koffice/kspread/formula.cc

/* This file is part of the KDE project
   Copyright (C) 2003,2004 Ariya Hidayat <ariya@kde.org>
   Copyright (C) 2005 Tomas Mecir <mecirt@gmail.com>

   This library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public
   License as published by the Free Software Foundation; either
   version 2 of the License.

   This library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Library General Public License for more details.

   You should have received a copy of the GNU Library General Public License
   along with this library; see the file COPYING.LIB.  If not, write to
   the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
 * Boston, MA 02110-1301, USA.
*/

#include "formula.h"

#include "kspread_cell.h"
#include "kspread_sheet.h"
#include "kspread_doc.h"
#include "kspread_util.h"
#include "kspread_value.h"

#include "valuecalc.h"
#include "valueconverter.h"
#include "valueparser.h"

#include "functions.h"

#include <limits.h>

#include <tqregexp.h>
#include <tqstring.h>
#include <tqvaluevector.h>
#include <tqvaluestack.h>

#include <klocale.h>

/*
  To understand how this formula engine works, please refer to the documentation
  in file DESIGN.html.

  Useful references:
   - "Principles of Compiler Design", A.V.Aho, J.D.Ullman, Addison Wesley, 1978
   - "Writing Interactive Compilers and Interpreters", P.J. Brown,
     John Wiley and Sons, 1979.
   - "The Theory and Practice of Compiler Writing", J.Tremblay, P.G.Sorenson,
     McGraw-Hill, 1985.
   - "The Java(TM) Virtual Machine Specification", T.Lindholm, F.Yellin,
     Addison-Wesley, 1997.
   - "Java Virtual Machine", J.Meyer, T.Downing, O'Reilly, 1997.

 */


 /*
TODO - features:
 - handle Intersection
 - cell reference is made relative (absolute now)
 - shared formula (different owner, same data)
 - relative internal representation (independent of owner)
 - OASIS support
TODO - optimizations:
 - handle initial formula marker = (and +)
 - reuse constant already in the pool
 - reuse references already in the pool
 - expression optimization (e.g. 1+2+A1 becomes 3+A1)
 */

namespace KSpread
{

class Opcode
{
public:

  enum { Nop = 0, Load, Ref, Cell, Range, Function, Add, Sub, Neg, Mul, Div,
    Pow, Concat, Not, Equal, Less, Greater };

  unsigned type;
  unsigned index;

  Opcode(): type(Nop), index(0) {};
  Opcode( unsigned t ): type(t), index(0) {};
  Opcode( unsigned t, unsigned i ): type(t), index(i) {};
};

class Formula::Private
{
public:
  Formula *formula;
  Cell *cell;
  Sheet *sheet;
  bool dirty;
  bool valid;
  TQString expression;
  TQValueVector<Opcode> codes;
  TQValueVector<Value> constants;
};

class TokenStack : public TQValueVector<Token>
{
public:
  TokenStack();
  bool isEmpty() const;
  unsigned itemCount() const;
  void push( const Token& token );
  Token pop();
  const Token& top();
  const Token& top( unsigned index );
private:
  void ensureSpace();
  unsigned topIndex;
};

}

using namespace KSpread;

// for null token
const Token Token::null;

// helper function: return operator of given token text
// e.g. "*" yields Operator::Asterisk, and so on
Token::Op KSpread::matchOperator( const TQString& text )
{
  Token::Op result = Token::InvalidOp;

  if( text.length() == 1 )
  {
    TQChar p = text[0];
    switch( p.tqunicode() )
    {
        case '+': result = Token::Plus; break;
        case '-': result = Token::Minus; break;
        case '*': result = Token::Asterisk; break;
        case '/': result = Token::Slash; break;
        case '^': result = Token::Caret; break;
        case ',': result = Token::Comma; break;
        case ';': result = Token::Semicolon; break;
        case '(': result = Token::LeftPar; break;
        case ')': result = Token::RightPar; break;
        case '&': result = Token::Ampersand; break;
        case '=': result = Token::Equal; break;
        case '<': result = Token::Less; break;
        case '>': result = Token::Greater; break;
        case '%': result = Token::Percent; break;
        default : result = Token::InvalidOp; break;
    }
  }

  if( text.length() == 2 )
  {
    if( text == "<>" ) result = Token::NotEqual;
    if( text == "<=" ) result = Token::LessEqual;
    if( text == ">=" ) result = Token::GreaterEqual;
    if( text == "==" ) result = Token::Equal;
  }

