
|
case closest:
|
|
const eps = 0.499999999
|
|
if res >= eps {
|
|
val++
|
|
} else if res <= -eps {
|
|
val--
|
|
}
|
|
case down:
|
|
case up:
|
|
if res > 0 {
|
|
val++
|
|
} else if res < 0 {
|
|
val--
|
|
}
|
|
}
|
|
return val * significance
|
|
}
|
|
|
|
// ROUND function rounds a supplied number up or down, to a specified number
|
|
// of decimal places. The syntax of the function is:
|
|
//
|
|
// ROUND(number,num_digits)
|
|
//
|
|
func (fn *formulaFuncs) ROUND(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ROUND requires 2 numeric arguments")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
digits := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if digits.Type == ArgError {
|
|
return digits
|
|
}
|
|
return newNumberFormulaArg(fn.round(number.Number, digits.Number, closest))
|
|
}
|
|
|
|
// ROUNDDOWN function rounds a supplied number down towards zero, to a
|
|
// specified number of decimal places. The syntax of the function is:
|
|
//
|
|
// ROUNDDOWN(number,num_digits)
|
|
//
|
|
func (fn *formulaFuncs) ROUNDDOWN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ROUNDDOWN requires 2 numeric arguments")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
digits := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if digits.Type == ArgError {
|
|
return digits
|
|
}
|
|
return newNumberFormulaArg(fn.round(number.Number, digits.Number, down))
|
|
}
|
|
|
|
// ROUNDUP function rounds a supplied number up, away from zero, to a
|
|
// specified number of decimal places. The syntax of the function is:
|
|
//
|
|
// ROUNDUP(number,num_digits)
|
|
//
|
|
func (fn *formulaFuncs) ROUNDUP(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ROUNDUP requires 2 numeric arguments")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
digits := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if digits.Type == ArgError {
|
|
return digits
|
|
}
|
|
return newNumberFormulaArg(fn.round(number.Number, digits.Number, up))
|
|
}
|
|
|
|
// SEC function calculates the secant of a given angle. The syntax of the
|
|
// function is:
|
|
//
|
|
// SEC(number)
|
|
//
|
|
func (fn *formulaFuncs) SEC(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SEC requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Cos(number.Number))
|
|
}
|
|
|
|
// SECH function calculates the hyperbolic secant (sech) of a supplied angle.
|
|
// The syntax of the function is:
|
|
//
|
|
// SECH(number)
|
|
//
|
|
func (fn *formulaFuncs) SECH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SECH requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(1 / math.Cosh(number.Number))
|
|
}
|
|
|
|
// SIGN function returns the arithmetic sign (+1, -1 or 0) of a supplied
|
|
// number. I.e. if the number is positive, the Sign function returns +1, if
|
|
// the number is negative, the function returns -1 and if the number is 0
|
|
// (zero), the function returns 0. The syntax of the function is:
|
|
//
|
|
// SIGN(number)
|
|
//
|
|
func (fn *formulaFuncs) SIGN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SIGN requires 1 numeric argument")
|
|
}
|
|
val := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if val.Type == ArgError {
|
|
return val
|
|
}
|
|
if val.Number < 0 {
|
|
return newNumberFormulaArg(-1)
|
|
}
|
|
if val.Number > 0 {
|
|
return newNumberFormulaArg(1)
|
|
}
|
|
return newNumberFormulaArg(0)
|
|
}
|
|
|
|
// SIN function calculates the sine of a given angle. The syntax of the
|
|
// function is:
|
|
//
|
|
// SIN(number)
|
|
//
|
|
func (fn *formulaFuncs) SIN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SIN requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Sin(number.Number))
|
|
}
|
|
|
|
// SINH function calculates the hyperbolic sine (sinh) of a supplied number.
|
|
// The syntax of the function is:
|
|
//
|
|
// SINH(number)
|
|
//
|
|
func (fn *formulaFuncs) SINH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SINH requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Sinh(number.Number))
|
|
}
|
|
|
|
// SQRT function calculates the positive square root of a supplied number. The
|
|
// syntax of the function is:
|
|
//
|
|
// SQRT(number)
|
|
//
|
|
func (fn *formulaFuncs) SQRT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SQRT requires 1 numeric argument")
|
|
}
|
|
value := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if value.Type == ArgError {
|
|
return value
|
|
}
|
|
if value.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorNUM, formulaErrorNUM)
|
|
}
|
|
return newNumberFormulaArg(math.Sqrt(value.Number))
|
|
}
|
|
|
|
// SQRTPI function returns the square root of a supplied number multiplied by
|
|
// the mathematical constant, π. The syntax of the function is:
|
|
//
|
|
// SQRTPI(number)
|
|
//
|
|
func (fn *formulaFuncs) SQRTPI(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SQRTPI requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Sqrt(number.Number * math.Pi))
|
|
}
|
|
|
|
// STDEV function calculates the sample standard deviation of a supplied set
|
|
// of values. The syntax of the function is:
|
|
//
|
|
// STDEV(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) STDEV(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "STDEV requires at least 1 argument")
|
|
}
|
|
return fn.stdev(false, argsList)
|
|
}
|
|
|
|
// STDEVA function estimates standard deviation based on a sample. The
|
|
// standard deviation is a measure of how widely values are dispersed from
|
|
// the average value (the mean). The syntax of the function is:
|
|
//
|
|
// STDEVA(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) STDEVA(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "STDEVA requires at least 1 argument")
|
|
}
|
|
return fn.stdev(true, argsList)
|
|
}
|
|
|
|
// stdev is an implementation of the formula function STDEV and STDEVA.
|
|
func (fn *formulaFuncs) stdev(stdeva bool, argsList *list.List) formulaArg {
|
|
pow := func(result, count float64, n, m formulaArg) (float64, float64) {
|
|
if result == -1 {
|
|
result = math.Pow((n.Number - m.Number), 2)
|
|
} else {
|
|
result += math.Pow((n.Number - m.Number), 2)
|
|
}
|
|
count++
|
|
return result, count
|
|
}
|
|
count, result := -1.0, -1.0
|
|
var mean formulaArg
|
|
if stdeva {
|
|
mean = fn.AVERAGEA(argsList)
|
|
} else {
|
|
mean = fn.AVERAGE(argsList)
|
|
}
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString, ArgNumber:
|
|
if !stdeva && (token.Value() == "TRUE" || token.Value() == "FALSE") {
|
|
continue
|
|
} else if stdeva && (token.Value() == "TRUE" || token.Value() == "FALSE") {
|
|
num := token.ToBool()
|
|
if num.Type == ArgNumber {
|
|
result, count = pow(result, count, num, mean)
|
|
continue
|
|
}
|
|
} else {
|
|
num := token.ToNumber()
|
|
if num.Type == ArgNumber {
|
|
result, count = pow(result, count, num, mean)
|
|
}
|
|
}
|
|
case ArgList, ArgMatrix:
|
|
for _, row := range token.ToList() {
|
|
if row.Type == ArgNumber || row.Type == ArgString {
|
|
if !stdeva && (row.Value() == "TRUE" || row.Value() == "FALSE") {
|
|
continue
|
|
} else if stdeva && (row.Value() == "TRUE" || row.Value() == "FALSE") {
|
|
num := row.ToBool()
|
|
if num.Type == ArgNumber {
|
|
result, count = pow(result, count, num, mean)
|
|
continue
|
|
}
|
|
} else {
|
|
num := row.ToNumber()
|
|
if num.Type == ArgNumber {
|
|
result, count = pow(result, count, num, mean)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if count > 0 && result >= 0 {
|
|
return newNumberFormulaArg(math.Sqrt(result / count))
|
|
}
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
|
|
// SUM function adds together a supplied set of numbers and returns the sum of
|
|
// these values. The syntax of the function is:
|
|
//
|
|
// SUM(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) SUM(argsList *list.List) formulaArg {
|
|
var sum float64
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgUnknown:
|
|
continue
|
|
case ArgString:
|
|
if num := token.ToNumber(); num.Type == ArgNumber {
|
|
sum += num.Number
|
|
}
|
|
case ArgNumber:
|
|
sum += token.Number
|
|
case ArgMatrix:
|
|
for _, row := range token.Matrix {
|
|
for _, value := range row {
|
|
if num := value.ToNumber(); num.Type == ArgNumber {
|
|
sum += num.Number
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return newNumberFormulaArg(sum)
|
|
}
|
|
|
|
// SUMIF function finds the values in a supplied array, that satisfy a given
|
|
// criteria, and returns the sum of the corresponding values in a second
|
|
// supplied array. The syntax of the function is:
|
|
//
|
|
// SUMIF(range,criteria,[sum_range])
|
|
//
|
|
func (fn *formulaFuncs) SUMIF(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SUMIF requires at least 2 argument")
|
|
}
|
|
var criteria = formulaCriteriaParser(argsList.Front().Next().Value.(formulaArg).String)
|
|
var rangeMtx = argsList.Front().Value.(formulaArg).Matrix
|
|
var sumRange [][]formulaArg
|
|
if argsList.Len() == 3 {
|
|
sumRange = argsList.Back().Value.(formulaArg).Matrix
|
|
}
|
|
var sum, val float64
|
|
var err error
|
|
for rowIdx, row := range rangeMtx {
|
|
for colIdx, col := range row {
|
|
var ok bool
|
|
fromVal := col.String
|
|
if col.String == "" {
|
|
continue
|
|
}
|
|
if ok, err = formulaCriteriaEval(fromVal, criteria); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
if ok {
|
|
if argsList.Len() == 3 {
|
|
if len(sumRange) <= rowIdx || len(sumRange[rowIdx]) <= colIdx {
|
|
continue
|
|
}
|
|
fromVal = sumRange[rowIdx][colIdx].String
|
|
}
|
|
if val, err = strconv.ParseFloat(fromVal, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
sum += val
|
|
}
|
|
}
|
|
}
|
|
return newNumberFormulaArg(sum)
|
|
}
|
|
|
|
// SUMSQ function returns the sum of squares of a supplied set of values. The
|
|
// syntax of the function is:
|
|
//
|
|
// SUMSQ(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) SUMSQ(argsList *list.List) formulaArg {
|
|
var val, sq float64
|
|
var err error
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
if token.String == "" {
|
|
continue
|
|
}
|
|
if val, err = strconv.ParseFloat(token.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
sq += val * val
|
|
case ArgNumber:
|
|
sq += token.Number
|
|
case ArgMatrix:
|
|
for _, row := range token.Matrix {
|
|
for _, value := range row {
|
|
if value.String == "" {
|
|
continue
|
|
}
|
|
if val, err = strconv.ParseFloat(value.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
sq += val * val
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return newNumberFormulaArg(sq)
|
|
}
|
|
|
|
// TAN function calculates the tangent of a given angle. The syntax of the
|
|
// function is:
|
|
//
|
|
// TAN(number)
|
|
//
|
|
func (fn *formulaFuncs) TAN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "TAN requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Tan(number.Number))
|
|
}
|
|
|
|
// TANH function calculates the hyperbolic tangent (tanh) of a supplied
|
|
// number. The syntax of the function is:
|
|
//
|
|
// TANH(number)
|
|
//
|
|
func (fn *formulaFuncs) TANH(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "TANH requires 1 numeric argument")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
return newNumberFormulaArg(math.Tanh(number.Number))
|
|
}
|
|
|
|
// TRUNC function truncates a supplied number to a specified number of decimal
|
|
// places. The syntax of the function is:
|
|
//
|
|
// TRUNC(number,[number_digits])
|
|
//
|
|
func (fn *formulaFuncs) TRUNC(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "TRUNC requires at least 1 argument")
|
|
}
|
|
var digits, adjust, rtrim float64
|
|
var err error
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if number.Type == ArgError {
|
|
return number
|
|
}
|
|
if argsList.Len() > 1 {
|
|
d := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if d.Type == ArgError {
|
|
return d
|
|
}
|
|
digits = d.Number
|
|
digits = math.Floor(digits)
|
|
}
|
|
adjust = math.Pow(10, digits)
|
|
x := int((math.Abs(number.Number) - math.Abs(float64(int(number.Number)))) * adjust)
|
|
if x != 0 {
|
|
if rtrim, err = strconv.ParseFloat(strings.TrimRight(strconv.Itoa(x), "0"), 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
}
|
|
if (digits > 0) && (rtrim < adjust/10) {
|
|
return newNumberFormulaArg(number.Number)
|
|
}
|
|
return newNumberFormulaArg(float64(int(number.Number*adjust)) / adjust)
|
|
}
|
|
|
|
// Statistical Functions
|
|
|
|
// AVERAGE function returns the arithmetic mean of a list of supplied numbers.