  return result;
}

// helper function: give operator precedence
// e.g. "+" is 1 while "*" is 3
static int opPrecedence( Token::Op op )
{
  int prec = -1;
  switch( op )
  {
    case Token::Percent      : prec = 8; break;
    case Token::Caret        : prec = 7; break;
    case Token::Asterisk     : prec = 5; break;
    case Token::Slash        : prec = 6; break;
    case Token::Plus         : prec = 3; break;
    case Token::Minus        : prec = 3; break;
    case Token::Ampersand    : prec = 2; break;
    case Token::Equal        : prec = 1; break;
    case Token::NotEqual     : prec = 1; break;
    case Token::Less         : prec = 1; break;
    case Token::Greater      : prec = 1; break;
    case Token::LessEqual    : prec = 1; break;
    case Token::GreaterEqual : prec = 1; break;
    case Token::Semicolon    : prec = 0; break;
    case Token::RightPar     : prec = 0; break;
    case Token::LeftPar      : prec = -1; break;
    default: prec = -1; break;
  }
  return prec;
}

// helper function
static Value tokenAsValue( const Token& token )
{
  Value value;
  if( token.isBoolean() ) value = Value( token.asBoolean() );
  else if( token.isInteger() ) value = Value( token.asInteger() );
  else if( token.isFloat() ) value = Value( token.asFloat() );
  else if( token.isString() ) value = Value( token.asString() );
  return value;
}

/**********************
    Token
 **********************/

// creates a token
Token::Token( Type type, const TQString& text, int pos )
{
  m_type = type;
  m_text = text;
  m_pos = pos;
}

// copy constructor
Token::Token( const Token& token )
{
  m_type = token.m_type;
  m_text = token.m_text;
  m_pos = token.m_pos;
}

// assignment operator
Token& Token::operator=( const Token& token )
{
  m_type = token.m_type;
  m_text = token.m_text;
  m_pos = token.m_pos;
  return *this;
}

bool Token::asBoolean() const
{
  if( !isBoolean() ) return false;
  return m_text.lower() == "true";
  // FIXME check also for i18n version
}

long Token::asInteger() const
{
  if( isInteger() ) return m_text.toLong();
  else return 0;
}

double Token::asFloat() const
{
  if( isFloat() ) return m_text.toDouble();
  else return 0.0;
}

TQString Token::asString() const
{
  if( isString() ) return m_text.mid( 1, m_text.length()-2 );
  else return TQString();
}

Token::Op Token::asOperator() const
{
  if( isOperator() ) return matchOperator( m_text );
  else return InvalidOp;
}

TQString Token::sheetName() const
{
  if( !isCell() && !isRange() ) return TQString();
  int i = m_text.tqfind( '!' );
  if( i < 0 ) return TQString();
  TQString sheet = m_text.left( i );
  return sheet;
}

TQString Token::description() const
{
  TQString desc;

  switch (m_type )
  {
    case  Boolean:    desc = "Boolean"; break;
    case  Integer:    desc = "Integer"; break;
    case  Float:      desc = "Float"; break;
    case  String:     desc = "String"; break;
    case  Identifier: desc = "Identifier"; break;
    case  Cell:       desc = "Cell"; break;
    case  Range:      desc = "Range"; break;
    case  Operator:   desc = "Operator"; break;
    default:          desc = "Unknown"; break;
  }

  while( desc.length() < 10 ) desc.prepend( ' ' );
  desc.prepend( "  " );
  desc.prepend( TQString::number( m_pos ) );
  desc.append( " : " ).append( m_text );

  return desc;
}


/**********************
    TokenStack
 **********************/

TokenStack::TokenStack(): TQValueVector<Token>()
{
  topIndex = 0;
  ensureSpace();
}

bool TokenStack::isEmpty() const
{
  return topIndex == 0;
}

unsigned TokenStack::itemCount() const
{
  return topIndex;
}

void TokenStack::push( const Token& token )
{
  ensureSpace();
  at( topIndex++ ) = token;
}

Token TokenStack::pop()
{
  return (topIndex > 0 ) ? Token( at( --topIndex ) ) : Token();
}

const Token& TokenStack::top()
{
  return top( 0 );
}

const Token& TokenStack::top( unsigned index )
{
  if( topIndex > index )
    return at( topIndex-index-1 );
  return Token::null;
}

void TokenStack::ensureSpace()
{
  while( topIndex >= size() )
    resize( size() + 10 );
}