|
|
// The syntax of the function is:
|
|
//
|
|
// AVERAGE(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) AVERAGE(argsList *list.List) formulaArg {
|
|
args := []formulaArg{}
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
args = append(args, arg.Value.(formulaArg))
|
|
}
|
|
count, sum := fn.countSum(false, args)
|
|
if count == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, "AVERAGE divide by zero")
|
|
}
|
|
return newNumberFormulaArg(sum / count)
|
|
}
|
|
|
|
// AVERAGEA function returns the arithmetic mean of a list of supplied numbers
|
|
// with text cell and zero values. The syntax of the function is:
|
|
//
|
|
// AVERAGEA(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) AVERAGEA(argsList *list.List) formulaArg {
|
|
args := []formulaArg{}
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
args = append(args, arg.Value.(formulaArg))
|
|
}
|
|
count, sum := fn.countSum(true, args)
|
|
if count == 0 {
|
|
return newErrorFormulaArg(formulaErrorDIV, "AVERAGEA divide by zero")
|
|
}
|
|
return newNumberFormulaArg(sum / count)
|
|
}
|
|
|
|
// countSum get count and sum for a formula arguments array.
|
|
func (fn *formulaFuncs) countSum(countText bool, args []formulaArg) (count, sum float64) {
|
|
for _, arg := range args {
|
|
switch arg.Type {
|
|
case ArgNumber:
|
|
if countText || !arg.Boolean {
|
|
sum += arg.Number
|
|
count++
|
|
}
|
|
case ArgString:
|
|
if !countText && (arg.Value() == "TRUE" || arg.Value() == "FALSE") {
|
|
continue
|
|
} else if countText && (arg.Value() == "TRUE" || arg.Value() == "FALSE") {
|
|
num := arg.ToBool()
|
|
if num.Type == ArgNumber {
|
|
count++
|
|
sum += num.Number
|
|
continue
|
|
}
|
|
}
|
|
num := arg.ToNumber()
|
|
if countText && num.Type == ArgError && arg.String != "" {
|
|
count++
|
|
}
|
|
if num.Type == ArgNumber {
|
|
sum += num.Number
|
|
count++
|
|
}
|
|
case ArgList, ArgMatrix:
|
|
cnt, summary := fn.countSum(countText, arg.ToList())
|
|
sum += summary
|
|
count += cnt
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// COUNT function returns the count of numeric values in a supplied set of
|
|
// cells or values. This count includes both numbers and dates. The syntax of
|
|
// the function is:
|
|
//
|
|
// COUNT(value1,[value2],...)
|
|
//
|
|
func (fn *formulaFuncs) COUNT(argsList *list.List) formulaArg {
|
|
var count int
|
|
for token := argsList.Front(); token != nil; token = token.Next() {
|
|
arg := token.Value.(formulaArg)
|
|
switch arg.Type {
|
|
case ArgString:
|
|
if arg.ToNumber().Type != ArgError {
|
|
count++
|
|
}
|
|
case ArgNumber:
|
|
count++
|
|
case ArgMatrix:
|
|
for _, row := range arg.Matrix {
|
|
for _, value := range row {
|
|
if value.ToNumber().Type != ArgError {
|
|
count++
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return newNumberFormulaArg(float64(count))
|
|
}
|
|
|
|
// COUNTA function returns the number of non-blanks within a supplied set of
|
|
// cells or values. The syntax of the function is:
|
|
//
|
|
// COUNTA(value1,[value2],...)
|
|
//
|
|
func (fn *formulaFuncs) COUNTA(argsList *list.List) formulaArg {
|
|
var count int
|
|
for token := argsList.Front(); token != nil; token = token.Next() {
|
|
arg := token.Value.(formulaArg)
|
|
switch arg.Type {
|
|
case ArgString:
|
|
if arg.String != "" {
|
|
count++
|
|
}
|
|
case ArgNumber:
|
|
count++
|
|
case ArgMatrix:
|
|
for _, row := range arg.ToList() {
|
|
switch row.Type {
|
|
case ArgString:
|
|
if row.String != "" {
|
|
count++
|
|
}
|
|
case ArgNumber:
|
|
count++
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return newNumberFormulaArg(float64(count))
|
|
}
|
|
|
|
// COUNTBLANK function returns the number of blank cells in a supplied range.
|
|
// The syntax of the function is:
|
|
//
|
|
// COUNTBLANK(range)
|
|
//
|
|
func (fn *formulaFuncs) COUNTBLANK(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COUNTBLANK requires 1 argument")
|
|
}
|
|
var count int
|
|
token := argsList.Front().Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
if token.String == "" {
|
|
count++
|
|
}
|
|
case ArgList, ArgMatrix:
|
|
for _, row := range token.ToList() {
|
|
switch row.Type {
|
|
case ArgString:
|
|
if row.String == "" {
|
|
count++
|
|
}
|
|
case ArgEmpty:
|
|
count++
|
|
}
|
|
}
|
|
case ArgEmpty:
|
|
count++
|
|
}
|
|
return newNumberFormulaArg(float64(count))
|
|
}
|
|
|
|
// FISHER function calculates the Fisher Transformation for a supplied value.
|
|
// The syntax of the function is:
|
|
//
|
|
// FISHER(x)
|
|
//
|
|
func (fn *formulaFuncs) FISHER(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FISHER requires 1 numeric argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
arg := token.ToNumber()
|
|
if arg.Type == ArgNumber {
|
|
if arg.Number <= -1 || arg.Number >= 1 {
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
return newNumberFormulaArg(0.5 * math.Log((1+arg.Number)/(1-arg.Number)))
|
|
}
|
|
case ArgNumber:
|
|
if token.Number <= -1 || token.Number >= 1 {
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
return newNumberFormulaArg(0.5 * math.Log((1+token.Number)/(1-token.Number)))
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FISHER requires 1 numeric argument")
|
|
}
|
|
|
|
// FISHERINV function calculates the inverse of the Fisher Transformation and
|
|
// returns a value between -1 and +1. The syntax of the function is:
|
|
//
|
|
// FISHERINV(y)
|
|
//
|
|
func (fn *formulaFuncs) FISHERINV(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FISHERINV requires 1 numeric argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
arg := token.ToNumber()
|
|
if arg.Type == ArgNumber {
|
|
return newNumberFormulaArg((math.Exp(2*arg.Number) - 1) / (math.Exp(2*arg.Number) + 1))
|
|
}
|
|
case ArgNumber:
|
|
return newNumberFormulaArg((math.Exp(2*token.Number) - 1) / (math.Exp(2*token.Number) + 1))
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FISHERINV requires 1 numeric argument")
|
|
}
|
|
|
|
// GAMMA function returns the value of the Gamma Function, Γ(n), for a
|
|
// specified number, n. The syntax of the function is:
|
|
//
|
|
// GAMMA(number)
|
|
//
|
|
func (fn *formulaFuncs) GAMMA(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "GAMMA requires 1 numeric argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
arg := token.ToNumber()
|
|
if arg.Type == ArgNumber {
|
|
if arg.Number <= 0 {
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
return newNumberFormulaArg(math.Gamma(arg.Number))
|
|
}
|
|
case ArgNumber:
|
|
if token.Number <= 0 {
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
return newNumberFormulaArg(math.Gamma(token.Number))
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, "GAMMA requires 1 numeric argument")
|
|
}
|
|
|
|
// GAMMALN function returns the natural logarithm of the Gamma Function, Γ
|
|
// (n). The syntax of the function is:
|
|
//
|
|
// GAMMALN(x)
|
|
//
|
|
func (fn *formulaFuncs) GAMMALN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "GAMMALN requires 1 numeric argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
arg := token.ToNumber()
|
|
if arg.Type == ArgNumber {
|
|
if arg.Number <= 0 {
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
return newNumberFormulaArg(math.Log(math.Gamma(arg.Number)))
|
|
}
|
|
case ArgNumber:
|
|
if token.Number <= 0 {
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
return newNumberFormulaArg(math.Log(math.Gamma(token.Number)))
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, "GAMMALN requires 1 numeric argument")
|
|
}
|
|
|
|
// KURT function calculates the kurtosis of a supplied set of values. The
|
|
// syntax of the function is:
|
|
//
|
|
// KURT(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) KURT(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "KURT requires at least 1 argument")
|
|
}
|
|
mean, stdev := fn.AVERAGE(argsList), fn.STDEV(argsList)
|
|
if stdev.Number > 0 {
|
|
count, summer := 0.0, 0.0
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString, ArgNumber:
|
|
num := token.ToNumber()
|
|
if num.Type == ArgError {
|
|
continue
|
|
}
|
|
summer += math.Pow((num.Number-mean.Number)/stdev.Number, 4)
|
|
count++
|
|
case ArgList, ArgMatrix:
|
|
for _, row := range token.ToList() {
|
|
if row.Type == ArgNumber || row.Type == ArgString {
|
|
num := row.ToNumber()
|
|
if num.Type == ArgError {
|
|
continue
|
|
}
|
|
summer += math.Pow((num.Number-mean.Number)/stdev.Number, 4)
|
|
count++
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if count > 3 {
|
|
return newNumberFormulaArg(summer*(count*(count+1)/((count-1)*(count-2)*(count-3))) - (3 * math.Pow(count-1, 2) / ((count - 2) * (count - 3))))
|
|
}
|
|
}
|
|
return newErrorFormulaArg(formulaErrorDIV, formulaErrorDIV)
|
|
}
|
|
|
|
// kth is an implementation of the formula function LARGE and SMALL.