/**********************
    FormulaPrivate
 **********************/

// helper function: return true for valid identifier character
bool KSpread::isIdentifier( TQChar ch )
{
  return ( ch.tqunicode() == '_' ) || (ch.tqunicode() == '$' ) || ( ch.isLetter() );
}




/**********************
    Formula
 **********************/

// Constructor

Formula::Formula (Sheet *sheet, Cell *cell)
{
  d = new Private;
  d->cell = cell;
  d->sheet = sheet;
  clear();
}

Formula::Formula()
{
  d = new Private;
  d->cell = 0;
  d->sheet = 0;
  clear();
}

// Destructor

Formula::~Formula()
{
  delete d;
}

Cell* Formula::cell() const
{
  return d->cell;
}

Sheet* Formula::sheet() const
{
  return d->sheet;
}

// Sets a new expression for this formula.
// note that both the real lex and parse processes will happen later on
// when needed (i.e. "lazy parse"), for example during formula evaluation.

void Formula::setExpression( const TQString& expr )
{
  d->expression = expr;
  d->dirty = true;
  d->valid = false;
}

// Returns the expression associated with this formula.

TQString Formula::expression() const
{
  return d->expression;
}

// Returns the validity of the formula.
// note: empty formula is always invalid.

bool Formula::isValid() const
{
  if( d->dirty )
  {
    KLocale* locale = d->cell ? d->cell->locale() : 0;
    if ((!locale) && d->sheet)
      locale = d->sheet->doc()->locale();
    Tokens tokens = scan( d->expression, locale );
    if( tokens.valid() )
      compile( tokens );
    else
      d->valid = false;
  }
  return d->valid;
}

// Clears everything, also mark the formula as invalid.

void Formula::clear()
{
  d->expression = TQString();
  d->dirty = true;
  d->valid = false;
  d->constants.clear();
  d->codes.clear();
}

// Returns list of token for the expression.
// this triggers again the lexical analysis step. it is however preferable
// (even when there's small performance penalty) because otherwise we need to
// store parsed tokens all the time which serves no good purpose.

Tokens Formula::tokens() const
{
  KLocale* locale = d->cell ? d->cell->locale() : 0;
  if ((!locale) && d->sheet)
    locale = d->sheet->doc()->locale();
  return scan( d->expression, locale );
}

Tokens Formula::scan( const TQString& expr, KLocale* locale ) const
{
  // to hold the result
  Tokens tokens;

  // parsing state
  enum { Start, Finish, Bad, InNumber, InDecimal, InExpIndicator, InExponent,
    InString, InIdentifier, InCell, InRange, InSheetOrAreaName } state;

  // use locale settings if specified
  TQString thousand = locale ? locale->thousandsSeparator() : "";
  TQString decimal = locale ? locale->decimalSymbol() : ".";

  // initialize variables
  state = Start;
  unsigned int i = 0;
  TQString ex = expr;
  TQString tokenText;
  int tokenStart = 0;

  // first character must be equal sign (=)
  if( ex[0] != '=' )
    return tokens;

  // but the scanner should not see this equal sign
  ex.remove( 0, 1 );

  // force a terminator
  ex.append( TQChar() );

  // main loop
  while( (state != Bad) && (state != Finish) && (i < ex.length()) )
  {
    TQChar ch = ex[i];

    switch( state )
    {

    case Start:

       tokenStart = i;

       // skip any whitespaces
       if( ch.isSpace() ) i++;

       // check for number
       else if( ch.isDigit() )
       {
         state = InNumber;
       }

       // a string ?
       else if ( ch == '"' )
       {
         tokenText.append( ex[i++] );
         state = InString;
       }

       // beginning with alphanumeric ?
       // could be identifier, cell, range, or function...
       else if( isIdentifier( ch ) )
       {
         state = InIdentifier;
       }

       // aposthrophe (') marks sheet name for 3-d cell, e.g 'Sales Q3'!A4, or a named range
       else if ( ch.tqunicode() == '\'' )
       {
         i++;
         state = InSheetOrAreaName;
       }

       // decimal dot ?
       else if ( ch == decimal )
       {
         tokenText.append( ex[i++] );
         state = InDecimal;
       }

       // terminator character
       else if ( ch == TQChar::null )
          state = Finish;

       // look for operator match
       else
       {
         int op;
         TQString s;

         // check for two-chars operator, such as '<=', '>=', etc
         s.append( ch ).append( ex[i+1] );
         op = matchOperator( s );

         // check for one-char operator, such as '+', ';', etc
         if( op == Token::InvalidOp )
         {
           s = TQString( ch );
           op = matchOperator( s );
         }

         // any matched operator ?
         if( op != Token::InvalidOp )
         {
           int len = s.length();
           i += len;
           tokens.append( Token( Token::Operator, s.left( len ), tokenStart ) );
         }
         else state = Bad;
        }
       break;

    case InIdentifier:

       // consume as long as alpha, dollar sign, underscore, or digit
       if( isIdentifier( ch )  || ch.isDigit() ) tokenText.append( ex[i++] );

       // a '!' ? then this must be sheet name, e.g "Sheet4!"
       else if( ch == '!' )
       {
          tokenText.append( ex[i++] );
          state = InCell;
       }

       // a '(' ? then this must be a function identifier
       else if( ch == '(' )
       {
           tokens.append (Token (Token::Identifier, tokenText, tokenStart));
           tokenStart = i;
           tokenText = "";
           state = Start;
       }

       // we're done with identifier
       else
       {
         // check for cell reference,  e.g A1, VV123, ...
         TQRegExp exp("(\\$?)([a-zA-Z]+)(\\$?)([0-9]+)$");
         int n = exp.search( tokenText );
         if( n >= 0 )
           state = InCell;
         else
         {
           if ( isNamedArea( tokenText ) )
             tokens.append (Token (Token::Range, tokenText, tokenStart));
           else
             tokens.append (Token (Token::Identifier, tokenText, tokenStart));
           tokenStart = i;
           tokenText = "";
           state = Start;
         }
       }
       break;

    case InCell:

       // consume as long as alpha, dollar sign, underscore, or digit
       if( isIdentifier( ch )  || ch.isDigit() ) tokenText.append( ex[i++] );

       // we're done with cell ref, possibly with sheet name (like "Sheet2!B2")
       // note that "Sheet2!TotalSales" is also possible, in which "TotalSales" is a named area
       else
       {

         // check if it's a cell ref like A32, not named area
         TQString cell;
         for( int j = tokenText.length()-1; j>=0; j-- )
           if( tokenText[j] == '!' )
               break;
           else
               cell.prepend( tokenText[j] );
         TQRegExp exp("(\\$?)([a-zA-Z]+)(\\$?)([0-9]+)$");
         if( exp.search( cell ) != 0 )
         {

           // we're done with named area
           // (Tomas) huh? this doesn't seem to check for named areas ...
           tokens.append( Token( Token::Range, tokenText, tokenStart ) );
           tokenText = "";
           state = Start;
         }

         else
         {

           // so up to now we've got something like A2 or Sheet2!F4
           // check for range reference
           if( ch == ':' )
           {
             tokenText.append( ex[i++] );
             state = InRange;
           }
           else
           {
             // we're done with cell reference
             tokens.append( Token( Token::Cell, tokenText, tokenStart ) );
             tokenText = "";
             state = Start;
           }
         }
       }
       break;

    case InRange:

       // consume as long as alpha, dollar sign, underscore, or digit
       if( isIdentifier( ch )  || ch.isDigit() ) tokenText.append( ex[i++] );

       // we're done with range reference
       else
       {
         tokens.append( Token( Token::Range, tokenText, tokenStart ) );
         tokenText = "";
         state = Start;
       }
       break;

    case InSheetOrAreaName:

       // consume until '
        if ( ch.tqunicode() != '\'' )
          tokenText.append( ex[i++] );

       else
       {
         // eat the aposthrophe itself
         ++i;
         // must be followed by '!' to be sheet name
         if( ex[i] == '!' )
         {
           tokenText.append( ex[i++] );
           state = InCell;
         }
         else
         {
           if ( isNamedArea( tokenText ) )
             tokens.append (Token (Token::Range, tokenText, tokenStart));
           else
             tokens.append (Token (Token::Identifier, tokenText, tokenStart));
           tokenStart = i;
           tokenText = "";
           state = Start;
         }
       }
       break;

    case InNumber:

       // consume as long as it's digit
       if( ch.isDigit() ) tokenText.append( ex[i++] );

       // skip thousand separator
       else if( !thousand.isEmpty() && ( ch ==thousand[0] ) ) i++;