|
|
func (fn *formulaFuncs) kth(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires 2 arguments", name))
|
|
}
|
|
array := argsList.Front().Value.(formulaArg).ToList()
|
|
kArg := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if kArg.Type != ArgNumber {
|
|
return kArg
|
|
}
|
|
k := int(kArg.Number)
|
|
if k < 1 {
|
|
return newErrorFormulaArg(formulaErrorNUM, "k should be > 0")
|
|
}
|
|
data := []float64{}
|
|
for _, arg := range array {
|
|
if numArg := arg.ToNumber(); numArg.Type == ArgNumber {
|
|
data = append(data, numArg.Number)
|
|
}
|
|
}
|
|
if len(data) < k {
|
|
return newErrorFormulaArg(formulaErrorNUM, "k should be <= length of array")
|
|
}
|
|
sort.Float64s(data)
|
|
if name == "LARGE" {
|
|
return newNumberFormulaArg(data[len(data)-k])
|
|
}
|
|
return newNumberFormulaArg(data[k-1])
|
|
}
|
|
|
|
// LARGE function returns the k'th largest value from an array of numeric
|
|
// values. The syntax of the function is:
|
|
//
|
|
// LARGE(array,k)
|
|
//
|
|
func (fn *formulaFuncs) LARGE(argsList *list.List) formulaArg {
|
|
return fn.kth("LARGE", argsList)
|
|
}
|
|
|
|
// MAX function returns the largest value from a supplied set of numeric
|
|
// values. The syntax of the function is:
|
|
//
|
|
// MAX(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) MAX(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MAX requires at least 1 argument")
|
|
}
|
|
return fn.max(false, argsList)
|
|
}
|
|
|
|
// MAXA function returns the largest value from a supplied set of numeric
|
|
// values, while counting text and the logical value FALSE as the value 0 and
|
|
// counting the logical value TRUE as the value 1. The syntax of the function
|
|
// is:
|
|
//
|
|
// MAXA(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) MAXA(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MAXA requires at least 1 argument")
|
|
}
|
|
return fn.max(true, argsList)
|
|
}
|
|
|
|
// max is an implementation of the formula function MAX and MAXA.
|
|
func (fn *formulaFuncs) max(maxa bool, argsList *list.List) formulaArg {
|
|
max := -math.MaxFloat64
|
|
for token := argsList.Front(); token != nil; token = token.Next() {
|
|
arg := token.Value.(formulaArg)
|
|
switch arg.Type {
|
|
case ArgString:
|
|
if !maxa && (arg.Value() == "TRUE" || arg.Value() == "FALSE") {
|
|
continue
|
|
} else {
|
|
num := arg.ToBool()
|
|
if num.Type == ArgNumber && num.Number > max {
|
|
max = num.Number
|
|
continue
|
|
}
|
|
}
|
|
num := arg.ToNumber()
|
|
if num.Type != ArgError && num.Number > max {
|
|
max = num.Number
|
|
}
|
|
case ArgNumber:
|
|
if arg.Number > max {
|
|
max = arg.Number
|
|
}
|
|
case ArgList, ArgMatrix:
|
|
for _, row := range arg.ToList() {
|
|
switch row.Type {
|
|
case ArgString:
|
|
if !maxa && (row.Value() == "TRUE" || row.Value() == "FALSE") {
|
|
continue
|
|
} else {
|
|
num := row.ToBool()
|
|
if num.Type == ArgNumber && num.Number > max {
|
|
max = num.Number
|
|
continue
|
|
}
|
|
}
|
|
num := row.ToNumber()
|
|
if num.Type != ArgError && num.Number > max {
|
|
max = num.Number
|
|
}
|
|
case ArgNumber:
|
|
if row.Number > max {
|
|
max = row.Number
|
|
}
|
|
}
|
|
}
|
|
case ArgError:
|
|
return arg
|
|
}
|
|
}
|
|
if max == -math.MaxFloat64 {
|
|
max = 0
|
|
}
|
|
return newNumberFormulaArg(max)
|
|
}
|
|
|
|
// MEDIAN function returns the statistical median (the middle value) of a list
|
|
// of supplied numbers. The syntax of the function is:
|
|
//
|
|
// MEDIAN(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) MEDIAN(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MEDIAN requires at least 1 argument")
|
|
}
|
|
var values = []float64{}
|
|
var median, digits float64
|
|
var err error
|
|
for token := argsList.Front(); token != nil; token = token.Next() {
|
|
arg := token.Value.(formulaArg)
|
|
switch arg.Type {
|
|
case ArgString:
|
|
num := arg.ToNumber()
|
|
if num.Type == ArgError {
|
|
return newErrorFormulaArg(formulaErrorVALUE, num.Error)
|
|
}
|
|
values = append(values, num.Number)
|
|
case ArgNumber:
|
|
values = append(values, arg.Number)
|
|
case ArgMatrix:
|
|
for _, row := range arg.Matrix {
|
|
for _, value := range row {
|
|
if value.String == "" {
|
|
continue
|
|
}
|
|
if digits, err = strconv.ParseFloat(value.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
values = append(values, digits)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
sort.Float64s(values)
|
|
if len(values)%2 == 0 {
|
|
median = (values[len(values)/2-1] + values[len(values)/2]) / 2
|
|
} else {
|
|
median = values[len(values)/2]
|
|
}
|
|
return newNumberFormulaArg(median)
|
|
}
|
|
|
|
// MIN function returns the smallest value from a supplied set of numeric
|
|
// values. The syntax of the function is:
|
|
//
|
|
// MIN(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) MIN(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MIN requires at least 1 argument")
|
|
}
|
|
return fn.min(false, argsList)
|
|
}
|
|
|
|
// MINA function returns the smallest value from a supplied set of numeric
|
|
// values, while counting text and the logical value FALSE as the value 0 and
|
|
// counting the logical value TRUE as the value 1. The syntax of the function
|
|
// is:
|
|
//
|
|
// MINA(number1,[number2],...)
|
|
//
|
|
func (fn *formulaFuncs) MINA(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "MINA requires at least 1 argument")
|
|
}
|
|
return fn.min(true, argsList)
|
|
}
|
|
|
|
// min is an implementation of the formula function MIN and MINA.
|
|
func (fn *formulaFuncs) min(mina bool, argsList *list.List) formulaArg {
|
|
min := math.MaxFloat64
|
|
for token := argsList.Front(); token != nil; token = token.Next() {
|
|
arg := token.Value.(formulaArg)
|
|
switch arg.Type {
|
|
case ArgString:
|
|
if !mina && (arg.Value() == "TRUE" || arg.Value() == "FALSE") {
|
|
continue
|
|
} else {
|
|
num := arg.ToBool()
|
|
if num.Type == ArgNumber && num.Number < min {
|
|
min = num.Number
|
|
continue
|
|
}
|
|
}
|
|
num := arg.ToNumber()
|
|
if num.Type != ArgError && num.Number < min {
|
|
min = num.Number
|
|
}
|
|
case ArgNumber:
|
|
if arg.Number < min {
|
|
min = arg.Number
|
|
}
|
|
case ArgList, ArgMatrix:
|
|
for _, row := range arg.ToList() {
|
|
switch row.Type {
|
|
case ArgString:
|
|
if !mina && (row.Value() == "TRUE" || row.Value() == "FALSE") {
|
|
continue
|
|
} else {
|
|
num := row.ToBool()
|
|
if num.Type == ArgNumber && num.Number < min {
|
|
min = num.Number
|
|
continue
|
|
}
|
|
}
|
|
num := row.ToNumber()
|
|
if num.Type != ArgError && num.Number < min {
|
|
min = num.Number
|
|
}
|
|
case ArgNumber:
|
|
if row.Number < min {
|
|
min = row.Number
|
|
}
|
|
}
|
|
}
|
|
case ArgError:
|
|
return arg
|
|
}
|
|
}
|
|
if min == math.MaxFloat64 {
|
|
min = 0
|
|
}
|
|
return newNumberFormulaArg(min)
|
|
}
|
|
|
|
// PERMUT function calculates the number of permutations of a specified number
|
|
// of objects from a set of objects. The syntax of the function is:
|
|
//
|
|
// PERMUT(number,number_chosen)
|
|
//
|
|
func (fn *formulaFuncs) PERMUT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "PERMUT requires 2 numeric arguments")
|
|
}
|
|
number := argsList.Front().Value.(formulaArg).ToNumber()
|
|
chosen := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if number.Type != ArgNumber {
|
|
return number
|
|
}
|
|
if chosen.Type != ArgNumber {
|
|
return chosen
|
|
}
|
|
if number.Number < chosen.Number {
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
return newNumberFormulaArg(math.Round(fact(number.Number) / fact(number.Number-chosen.Number)))
|
|
}
|
|
|
|
// SMALL function returns the k'th smallest value from an array of numeric
|
|
// values. The syntax of the function is:
|
|
//
|
|
// SMALL(array,k)
|
|
//
|
|
func (fn *formulaFuncs) SMALL(argsList *list.List) formulaArg {
|
|
return fn.kth("SMALL", argsList)
|
|
}
|
|
|
|
// Information Functions
|
|
|
|
// ISBLANK function tests if a specified cell is blank (empty) and if so,
|
|
// returns TRUE; Otherwise the function returns FALSE. The syntax of the
|
|
// function is:
|
|
//
|
|
// ISBLANK(value)
|
|
//
|
|
func (fn *formulaFuncs) ISBLANK(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISBLANK requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
result := "FALSE"
|
|
switch token.Type {
|
|
case ArgUnknown:
|
|
result = "TRUE"
|
|
case ArgString:
|
|
if token.String == "" {
|
|
result = "TRUE"
|
|
}
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// ISERR function tests if an initial supplied expression (or value) returns
|
|
// any Excel Error, except the #N/A error. If so, the function returns the
|
|
// logical value TRUE; If the supplied value is not an error or is the #N/A
|
|
// error, the ISERR function returns FALSE. The syntax of the function is:
|
|
//
|
|
// ISERR(value)
|
|
//
|
|
func (fn *formulaFuncs) ISERR(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISERR requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
result := "FALSE"
|
|
if token.Type == ArgError {
|
|
for _, errType := range []string{
|
|
formulaErrorDIV, formulaErrorNAME, formulaErrorNUM,
|
|
formulaErrorVALUE, formulaErrorREF, formulaErrorNULL,
|
|
formulaErrorSPILL, formulaErrorCALC, formulaErrorGETTINGDATA,
|
|
} {
|
|
if errType == token.String {
|
|
result = "TRUE"
|
|
}
|
|
}
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// ISERROR function tests if an initial supplied expression (or value) returns
|
|
// an Excel Error, and if so, returns the logical value TRUE; Otherwise the
|
|
// function returns FALSE. The syntax of the function is:
|
|
//
|
|
// ISERROR(value)
|
|
//
|
|
func (fn *formulaFuncs) ISERROR(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISERROR requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
result := "FALSE"
|
|
if token.Type == ArgError {
|
|
for _, errType := range []string{
|
|
formulaErrorDIV, formulaErrorNAME, formulaErrorNA, formulaErrorNUM,
|
|
formulaErrorVALUE, formulaErrorREF, formulaErrorNULL, formulaErrorSPILL,
|
|
formulaErrorCALC, formulaErrorGETTINGDATA,
|
|
} {
|
|
if errType == token.String {
|
|
result = "TRUE"
|
|
}
|
|
}
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// ISEVEN function tests if a supplied number (or numeric expression)
|
|
// evaluates to an even number, and if so, returns TRUE; Otherwise, the
|
|
// function returns FALSE. The syntax of the function is:
|
|
//
|
|
// ISEVEN(value)
|
|
//
|
|
func (fn *formulaFuncs) ISEVEN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISEVEN requires 1 argument")
|
|
}
|
|
var (
|
|
token = argsList.Front().Value.(formulaArg)
|
|
result = "FALSE"
|
|
numeric int
|
|
err error
|
|
)
|
|
if token.Type == ArgString {
|
|
if numeric, err = strconv.Atoi(token.String); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
if numeric == numeric/2*2 {
|
|
return newStringFormulaArg("TRUE")
|
|
}
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// ISNA function tests if an initial supplied expression (or value) returns
|
|
// the Excel #N/A Error, and if so, returns TRUE; Otherwise the function
|
|
// returns FALSE. The syntax of the function is:
|
|
//
|
|
// ISNA(value)
|
|
//
|
|