       // convert decimal separator to '.', also support '.' directly
       // we always support '.' because of bug #98455
       else if(( !decimal.isEmpty() && ( ch == decimal[0] ) ) || (ch == '.'))
       {
         tokenText.append( '.' );
         i++;
         state = InDecimal;
       }

       // exponent ?
       else if( ch.upper() == 'E' )
       {
         tokenText.append( 'E' );
         i++;
         state = InExpIndicator;
       }

       // reference sheet delimiter?
       else if( ch == '!' )
       {
         tokenText.append( ex[i++] );
         state = InCell;
       }

       // identifier?
       else if( isIdentifier( ch ) )
       {
         // has to be a sheet or area name then
         state = InIdentifier;
       }

       // we're done with integer number
       else
       {
         tokens.append( Token( Token::Integer, tokenText, tokenStart ) );
         tokenText = "";
         state = Start;
       };
       break;

    case InDecimal:

       // consume as long as it's digit
       if( ch.isDigit() ) tokenText.append( ex[i++] );

       // exponent ?
       else if( ch.upper() == 'E' )
       {
         tokenText.append( 'E' );
         i++;
         state = InExpIndicator;
       }

       // we're done with floating-point number
       else
       {
         tokens.append( Token( Token::Float, tokenText, tokenStart ) );
         tokenText = "";
         state = Start;
       };
       break;

    case InExpIndicator:

       // possible + or - right after E, e.g 1.23E+12 or 4.67E-8
       if( ( ch == '+' ) || ( ch == '-' ) ) tokenText.append( ex[i++] );

       // consume as long as it's digit
       else if( ch.isDigit() ) state = InExponent;

       // invalid thing here
       else state = Bad;

       break;

    case InExponent:

       // consume as long as it's digit
       if( ch.isDigit() ) tokenText.append( ex[i++] );

       // we're done with floating-point number
       else
       {
         tokens.append( Token( Token::Float, tokenText, tokenStart ) );
         tokenText = "";
         state = Start;
       };
       break;

    case InString:

       // consume until "
       if( ch != '"' ) tokenText.append( ex[i++] );

       else
       {
         tokenText.append( ch ); i++;
         tokens.append( Token( Token::String, tokenText, tokenStart ) );
         tokenText = "";
         state = Start;
       }
       break;

    case Bad:
    default:
       break;
    };
  };

  if( state == Bad )
    tokens.setValid( false );

  return tokens;
}

// will affect: dirty, valid, codes, constants
void Formula::compile( const Tokens& tokens ) const
{
  // initialize variables
  d->dirty = false;
  d->valid = false;
  d->codes.clear();
  d->constants.clear();

  // sanity check
  if( tokens.count() == 0 ) return;

  TokenStack syntaxStack;
  TQValueStack<int> argStack;
  unsigned argCount = 1;

  for( unsigned i = 0; i <= tokens.count(); i++ )
  {
    // helper token: InvalidOp is end-of-formula
    Token token =  ( i < tokens.count() ) ? tokens[i] : Token( Token::Operator );
    Token::Type tokenType = token.type();

    // unknown token is invalid
    if( tokenType == Token::Unknown ) break;

    // for constants, push immediately to stack
    // generate code to load from a constant
    if ( ( tokenType == Token::Integer ) || ( tokenType == Token::Float ) ||
    ( tokenType == Token::String ) || ( tokenType == Token::Boolean ) )
    {
      syntaxStack.push( token );
      d->constants.append( tokenAsValue( token ) );
      d->codes.append( Opcode( Opcode::Load, d->constants.count()-1 ) );
    }

    // for cell, range, or identifier, push immediately to stack
    // generate code to load from reference
    if( ( tokenType == Token::Cell ) || ( tokenType == Token::Range ) ||
    ( tokenType == Token::Identifier ) )
    {
      syntaxStack.push( token );
      d->constants.append( Value( token.text() ) );
      if (tokenType == Token::Cell)
        d->codes.append( Opcode( Opcode::Cell, d->constants.count()-1 ) );
      else if (tokenType == Token::Range)
        d->codes.append( Opcode( Opcode::Range, d->constants.count()-1 ) );
      else
        d->codes.append( Opcode( Opcode::Ref, d->constants.count()-1 ) );
    }

    // are we entering a function ?
    // if token is operator, and stack already has: id ( arg
    if( tokenType == Token::Operator )
    if( syntaxStack.itemCount() >= 3 )
    {
        Token arg = syntaxStack.top();
        Token par = syntaxStack.top( 1 );
        Token id = syntaxStack.top( 2 );
        if( !arg.isOperator() )
        if( par.asOperator() == Token::LeftPar )
        if( id.isIdentifier() )
        {
          argStack.push( argCount );
          argCount = 1;
        }
     }