func (fn *formulaFuncs) ISNA(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISNA requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
result := "FALSE"
|
|
if token.Type == ArgError && token.String == formulaErrorNA {
|
|
result = "TRUE"
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// ISNONTEXT function function tests if a supplied value is text. If not, the
|
|
// function returns TRUE; If the supplied value is text, the function returns
|
|
// FALSE. The syntax of the function is:
|
|
//
|
|
// ISNONTEXT(value)
|
|
//
|
|
func (fn *formulaFuncs) ISNONTEXT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISNONTEXT requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
result := "TRUE"
|
|
if token.Type == ArgString && token.String != "" {
|
|
result = "FALSE"
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// ISNUMBER function function tests if a supplied value is a number. If so,
|
|
// the function returns TRUE; Otherwise it returns FALSE. The syntax of the
|
|
// function is:
|
|
//
|
|
// ISNUMBER(value)
|
|
//
|
|
func (fn *formulaFuncs) ISNUMBER(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISNUMBER requires 1 argument")
|
|
}
|
|
token, result := argsList.Front().Value.(formulaArg), false
|
|
if token.Type == ArgString && token.String != "" {
|
|
if _, err := strconv.Atoi(token.String); err == nil {
|
|
result = true
|
|
}
|
|
}
|
|
return newBoolFormulaArg(result)
|
|
}
|
|
|
|
// ISODD function tests if a supplied number (or numeric expression) evaluates
|
|
// to an odd number, and if so, returns TRUE; Otherwise, the function returns
|
|
// FALSE. The syntax of the function is:
|
|
//
|
|
// ISODD(value)
|
|
//
|
|
func (fn *formulaFuncs) ISODD(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISODD requires 1 argument")
|
|
}
|
|
var (
|
|
token = argsList.Front().Value.(formulaArg)
|
|
result = "FALSE"
|
|
numeric int
|
|
err error
|
|
)
|
|
if token.Type == ArgString {
|
|
if numeric, err = strconv.Atoi(token.String); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
if numeric != numeric/2*2 {
|
|
return newStringFormulaArg("TRUE")
|
|
}
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// ISTEXT function tests if a supplied value is text, and if so, returns TRUE;
|
|
// Otherwise, the function returns FALSE. The syntax of the function is:
|
|
//
|
|
// ISTEXT(value)
|
|
//
|
|
func (fn *formulaFuncs) ISTEXT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ISTEXT requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
if token.ToNumber().Type != ArgError {
|
|
return newBoolFormulaArg(false)
|
|
}
|
|
return newBoolFormulaArg(token.Type == ArgString)
|
|
}
|
|
|
|
// NA function returns the Excel #N/A error. This error message has the
|
|
// meaning 'value not available' and is produced when an Excel Formula is
|
|
// unable to find a value that it needs. The syntax of the function is:
|
|
//
|
|
// NA()
|
|
//
|
|
func (fn *formulaFuncs) NA(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "NA accepts no arguments")
|
|
}
|
|
return newErrorFormulaArg(formulaErrorNA, formulaErrorNA)
|
|
}
|
|
|
|
// SHEET function returns the Sheet number for a specified reference. The
|
|
// syntax of the function is:
|
|
//
|
|
// SHEET()
|
|
//
|
|
func (fn *formulaFuncs) SHEET(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SHEET accepts no arguments")
|
|
}
|
|
return newNumberFormulaArg(float64(fn.f.GetSheetIndex(fn.sheet) + 1))
|
|
}
|
|
|
|
// Logical Functions
|
|
|
|
// AND function tests a number of supplied conditions and returns TRUE or
|
|
// FALSE. The syntax of the function is:
|
|
//
|
|
// AND(logical_test1,[logical_test2],...)
|
|
//
|
|
func (fn *formulaFuncs) AND(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "AND requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 30 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "AND accepts at most 30 arguments")
|
|
}
|
|
var (
|
|
and = true
|
|
val float64
|
|
err error
|
|
)
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgUnknown:
|
|
continue
|
|
case ArgString:
|
|
if token.String == "TRUE" {
|
|
continue
|
|
}
|
|
if token.String == "FALSE" {
|
|
return newStringFormulaArg(token.String)
|
|
}
|
|
if val, err = strconv.ParseFloat(token.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
and = and && (val != 0)
|
|
case ArgMatrix:
|
|
// TODO
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
}
|
|
return newBoolFormulaArg(and)
|
|
}
|
|
|
|
// FALSE function function returns the logical value FALSE. The syntax of the
|
|
// function is:
|
|
//
|
|
// FALSE()
|
|
//
|
|
func (fn *formulaFuncs) FALSE(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "FALSE takes no arguments")
|
|
}
|
|
return newBoolFormulaArg(false)
|
|
}
|
|
|
|
// IFERROR function receives two values (or expressions) and tests if the
|
|
// first of these evaluates to an error. The syntax of the function is:
|
|
//
|
|
// IFERROR(value,value_if_error)
|
|
//
|
|
func (fn *formulaFuncs) IFERROR(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "IFERROR requires 2 arguments")
|
|
}
|
|
value := argsList.Front().Value.(formulaArg)
|
|
if value.Type != ArgError {
|
|
if value.Type == ArgEmpty {
|
|
return newNumberFormulaArg(0)
|
|
}
|
|
return value
|
|
}
|
|
return argsList.Back().Value.(formulaArg)
|
|
}
|
|
|
|
// NOT function returns the opposite to a supplied logical value. The syntax
|
|
// of the function is:
|
|
//
|
|
// NOT(logical)
|
|
//
|
|
func (fn *formulaFuncs) NOT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "NOT requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString, ArgList:
|
|
if strings.ToUpper(token.String) == "TRUE" {
|
|
return newBoolFormulaArg(false)
|
|
}
|
|
if strings.ToUpper(token.String) == "FALSE" {
|
|
return newBoolFormulaArg(true)
|
|
}
|
|
case ArgNumber:
|
|
return newBoolFormulaArg(!(token.Number != 0))
|
|
case ArgError:
|
|
|
|
return token
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, "NOT expects 1 boolean or numeric argument")
|
|
}
|
|
|
|
// OR function tests a number of supplied conditions and returns either TRUE
|
|
// or FALSE. The syntax of the function is:
|
|
//
|
|
// OR(logical_test1,[logical_test2],...)
|
|
//
|
|
func (fn *formulaFuncs) OR(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "OR requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 30 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "OR accepts at most 30 arguments")
|
|
}
|
|
var (
|
|
or bool
|
|
val float64
|
|
err error
|
|
)
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgUnknown:
|
|
continue
|
|
case ArgString:
|
|
if token.String == "FALSE" {
|
|
continue
|
|
}
|
|
if token.String == "TRUE" {
|
|
or = true
|
|
continue
|
|
}
|
|
if val, err = strconv.ParseFloat(token.String, 64); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
or = val != 0
|
|
case ArgMatrix:
|
|
// TODO
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(strconv.FormatBool(or)))
|
|
}
|
|
|
|
// TRUE function returns the logical value TRUE. The syntax of the function
|
|
// is:
|
|
//
|
|
// TRUE()
|
|
//
|
|
func (fn *formulaFuncs) TRUE(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "TRUE takes no arguments")
|
|
}
|
|
return newBoolFormulaArg(true)
|
|
}
|
|
|
|
// Date and Time Functions
|
|
|
|
// DATE returns a date, from a user-supplied year, month and day. The syntax
|
|
// of the function is:
|
|
//
|
|
// DATE(year,month,day)
|
|
//
|
|
func (fn *formulaFuncs) DATE(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "DATE requires 3 number arguments")
|
|
}
|
|
var year, month, day int
|
|
var err error
|
|
if year, err = strconv.Atoi(argsList.Front().Value.(formulaArg).String); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "DATE requires 3 number arguments")
|
|
}
|
|
if month, err = strconv.Atoi(argsList.Front().Next().Value.(formulaArg).String); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "DATE requires 3 number arguments")
|
|
}
|
|
if day, err = strconv.Atoi(argsList.Back().Value.(formulaArg).String); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "DATE requires 3 number arguments")
|
|
}
|
|
d := makeDate(year, time.Month(month), day)
|
|
return newStringFormulaArg(timeFromExcelTime(daysBetween(excelMinTime1900.Unix(), d)+1, false).String())
|
|
}
|
|
|
|
// NOW function returns the current date and time. The function receives no arguments and therefore. The syntax of the function is:
|
|
//
|
|
// NOW()
|
|
//
|
|
func (fn *formulaFuncs) NOW(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "NOW accepts no arguments")
|
|
}
|
|
now := time.Now()
|
|
_, offset := now.Zone()
|
|
return newNumberFormulaArg(25569.0 + float64(now.Unix()+int64(offset))/86400)
|
|
}
|
|
|
|
// TODAY function returns the current date. The function has no arguments and
|
|
// therefore. The syntax of the function is:
|
|
//
|
|
// TODAY()
|
|
//
|
|
func (fn *formulaFuncs) TODAY(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "TODAY accepts no arguments")
|
|
}
|
|
now := time.Now()
|
|
_, offset := now.Zone()
|
|
return newNumberFormulaArg(daysBetween(excelMinTime1900.Unix(), now.Unix()+int64(offset)) + 1)
|
|
}
|
|
|
|
// makeDate return date as a Unix time, the number of seconds elapsed since
|
|
// January 1, 1970 UTC.
|
|
func makeDate(y int, m time.Month, d int) int64 {
|
|
if y == 1900 && int(m) <= 2 {
|
|
d--
|
|
}
|
|
date := time.Date(y, m, d, 0, 0, 0, 0, time.UTC)
|
|
return date.Unix()
|
|
}
|
|
|
|
// daysBetween return time interval of the given start timestamp and end
|
|
// timestamp.
|
|
func daysBetween(startDate, endDate int64) float64 {
|
|
return float64(int(0.5 + float64((endDate-startDate)/86400)))
|
|
}
|
|
|
|
// Text Functions
|
|
|
|
// CHAR function returns the character relating to a supplied character set
|
|
// number (from 1 to 255). syntax of the function is:
|
|
//
|
|
// CHAR(number)
|
|
//
|
|
func (fn *formulaFuncs) CHAR(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CHAR requires 1 argument")
|
|
}
|
|
arg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if arg.Type != ArgNumber {
|
|
return arg
|
|
}
|
|
num := int(arg.Number)
|
|
if num < 0 || num > 255 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
return newStringFormulaArg(fmt.Sprintf("%c", num))
|
|
}
|
|
|
|
// CLEAN removes all non-printable characters from a supplied text string. The
|
|
// syntax of the function is:
|
|
//
|
|
// CLEAN(text)
|
|
//
|
|
func (fn *formulaFuncs) CLEAN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CLEAN requires 1 argument")
|
|
}
|
|
b := bytes.Buffer{}
|
|
for _, c := range argsList.Front().Value.(formulaArg).String {
|
|
if c > 31 {
|
|
b.WriteRune(c)
|
|
}
|
|
}
|
|
return newStringFormulaArg(b.String())
|
|
}
|
|
|
|
// CODE function converts the first character of a supplied text string into
|
|
// the associated numeric character set code used by your computer. The
|
|
// syntax of the function is:
|
|
//
|
|
// CODE(text)
|
|
//
|
|
func (fn *formulaFuncs) CODE(argsList *list.List) formulaArg {
|
|
return fn.code("CODE", argsList)
|
|
}
|
|
|
|
// code is an implementation of the formula function CODE and UNICODE.
|
|
func (fn *formulaFuncs) code(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires 1 argument", name))
|
|
}
|
|
text := argsList.Front().Value.(formulaArg).Value()
|
|
if len(text) == 0 {
|
|
if name == "CODE" {
|
|
return newNumberFormulaArg(0)
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
return newNumberFormulaArg(float64(text[0]))
|
|
}
|
|
|
|
// CONCAT function joins together a series of supplied text strings into one
|
|
// combined text string.
|
|
//
|
|
// CONCAT(text1,[text2],...)
|
|
//
|
|
func (fn *formulaFuncs) CONCAT(argsList *list.List) formulaArg {
|
|
return fn.concat("CONCAT", argsList)
|
|
}
|
|
|
|
// CONCATENATE function joins together a series of supplied text strings into
|
|
// one combined text string.
|
|
//
|
|
// CONCATENATE(text1,[text2],...)
|
|
//
|
|
func (fn *formulaFuncs) CONCATENATE(argsList *list.List) formulaArg {
|
|
return fn.concat("CONCATENATE", argsList)
|
|
}
|
|
|
|
// concat is an implementation of the formula function CONCAT and CONCATENATE.
|
|
func (fn *formulaFuncs) concat(name string, argsList *list.List) formulaArg {
|
|
buf := bytes.Buffer{}
|
|
for arg := argsList.Front(); arg != nil; arg = arg.Next() {
|
|
token := arg.Value.(formulaArg)
|
|
switch token.Type {
|
|
case ArgString:
|
|
buf.WriteString(token.String)
|
|
case ArgNumber:
|
|
if token.Boolean {
|
|
if token.Number == 0 {
|
|
buf.WriteString("FALSE")
|
|
} else {
|
|
buf.WriteString("TRUE")
|
|
}
|
|
} else {
|
|
buf.WriteString(token.Value())
|
|
}
|
|
default:
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires arguments to be strings", name))
|
|
}
|
|
}
|
|
return newStringFormulaArg(buf.String())
|
|
}
|
|
|
|
// EXACT function tests if two supplied text strings or values are exactly
|
|
// equal and if so, returns TRUE; Otherwise, the function returns FALSE. The
|
|
// function is case-sensitive. The syntax of the function is:
|
|
//
|
|
// EXACT(text1,text2)
|
|
//
|
|
func (fn *formulaFuncs) EXACT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "EXACT requires 2 arguments")
|
|
}
|
|
text1 := argsList.Front().Value.(formulaArg).Value()
|
|
text2 := argsList.Back().Value.(formulaArg).Value()
|
|
return newBoolFormulaArg(text1 == text2)
|
|
}
|
|
|
|
// FIND function returns the position of a specified character or sub-string
|
|
// within a supplied text string. The function is case-sensitive. The syntax
|
|
// of the function is:
|
|
//
|
|
// FIND(find_text,within_text,[start_num])
|
|
//
|
|
func (fn *formulaFuncs) FIND(argsList *list.List) formulaArg {
|
|
return fn.find("FIND", argsList)
|
|
}
|
|
|
|
// FINDB counts each double-byte character as 2 when you have enabled the
|
|
// editing of a language that supports DBCS and then set it as the default
|
|
// language. Otherwise, FINDB counts each character as 1. The syntax of the
|
|
// function is:
|
|
//
|
|
// FINDB(find_text,within_text,[start_num])
|
|
//
|
|
func (fn *formulaFuncs) FINDB(argsList *list.List) formulaArg {
|
|
return fn.find("FINDB", argsList)
|
|
}
|
|
|
|
// find is an implementation of the formula function FIND and FINDB.
|
|
func (fn *formulaFuncs) find(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() < 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires at least 2 arguments", name))
|
|
}
|
|
if argsList.Len() > 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s allows at most 3 arguments", name))
|
|
}
|
|
findText := argsList.Front().Value.(formulaArg).Value()
|
|
withinText := argsList.Front().Next().Value.(formulaArg).Value()
|
|
startNum, result := 1, 1
|
|
if argsList.Len() == 3 {
|
|
numArg := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if numArg.Type != ArgNumber {
|
|
return numArg
|
|
}
|
|
if numArg.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
startNum = int(numArg.Number)
|
|
}
|
|
if findText == "" {
|
|
return newNumberFormulaArg(float64(startNum))
|
|
}
|
|
for idx := range withinText {
|
|
if result < startNum {
|
|
result++
|
|
}
|
|
if strings.Index(withinText[idx:], findText) == 0 {
|
|
return newNumberFormulaArg(float64(result))
|
|
}
|
|
result++
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
|
|
// LEFT function returns a specified number of characters from the start of a
|
|
// supplied text string. The syntax of the function is:
|
|
//
|
|
// LEFT(text,[num_chars])
|
|
//
|
|
func (fn *formulaFuncs) LEFT(argsList *list.List) formulaArg {
|
|
return fn.leftRight("LEFT", argsList)
|
|
}
|
|
|
|
// LEFTB returns the first character or characters in a text string, based on
|
|
// the number of bytes you specify. The syntax of the function is:
|
|
//
|
|
// LEFTB(text,[num_bytes])
|
|
//
|
|
func (fn *formulaFuncs) LEFTB(argsList *list.List) formulaArg {
|
|
return fn.leftRight("LEFTB", argsList)
|
|
}
|
|
|
|
// leftRight is an implementation of the formula function LEFT, LEFTB, RIGHT,
|
|
// RIGHTB. TODO: support DBCS include Japanese, Chinese (Simplified), Chinese
|
|
// (Traditional), and Korean.
|
|
func (fn *formulaFuncs) leftRight(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires at least 1 argument", name))
|
|
}
|
|
if argsList.Len() > 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s allows at most 2 arguments", name))
|
|
}
|
|
text, numChars := argsList.Front().Value.(formulaArg).Value(), 1
|
|
if argsList.Len() == 2 {
|
|
numArg := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if numArg.Type != ArgNumber {
|
|
return numArg
|
|
}
|
|
if numArg.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
numChars = int(numArg.Number)
|
|
}
|
|
if len(text) > numChars {
|
|
if name == "LEFT" || name == "LEFTB" {
|
|
return newStringFormulaArg(text[:numChars])
|
|
}
|
|
return newStringFormulaArg(text[len(text)-numChars:])
|
|
}
|
|
return newStringFormulaArg(text)
|
|
}
|
|
|
|
// LEN returns the length of a supplied text string. The syntax of the
|
|
// function is:
|
|
//
|
|
// LEN(text)
|
|
//
|
|
func (fn *formulaFuncs) LEN(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LEN requires 1 string argument")
|
|
}
|
|
return newStringFormulaArg(strconv.Itoa(len(argsList.Front().Value.(formulaArg).String)))
|
|
}
|
|
|
|
// LENB returns the number of bytes used to represent the characters in a text
|
|
// string. LENB counts 2 bytes per character only when a DBCS language is set
|
|
// as the default language. Otherwise LENB behaves the same as LEN, counting
|
|
// 1 byte per character. The syntax of the function is:
|
|
//
|
|
// LENB(text)
|
|
//
|
|
// TODO: the languages that support DBCS include Japanese, Chinese
|
|
// (Simplified), Chinese (Traditional), and Korean.
|
|
func (fn *formulaFuncs) LENB(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LENB requires 1 string argument")
|
|
}
|
|
return newStringFormulaArg(strconv.Itoa(len(argsList.Front().Value.(formulaArg).String)))
|
|
}
|
|
|
|
// LOWER converts all characters in a supplied text string to lower case. The
|
|
// syntax of the function is:
|
|
//
|
|
// LOWER(text)
|
|
//
|
|
func (fn *formulaFuncs) LOWER(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LOWER requires 1 argument")
|
|
}
|
|
return newStringFormulaArg(strings.ToLower(argsList.Front().Value.(formulaArg).String))
|
|
}
|
|
|
|
// MID function returns a specified number of characters from the middle of a
|
|
// supplied text string. The syntax of the function is:
|
|
//
|
|
// MID(text,start_num,num_chars)
|
|
//
|
|
func (fn *formulaFuncs) MID(argsList *list.List) formulaArg {
|
|
return fn.mid("MID", argsList)
|
|
}
|
|
|
|
// MIDB returns a specific number of characters from a text string, starting
|
|
// at the position you specify, based on the number of bytes you specify. The
|
|
// syntax of the function is:
|
|
//
|
|
// MID(text,start_num,num_chars)
|
|
//
|
|
func (fn *formulaFuncs) MIDB(argsList *list.List) formulaArg {
|
|
return fn.mid("MIDB", argsList)
|
|
}
|
|
|
|
// mid is an implementation of the formula function MID and MIDB. TODO:
|
|
// support DBCS include Japanese, Chinese (Simplified), Chinese
|
|
// (Traditional), and Korean.
|
|
func (fn *formulaFuncs) mid(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() != 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires 3 arguments", name))
|
|
}
|
|
text := argsList.Front().Value.(formulaArg).Value()
|
|
startNumArg, numCharsArg := argsList.Front().Next().Value.(formulaArg).ToNumber(), argsList.Front().Next().Next().Value.(formulaArg).ToNumber()
|
|
if startNumArg.Type != ArgNumber {
|
|
return startNumArg
|
|
}
|
|
if numCharsArg.Type != ArgNumber {
|
|
return numCharsArg
|
|
}
|
|
startNum := int(startNumArg.Number)
|
|
if startNum < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
textLen := len(text)
|
|
if startNum > textLen {
|
|
return newStringFormulaArg("")
|
|
}
|
|
startNum--
|
|
endNum := startNum + int(numCharsArg.Number)
|
|
if endNum > textLen+1 {
|
|
return newStringFormulaArg(text[startNum:])
|
|
}
|
|
return newStringFormulaArg(text[startNum:endNum])
|
|
}
|
|
|
|
// PROPER converts all characters in a supplied text string to proper case
|
|
// (i.e. all letters that do not immediately follow another letter are set to
|
|
// upper case and all other characters are lower case). The syntax of the
|
|
// function is:
|
|
//
|
|
// PROPER(text)
|
|
//
|
|
func (fn *formulaFuncs) PROPER(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "PROPER requires 1 argument")
|
|
}
|
|
buf := bytes.Buffer{}
|
|
isLetter := false
|
|
for _, char := range argsList.Front().Value.(formulaArg).String {
|
|
if !isLetter && unicode.IsLetter(char) {
|
|
buf.WriteRune(unicode.ToUpper(char))
|
|
} else {
|
|
buf.WriteRune(unicode.ToLower(char))
|
|
}
|
|
isLetter = unicode.IsLetter(char)
|
|
}
|
|
return newStringFormulaArg(buf.String())
|
|
}
|
|
|
|
// REPLACE function replaces all or part of a text string with another string.
|
|
// The syntax of the function is:
|
|
//
|
|
// REPLACE(old_text,start_num,num_chars,new_text)
|
|
//
|
|
func (fn *formulaFuncs) REPLACE(argsList *list.List) formulaArg {
|
|
return fn.replace("REPLACE", argsList)
|
|
}
|
|
|
|
// REPLACEB replaces part of a text string, based on the number of bytes you
|
|
// specify, with a different text string.
|
|
//
|
|
// REPLACEB(old_text,start_num,num_chars,new_text)
|
|
//
|
|
func (fn *formulaFuncs) REPLACEB(argsList *list.List) formulaArg {
|
|
return fn.replace("REPLACEB", argsList)
|
|
}
|
|
|
|
// replace is an implementation of the formula function REPLACE and REPLACEB.
|
|
// TODO: support DBCS include Japanese, Chinese (Simplified), Chinese
|
|
// (Traditional), and Korean.
|
|
func (fn *formulaFuncs) replace(name string, argsList *list.List) formulaArg {
|
|
if argsList.Len() != 4 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, fmt.Sprintf("%s requires 4 arguments", name))
|
|
}
|
|
oldText, newText := argsList.Front().Value.(formulaArg).Value(), argsList.Back().Value.(formulaArg).Value()
|
|
startNumArg, numCharsArg := argsList.Front().Next().Value.(formulaArg).ToNumber(), argsList.Front().Next().Next().Value.(formulaArg).ToNumber()
|
|
if startNumArg.Type != ArgNumber {
|
|
return startNumArg
|
|
}
|
|
if numCharsArg.Type != ArgNumber {
|
|
return numCharsArg
|
|
}
|
|
oldTextLen, startIdx := len(oldText), int(startNumArg.Number)
|
|
if startIdx > oldTextLen {
|
|
startIdx = oldTextLen + 1
|
|
}
|
|
endIdx := startIdx + int(numCharsArg.Number)
|
|
if endIdx > oldTextLen {
|
|
endIdx = oldTextLen + 1
|
|
}
|
|
if startIdx < 1 || endIdx < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
result := oldText[:startIdx-1] + newText + oldText[endIdx-1:]
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// REPT function returns a supplied text string, repeated a specified number
|
|
// of times. The syntax of the function is:
|
|
//
|
|
// REPT(text,number_times)
|
|
//
|
|
func (fn *formulaFuncs) REPT(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "REPT requires 2 arguments")
|
|
}
|
|
text := argsList.Front().Value.(formulaArg)
|
|
if text.Type != ArgString {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "REPT requires first argument to be a string")
|
|
}
|
|
times := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if times.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "REPT requires second argument to be a number")
|
|
}
|
|
if times.Number < 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "REPT requires second argument to be >= 0")
|
|
}
|
|
if times.Number == 0 {
|
|
return newStringFormulaArg("")
|
|
}
|
|
buf := bytes.Buffer{}
|
|
for i := 0; i < int(times.Number); i++ {
|
|
buf.WriteString(text.String)
|
|
}
|
|
return newStringFormulaArg(buf.String())
|
|
}
|
|
|
|
// RIGHT function returns a specified number of characters from the end of a
|
|
// supplied text string. The syntax of the function is:
|
|
//
|
|
// RIGHT(text,[num_chars])
|
|
//
|
|
func (fn *formulaFuncs) RIGHT(argsList *list.List) formulaArg {
|
|
return fn.leftRight("RIGHT", argsList)
|
|
}
|
|
|
|
// RIGHTB returns the last character or characters in a text string, based on
|
|
// the number of bytes you specify. The syntax of the function is:
|
|
//
|
|
// RIGHTB(text,[num_bytes])
|
|
//
|
|
func (fn *formulaFuncs) RIGHTB(argsList *list.List) formulaArg {
|
|
return fn.leftRight("RIGHTB", argsList)
|
|
}
|
|
|
|
// SUBSTITUTE function replaces one or more instances of a given text string,
|
|
// within an original text string. The syntax of the function is:
|
|
//
|
|
// SUBSTITUTE(text,old_text,new_text,[instance_num])
|
|
//
|
|
func (fn *formulaFuncs) SUBSTITUTE(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 3 && argsList.Len() != 4 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "SUBSTITUTE requires 3 or 4 arguments")
|
|
}
|
|
text, oldText := argsList.Front().Value.(formulaArg), argsList.Front().Next().Value.(formulaArg)
|
|
newText, instanceNum := argsList.Front().Next().Next().Value.(formulaArg), 0
|
|
if argsList.Len() == 3 {
|
|
return newStringFormulaArg(strings.Replace(text.Value(), oldText.Value(), newText.Value(), -1))
|
|
}
|
|
instanceNumArg := argsList.Back().Value.(formulaArg).ToNumber()
|
|
if instanceNumArg.Type != ArgNumber {
|
|
return instanceNumArg
|
|
}
|
|
instanceNum = int(instanceNumArg.Number)
|
|
if instanceNum < 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "instance_num should be > 0")
|
|
}
|
|
str, oldTextLen, count, chars, pos := text.Value(), len(oldText.Value()), instanceNum, 0, -1
|
|
for {
|
|
count--
|
|
index := strings.Index(str, oldText.Value())
|
|
if index == -1 {
|
|
pos = -1
|
|
break
|
|
} else {
|
|
pos = index + chars
|
|
if count == 0 {
|
|
break
|
|
}
|
|
idx := oldTextLen + index
|
|
chars += idx
|
|
str = str[idx:]
|
|
}
|
|
}
|
|
if pos == -1 {
|
|
return newStringFormulaArg(text.Value())
|
|
}
|
|
pre, post := text.Value()[:pos], text.Value()[pos+oldTextLen:]
|
|
return newStringFormulaArg(pre + newText.Value() + post)
|
|
}
|
|
|
|
// TRIM removes extra spaces (i.e. all spaces except for single spaces between
|
|
// words or characters) from a supplied text string. The syntax of the
|
|
// function is:
|
|
//
|
|
// TRIM(text)
|
|
//
|
|
func (fn *formulaFuncs) TRIM(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "TRIM requires 1 argument")
|
|
}
|
|
return newStringFormulaArg(strings.TrimSpace(argsList.Front().Value.(formulaArg).String))
|
|
}
|
|
|
|
// UNICHAR returns the Unicode character that is referenced by the given
|
|
// numeric value. The syntax of the function is:
|
|
//
|
|
// UNICHAR(number)
|
|
//
|
|
func (fn *formulaFuncs) UNICHAR(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "UNICHAR requires 1 argument")
|
|
}
|
|
numArg := argsList.Front().Value.(formulaArg).ToNumber()
|
|
if numArg.Type != ArgNumber {
|
|
return numArg
|
|
}
|
|
if numArg.Number <= 0 || numArg.Number > 55295 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, formulaErrorVALUE)
|
|
}
|
|
return newStringFormulaArg(string(rune(numArg.Number)))
|
|
}
|
|
|
|
// UNICODE function returns the code point for the first character of a
|
|
// supplied text string. The syntax of the function is:
|
|
//
|
|
// UNICODE(text)
|
|
//
|
|
func (fn *formulaFuncs) UNICODE(argsList *list.List) formulaArg {
|
|
return fn.code("UNICODE", argsList)
|
|
}
|
|
|
|
// UPPER converts all characters in a supplied text string to upper case. The
|
|
// syntax of the function is:
|
|
//
|
|
// UPPER(text)
|
|
//
|
|
func (fn *formulaFuncs) UPPER(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "UPPER requires 1 argument")
|
|
}
|
|
return newStringFormulaArg(strings.ToUpper(argsList.Front().Value.(formulaArg).String))
|
|
}
|
|
|
|
// Conditional Functions
|
|
|
|
// IF function tests a supplied condition and returns one result if the
|
|
// condition evaluates to TRUE, and another result if the condition evaluates
|
|
// to FALSE. The syntax of the function is:
|
|
//
|
|
// IF(logical_test,value_if_true,value_if_false)
|
|
//
|
|
func (fn *formulaFuncs) IF(argsList *list.List) formulaArg {
|
|
if argsList.Len() == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "IF requires at least 1 argument")
|
|
}
|
|
if argsList.Len() > 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "IF accepts at most 3 arguments")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg)
|
|
var (
|
|
cond bool
|
|
err error
|
|
result string
|
|
)
|
|
switch token.Type {
|
|
case ArgString:
|
|
if cond, err = strconv.ParseBool(token.String); err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, err.Error())
|
|
}
|
|
if argsList.Len() == 1 {
|
|
return newBoolFormulaArg(cond)
|
|
}
|
|
if cond {
|
|
return newStringFormulaArg(argsList.Front().Next().Value.(formulaArg).String)
|
|
}
|
|
if argsList.Len() == 3 {
|
|
result = argsList.Back().Value.(formulaArg).String
|
|
}
|
|
}
|
|
return newStringFormulaArg(result)
|
|
}
|
|
|
|
// Lookup and Reference Functions
|
|
|
|
// CHOOSE function returns a value from an array, that corresponds to a
|
|
// supplied index number (position). The syntax of the function is:
|
|
//
|
|
// CHOOSE(index_num,value1,[value2],...)
|
|
//
|
|
func (fn *formulaFuncs) CHOOSE(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CHOOSE requires 2 arguments")
|
|
}
|
|
idx, err := strconv.Atoi(argsList.Front().Value.(formulaArg).String)
|
|
if err != nil {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "CHOOSE requires first argument of type number")
|
|
}
|
|
if argsList.Len() <= idx {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "index_num should be <= to the number of values")
|
|
}
|
|
arg := argsList.Front()
|
|
for i := 0; i < idx; i++ {
|
|
arg = arg.Next()
|
|
}
|
|
var result formulaArg
|
|
switch arg.Value.(formulaArg).Type {
|
|
case ArgString:
|
|
result = newStringFormulaArg(arg.Value.(formulaArg).String)
|
|
case ArgMatrix:
|
|
result = newMatrixFormulaArg(arg.Value.(formulaArg).Matrix)
|
|
}
|
|
return result
|
|
}
|
|
|
|
// deepMatchRune finds whether the text deep matches/satisfies the pattern
|
|
// string.
|
|
func deepMatchRune(str, pattern []rune, simple bool) bool {
|
|
for len(pattern) > 0 {
|
|
switch pattern[0] {
|
|
default:
|
|
if len(str) == 0 || str[0] != pattern[0] {
|
|
return false
|
|
}
|
|
case '?':
|
|
if len(str) == 0 && !simple {
|
|
return false
|
|
}
|
|
case '*':
|
|
return deepMatchRune(str, pattern[1:], simple) ||
|
|
(len(str) > 0 && deepMatchRune(str[1:], pattern, simple))
|
|
}
|
|
str = str[1:]
|
|
pattern = pattern[1:]
|
|
}
|
|
return len(str) == 0 && len(pattern) == 0
|
|
}
|
|
|
|
// matchPattern finds whether the text matches or satisfies the pattern
|
|
// string. The pattern supports '*' and '?' wildcards in the pattern string.
|
|
func matchPattern(pattern, name string) (matched bool) {
|
|
if pattern == "" {
|
|
return name == pattern
|
|
}
|
|
if pattern == "*" {
|
|
return true
|
|
}
|
|
rname, rpattern := make([]rune, 0, len(name)), make([]rune, 0, len(pattern))
|
|
for _, r := range name {
|
|
rname = append(rname, r)
|
|
}
|
|
for _, r := range pattern {
|
|
rpattern = append(rpattern, r)
|
|
}
|
|
simple := false // Does extended wildcard '*' and '?' match.
|
|
return deepMatchRune(rname, rpattern, simple)
|
|
}
|
|
|
|
// compareFormulaArg compares the left-hand sides and the right-hand sides
|
|
// formula arguments by given conditions such as case sensitive, if exact
|
|
// match, and make compare result as formula criteria condition type.
|
|
func compareFormulaArg(lhs, rhs formulaArg, caseSensitive, exactMatch bool) byte {
|
|
if lhs.Type != rhs.Type {
|
|
return criteriaErr
|
|
}
|
|
switch lhs.Type {
|
|
case ArgNumber:
|
|
if lhs.Number == rhs.Number {
|
|
return criteriaEq
|
|
}
|
|
if lhs.Number < rhs.Number {
|
|
return criteriaL
|
|
}
|
|
return criteriaG
|
|
case ArgString:
|
|
ls, rs := lhs.String, rhs.String
|
|
if !caseSensitive {
|
|
ls, rs = strings.ToLower(ls), strings.ToLower(rs)
|
|
}
|
|
if exactMatch {
|
|
match := matchPattern(rs, ls)
|
|
if match {
|
|
return criteriaEq
|
|
}
|
|
return criteriaG
|
|
}
|
|
switch strings.Compare(ls, rs) {
|
|
case 1:
|
|
return criteriaG
|
|
case -1:
|
|
return criteriaL
|
|
case 0:
|
|
return criteriaEq
|
|
}
|
|
return criteriaErr
|
|
case ArgEmpty:
|
|
return criteriaEq
|
|
case ArgList:
|
|
return compareFormulaArgList(lhs, rhs, caseSensitive, exactMatch)
|
|
case ArgMatrix:
|
|
return compareFormulaArgMatrix(lhs, rhs, caseSensitive, exactMatch)
|
|
}
|
|
return criteriaErr
|
|
}
|
|
|
|
// compareFormulaArgList compares the left-hand sides and the right-hand sides
|
|
// list type formula arguments.
|
|
func compareFormulaArgList(lhs, rhs formulaArg, caseSensitive, exactMatch bool) byte {
|
|
if len(lhs.List) < len(rhs.List) {
|
|
return criteriaL
|
|
}
|
|
if len(lhs.List) > len(rhs.List) {
|
|
return criteriaG
|
|
}
|
|
for arg := range lhs.List {
|
|
criteria := compareFormulaArg(lhs.List[arg], rhs.List[arg], caseSensitive, exactMatch)
|
|
if criteria != criteriaEq {
|
|
return criteria
|
|
}
|
|
}
|
|
return criteriaEq
|
|
}
|
|
|
|
// compareFormulaArgMatrix compares the left-hand sides and the right-hand sides
|
|
// matrix type formula arguments.
|
|
func compareFormulaArgMatrix(lhs, rhs formulaArg, caseSensitive, exactMatch bool) byte {
|
|
if len(lhs.Matrix) < len(rhs.Matrix) {
|
|
return criteriaL
|
|
}
|
|
if len(lhs.Matrix) > len(rhs.Matrix) {
|
|
return criteriaG
|
|
}
|
|
for i := range lhs.Matrix {
|
|
left := lhs.Matrix[i]
|
|
right := lhs.Matrix[i]
|
|
if len(left) < len(right) {
|
|
return criteriaL
|
|
}
|
|
if len(left) > len(right) {
|
|
return criteriaG
|
|
}
|
|
for arg := range left {
|
|
criteria := compareFormulaArg(left[arg], right[arg], caseSensitive, exactMatch)
|
|
if criteria != criteriaEq {
|
|
return criteria
|
|
}
|
|
}
|
|
}
|
|
return criteriaEq
|
|
}
|
|
|
|
// COLUMN function returns the first column number within a supplied reference
|
|
// or the number of the current column. The syntax of the function is:
|
|
//
|
|
// COLUMN([reference])
|
|
//
|
|
func (fn *formulaFuncs) COLUMN(argsList *list.List) formulaArg {
|
|
if argsList.Len() > 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COLUMN requires at most 1 argument")
|
|
}
|
|
if argsList.Len() == 1 {
|
|
if argsList.Front().Value.(formulaArg).cellRanges != nil && argsList.Front().Value.(formulaArg).cellRanges.Len() > 0 {
|
|
return newNumberFormulaArg(float64(argsList.Front().Value.(formulaArg).cellRanges.Front().Value.(cellRange).From.Col))
|
|
}
|
|
if argsList.Front().Value.(formulaArg).cellRefs != nil && argsList.Front().Value.(formulaArg).cellRefs.Len() > 0 {
|
|
return newNumberFormulaArg(float64(argsList.Front().Value.(formulaArg).cellRefs.Front().Value.(cellRef).Col))
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, "invalid reference")
|
|
}
|
|
col, _, _ := CellNameToCoordinates(fn.cell)
|
|
return newNumberFormulaArg(float64(col))
|
|
}
|
|
|
|
// COLUMNS function receives an Excel range and returns the number of columns
|
|
// that are contained within the range. The syntax of the function is:
|
|
//
|
|
// COLUMNS(array)
|
|
//
|
|
func (fn *formulaFuncs) COLUMNS(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "COLUMNS requires 1 argument")
|
|
}
|
|
var min, max int
|
|
if argsList.Front().Value.(formulaArg).cellRanges != nil && argsList.Front().Value.(formulaArg).cellRanges.Len() > 0 {
|
|
crs := argsList.Front().Value.(formulaArg).cellRanges
|
|
for cr := crs.Front(); cr != nil; cr = cr.Next() {
|
|
if min == 0 {
|
|
min = cr.Value.(cellRange).From.Col
|
|
}
|
|
if min > cr.Value.(cellRange).From.Col {
|
|
min = cr.Value.(cellRange).From.Col
|
|
}
|
|
if min > cr.Value.(cellRange).To.Col {
|
|
min = cr.Value.(cellRange).To.Col
|
|
}
|
|
if max < cr.Value.(cellRange).To.Col {
|
|
max = cr.Value.(cellRange).To.Col
|
|
}
|
|
if max < cr.Value.(cellRange).From.Col {
|
|
max = cr.Value.(cellRange).From.Col
|
|
}
|
|
}
|
|
}
|
|
if argsList.Front().Value.(formulaArg).cellRefs != nil && argsList.Front().Value.(formulaArg).cellRefs.Len() > 0 {
|
|
cr := argsList.Front().Value.(formulaArg).cellRefs
|
|
for refs := cr.Front(); refs != nil; refs = refs.Next() {
|
|
if min == 0 {
|
|
min = refs.Value.(cellRef).Col
|
|
}
|
|
if min > refs.Value.(cellRef).Col {
|
|
min = refs.Value.(cellRef).Col
|
|
}
|
|
if max < refs.Value.(cellRef).Col {
|
|
max = refs.Value.(cellRef).Col
|
|
}
|
|
}
|
|
}
|
|
if max == TotalColumns {
|
|
return newNumberFormulaArg(float64(TotalColumns))
|
|
}
|
|
result := max - min + 1
|
|
if max == min {
|
|
if min == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "invalid reference")
|
|
}
|
|
return newNumberFormulaArg(float64(1))
|
|
}
|
|
return newNumberFormulaArg(float64(result))
|
|
}
|
|
|
|
// HLOOKUP function 'looks up' a given value in the top row of a data array
|
|
// (or table), and returns the corresponding value from another row of the
|
|
// array. The syntax of the function is:
|
|
//
|
|
// HLOOKUP(lookup_value,table_array,row_index_num,[range_lookup])
|
|
//
|
|
func (fn *formulaFuncs) HLOOKUP(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HLOOKUP requires at least 3 arguments")
|
|
}
|
|
if argsList.Len() > 4 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HLOOKUP requires at most 4 arguments")
|
|
}
|
|
lookupValue := argsList.Front().Value.(formulaArg)
|
|
tableArray := argsList.Front().Next().Value.(formulaArg)
|
|
if tableArray.Type != ArgMatrix {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HLOOKUP requires second argument of table array")
|
|
}
|
|
rowArg := argsList.Front().Next().Next().Value.(formulaArg).ToNumber()
|
|
if rowArg.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "HLOOKUP requires numeric row argument")
|
|
}
|
|
rowIdx, matchIdx, wasExact, exactMatch := int(rowArg.Number)-1, -1, false, false
|
|
if argsList.Len() == 4 {
|
|
rangeLookup := argsList.Back().Value.(formulaArg).ToBool()
|
|
if rangeLookup.Type == ArgError {
|
|
return newErrorFormulaArg(formulaErrorVALUE, rangeLookup.Error)
|
|
}
|
|
if rangeLookup.Number == 0 {
|
|
exactMatch = true
|
|
}
|
|
}
|
|
row := tableArray.Matrix[0]
|
|
if exactMatch || len(tableArray.Matrix) == TotalRows {
|
|
start:
|
|
for idx, mtx := range row {
|
|
lhs := mtx
|
|
switch lookupValue.Type {
|
|
case ArgNumber:
|
|
if !lookupValue.Boolean {
|
|
lhs = mtx.ToNumber()
|
|
if lhs.Type == ArgError {
|
|
lhs = mtx
|
|
}
|
|
}
|
|
case ArgMatrix:
|
|
lhs = tableArray
|
|
}
|
|
if compareFormulaArg(lhs, lookupValue, false, exactMatch) == criteriaEq {
|
|
matchIdx = idx
|
|
wasExact = true
|
|
break start
|
|
}
|
|
}
|
|
} else {
|
|
matchIdx, wasExact = hlookupBinarySearch(row, lookupValue)
|
|
}
|
|
if matchIdx == -1 {
|
|
return newErrorFormulaArg(formulaErrorNA, "HLOOKUP no result found")
|
|
}
|
|
if rowIdx < 0 || rowIdx >= len(tableArray.Matrix) {
|
|
return newErrorFormulaArg(formulaErrorNA, "HLOOKUP has invalid row index")
|
|
}
|
|
row = tableArray.Matrix[rowIdx]
|
|
if wasExact || !exactMatch {
|
|
return row[matchIdx]
|
|
}
|
|
return newErrorFormulaArg(formulaErrorNA, "HLOOKUP no result found")
|
|
}
|
|
|
|
// VLOOKUP function 'looks up' a given value in the left-hand column of a
|
|
// data array (or table), and returns the corresponding value from another
|
|
// column of the array. The syntax of the function is:
|
|
//
|
|
// VLOOKUP(lookup_value,table_array,col_index_num,[range_lookup])
|
|
//
|
|
func (fn *formulaFuncs) VLOOKUP(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "VLOOKUP requires at least 3 arguments")
|
|
}
|
|
if argsList.Len() > 4 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "VLOOKUP requires at most 4 arguments")
|
|
}
|
|
lookupValue := argsList.Front().Value.(formulaArg)
|
|
tableArray := argsList.Front().Next().Value.(formulaArg)
|
|
if tableArray.Type != ArgMatrix {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "VLOOKUP requires second argument of table array")
|
|
}
|
|
colIdx := argsList.Front().Next().Next().Value.(formulaArg).ToNumber()
|
|
if colIdx.Type != ArgNumber {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "VLOOKUP requires numeric col argument")
|
|
}
|
|
col, matchIdx, wasExact, exactMatch := int(colIdx.Number)-1, -1, false, false
|
|
if argsList.Len() == 4 {
|
|
rangeLookup := argsList.Back().Value.(formulaArg).ToBool()
|
|
if rangeLookup.Type == ArgError {
|
|
return newErrorFormulaArg(formulaErrorVALUE, rangeLookup.Error)
|
|
}
|
|
if rangeLookup.Number == 0 {
|
|
exactMatch = true
|
|
}
|
|
}
|
|
if exactMatch || len(tableArray.Matrix) == TotalRows {
|
|
start:
|
|
for idx, mtx := range tableArray.Matrix {
|
|
lhs := mtx[0]
|
|
switch lookupValue.Type {
|
|
case ArgNumber:
|
|
if !lookupValue.Boolean {
|
|
lhs = mtx[0].ToNumber()
|
|
if lhs.Type == ArgError {
|
|
lhs = mtx[0]
|
|
}
|
|
}
|
|
case ArgMatrix:
|
|
lhs = tableArray
|
|
}
|
|
if compareFormulaArg(lhs, lookupValue, false, exactMatch) == criteriaEq {
|
|
matchIdx = idx
|
|
wasExact = true
|
|
break start
|
|
}
|
|
}
|
|
} else {
|
|
matchIdx, wasExact = vlookupBinarySearch(tableArray, lookupValue)
|
|
}
|
|
if matchIdx == -1 {
|
|
return newErrorFormulaArg(formulaErrorNA, "VLOOKUP no result found")
|
|
}
|
|
mtx := tableArray.Matrix[matchIdx]
|
|
if col < 0 || col >= len(mtx) {
|
|
return newErrorFormulaArg(formulaErrorNA, "VLOOKUP has invalid column index")
|
|
}
|
|
if wasExact || !exactMatch {
|
|
return mtx[col]
|
|
}
|
|
return newErrorFormulaArg(formulaErrorNA, "VLOOKUP no result found")
|
|
}
|
|
|
|
// vlookupBinarySearch finds the position of a target value when range lookup
|
|
// is TRUE, if the data of table array can't guarantee be sorted, it will
|
|
// return wrong result.
|
|
func vlookupBinarySearch(tableArray, lookupValue formulaArg) (matchIdx int, wasExact bool) {
|
|
var low, high, lastMatchIdx int = 0, len(tableArray.Matrix) - 1, -1
|
|
for low <= high {
|
|
var mid int = low + (high-low)/2
|
|
mtx := tableArray.Matrix[mid]
|
|
lhs := mtx[0]
|
|
switch lookupValue.Type {
|
|
case ArgNumber:
|
|
if !lookupValue.Boolean {
|
|
lhs = mtx[0].ToNumber()
|
|
if lhs.Type == ArgError {
|
|
lhs = mtx[0]
|
|
}
|
|
}
|
|
case ArgMatrix:
|
|
lhs = tableArray
|
|
}
|
|
result := compareFormulaArg(lhs, lookupValue, false, false)
|
|
if result == criteriaEq {
|
|
matchIdx, wasExact = mid, true
|
|
return
|
|
} else if result == criteriaG {
|
|
high = mid - 1
|
|
} else if result == criteriaL {
|
|
matchIdx, low = mid, mid+1
|
|
if lhs.Value() != "" {
|
|
lastMatchIdx = matchIdx
|
|
}
|
|
} else {
|
|
return -1, false
|
|
}
|
|
}
|
|
matchIdx, wasExact = lastMatchIdx, true
|
|
return
|
|
}
|
|
|
|
// vlookupBinarySearch finds the position of a target value when range lookup
|
|
// is TRUE, if the data of table array can't guarantee be sorted, it will
|
|
// return wrong result.
|
|
func hlookupBinarySearch(row []formulaArg, lookupValue formulaArg) (matchIdx int, wasExact bool) {
|
|
var low, high, lastMatchIdx int = 0, len(row) - 1, -1
|
|
for low <= high {
|
|
var mid int = low + (high-low)/2
|
|
mtx := row[mid]
|
|
result := compareFormulaArg(mtx, lookupValue, false, false)
|
|
if result == criteriaEq {
|
|
matchIdx, wasExact = mid, true
|
|
return
|
|
} else if result == criteriaG {
|
|
high = mid - 1
|
|
} else if result == criteriaL {
|
|
low, lastMatchIdx = mid+1, mid
|
|
} else {
|
|
return -1, false
|
|
}
|
|
}
|
|
matchIdx, wasExact = lastMatchIdx, true
|
|
return
|
|
}
|
|
|
|
// LOOKUP function performs an approximate match lookup in a one-column or
|
|
// one-row range, and returns the corresponding value from another one-column
|
|
// or one-row range. The syntax of the function is:
|
|
//
|
|
// LOOKUP(lookup_value,lookup_vector,[result_vector])
|
|
//
|
|
func (fn *formulaFuncs) LOOKUP(argsList *list.List) formulaArg {
|
|
if argsList.Len() < 2 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LOOKUP requires at least 2 arguments")
|
|
}
|
|
if argsList.Len() > 3 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LOOKUP requires at most 3 arguments")
|
|
}
|
|
lookupValue := argsList.Front().Value.(formulaArg)
|
|
lookupVector := argsList.Front().Next().Value.(formulaArg)
|
|
if lookupVector.Type != ArgMatrix && lookupVector.Type != ArgList {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "LOOKUP requires second argument of table array")
|
|
}
|
|
cols, matchIdx := lookupCol(lookupVector), -1
|
|
for idx, col := range cols {
|
|
lhs := lookupValue
|
|
switch col.Type {
|
|
case ArgNumber:
|
|
lhs = lhs.ToNumber()
|
|
if !col.Boolean {
|
|
if lhs.Type == ArgError {
|
|
lhs = lookupValue
|
|
}
|
|
}
|
|
}
|
|
if compareFormulaArg(lhs, col, false, false) == criteriaEq {
|
|
matchIdx = idx
|
|
break
|
|
}
|
|
}
|
|
column := cols
|
|
if argsList.Len() == 3 {
|
|
column = lookupCol(argsList.Back().Value.(formulaArg))
|
|
}
|
|
if matchIdx < 0 || matchIdx >= len(column) {
|
|
return newErrorFormulaArg(formulaErrorNA, "LOOKUP no result found")
|
|
}
|
|
return column[matchIdx]
|
|
}
|
|
|
|
// lookupCol extract columns for LOOKUP.
|
|
func lookupCol(arr formulaArg) []formulaArg {
|
|
col := arr.List
|
|
if arr.Type == ArgMatrix {
|
|
col = nil
|
|
for _, r := range arr.Matrix {
|
|
if len(r) > 0 {
|
|
col = append(col, r[0])
|
|
continue
|
|
}
|
|
col = append(col, newEmptyFormulaArg())
|
|
}
|
|
}
|
|
return col
|
|
}
|
|
|
|
// ROW function returns the first row number within a supplied reference or
|
|
// the number of the current row. The syntax of the function is:
|
|
//
|
|
// ROW([reference])
|
|
//
|
|
func (fn *formulaFuncs) ROW(argsList *list.List) formulaArg {
|
|
if argsList.Len() > 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ROW requires at most 1 argument")
|
|
}
|
|
if argsList.Len() == 1 {
|
|
if argsList.Front().Value.(formulaArg).cellRanges != nil && argsList.Front().Value.(formulaArg).cellRanges.Len() > 0 {
|
|
return newNumberFormulaArg(float64(argsList.Front().Value.(formulaArg).cellRanges.Front().Value.(cellRange).From.Row))
|
|
}
|
|
if argsList.Front().Value.(formulaArg).cellRefs != nil && argsList.Front().Value.(formulaArg).cellRefs.Len() > 0 {
|
|
return newNumberFormulaArg(float64(argsList.Front().Value.(formulaArg).cellRefs.Front().Value.(cellRef).Row))
|
|
}
|
|
return newErrorFormulaArg(formulaErrorVALUE, "invalid reference")
|
|
}
|
|
_, row, _ := CellNameToCoordinates(fn.cell)
|
|
return newNumberFormulaArg(float64(row))
|
|
}
|
|
|
|
// ROWS function takes an Excel range and returns the number of rows that are
|
|
// contained within the range. The syntax of the function is:
|
|
//
|
|
// ROWS(array)
|
|
//
|
|
func (fn *formulaFuncs) ROWS(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ROWS requires 1 argument")
|
|
}
|
|
var min, max int
|
|
if argsList.Front().Value.(formulaArg).cellRanges != nil && argsList.Front().Value.(formulaArg).cellRanges.Len() > 0 {
|
|
crs := argsList.Front().Value.(formulaArg).cellRanges
|
|
for cr := crs.Front(); cr != nil; cr = cr.Next() {
|
|
if min == 0 {
|
|
min = cr.Value.(cellRange).From.Row
|
|
}
|
|
if min > cr.Value.(cellRange).From.Row {
|
|
min = cr.Value.(cellRange).From.Row
|
|
}
|
|
if min > cr.Value.(cellRange).To.Row {
|
|
min = cr.Value.(cellRange).To.Row
|
|
}
|
|
if max < cr.Value.(cellRange).To.Row {
|
|
max = cr.Value.(cellRange).To.Row
|
|
}
|
|
if max < cr.Value.(cellRange).From.Row {
|
|
max = cr.Value.(cellRange).From.Row
|
|
}
|
|
}
|
|
}
|
|
if argsList.Front().Value.(formulaArg).cellRefs != nil && argsList.Front().Value.(formulaArg).cellRefs.Len() > 0 {
|
|
cr := argsList.Front().Value.(formulaArg).cellRefs
|
|
for refs := cr.Front(); refs != nil; refs = refs.Next() {
|
|
if min == 0 {
|
|
min = refs.Value.(cellRef).Row
|
|
}
|
|
if min > refs.Value.(cellRef).Row {
|
|
min = refs.Value.(cellRef).Row
|
|
}
|
|
if max < refs.Value.(cellRef).Row {
|
|
max = refs.Value.(cellRef).Row
|
|
}
|
|
}
|
|
}
|
|
if max == TotalRows {
|
|
return newStringFormulaArg(strconv.Itoa(TotalRows))
|
|
}
|
|
result := max - min + 1
|
|
if max == min {
|
|
if min == 0 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "invalid reference")
|
|
}
|
|
return newNumberFormulaArg(float64(1))
|
|
}
|
|
return newStringFormulaArg(strconv.Itoa(result))
|
|
}
|
|
|
|
// Web Functions
|
|
|
|
// ENCODEURL function returns a URL-encoded string, replacing certain
|
|
// non-alphanumeric characters with the percentage symbol (%) and a
|
|
// hexadecimal number. The syntax of the function is:
|
|
//
|
|
// ENCODEURL(url)
|
|
//
|
|
func (fn *formulaFuncs) ENCODEURL(argsList *list.List) formulaArg {
|
|
if argsList.Len() != 1 {
|
|
return newErrorFormulaArg(formulaErrorVALUE, "ENCODEURL requires 1 argument")
|
|
}
|
|
token := argsList.Front().Value.(formulaArg).Value()
|
|
return newStringFormulaArg(strings.Replace(url.QueryEscape(token), "+", "%20", -1))
|
|
}
|