     // special case for percentage
    if( tokenType == Token::Operator )
    if( token.asOperator() == Token::Percent )
    if( syntaxStack.itemCount() >= 1 )
    if( !syntaxStack.top().isOperator() )
    {
      d->constants.append( Value( 0.01 ) );
      d->codes.append( Opcode( Opcode::Load, d->constants.count()-1 ) );
      d->codes.append( Opcode( Opcode::Mul ) );
    }

    // for any other operator, try to apply all parsing rules
    if( tokenType == Token::Operator )
    if( token.asOperator() != Token::Percent )
    {
      // repeat until no more rule applies
      for( ; ; )
      {
        bool ruleFound = false;

        // rule for function arguments, if token is ; or )
        // id ( arg1 ; arg2 -> id ( arg
        if( !ruleFound )
        if( syntaxStack.itemCount() >= 5 )
        if( ( token.asOperator() == Token::RightPar ) ||
        ( token.asOperator() == Token::Semicolon ) )
        {
          Token arg2 = syntaxStack.top();
          Token sep = syntaxStack.top( 1 );
          Token arg1 = syntaxStack.top( 2 );
          Token par = syntaxStack.top( 3 );
          Token id = syntaxStack.top( 4 );
          if( !arg2.isOperator() )
          if( sep.asOperator() == Token::Semicolon )
          if( !arg1.isOperator() )
          if( par.asOperator() == Token::LeftPar )
          if( id.isIdentifier() )
          {
            ruleFound = true;
            syntaxStack.pop();
            syntaxStack.pop();
            argCount++;
          }
        }

        // rule for function last argument:
        //  id ( arg ) -> arg
        if( !ruleFound )
        if( syntaxStack.itemCount() >= 4 )
        {
          Token par2 = syntaxStack.top();
          Token arg = syntaxStack.top( 1 );
          Token par1 = syntaxStack.top( 2 );
          Token id = syntaxStack.top( 3 );
          if( par2.asOperator() == Token::RightPar )
          if( !arg.isOperator() )
          if( par1.asOperator() == Token::LeftPar )
          if( id.isIdentifier() )
          {
            ruleFound = true;
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.push( arg );
            d->codes.append( Opcode( Opcode::Function, argCount ) );
            argCount = argStack.empty() ? 0 : argStack.pop();
          }
        }

        // rule for function call with tqparentheses, but without argument
        // e.g. "2*PI()"
        if( !ruleFound )
        if( syntaxStack.itemCount() >= 3 )
        {
          Token par2 = syntaxStack.top();
          Token par1 = syntaxStack.top( 1 );
          Token id = syntaxStack.top( 2 );
          if( par2.asOperator() == Token::RightPar )
          if( par1.asOperator() == Token::LeftPar )
          if( id.isIdentifier() )
          {
            ruleFound = true;
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.push( Token( Token::Integer ) );
            d->codes.append( Opcode( Opcode::Function, 0 ) );
          }
        }

        // rule for tqparenthesis:  ( Y ) -> Y
        if( !ruleFound )
        if( syntaxStack.itemCount() >= 3 )
        {
          Token right = syntaxStack.top();
          Token y = syntaxStack.top( 1 );
          Token left = syntaxStack.top( 2 );
          if( right.isOperator() )
          if( !y.isOperator() )
          if( left.isOperator() )
          if( right.asOperator() == Token::RightPar )
          if( left.asOperator() == Token::LeftPar )
          {
            ruleFound = true;
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.push( y );
          }
        }

        // rule for binary operator:  A (op) B -> A
        // conditions: precedence of op >= precedence of token
        // action: push (op) to result
        // e.g. "A * B" becomes "A" if token is operator "+"
        // exception: for caret (power operator), if op is another caret
        // then the rule doesn't apply, e.g. "2^3^2" is evaluated as "2^(3^2)"
        if( !ruleFound )
        if( syntaxStack.itemCount() >= 3 )
        {
          Token b = syntaxStack.top();
          Token op = syntaxStack.top( 1 );
          Token a = syntaxStack.top( 2 );
          if( !a.isOperator() )
          if( !b.isOperator() )
          if( op.isOperator() )
          if( token.asOperator() != Token::LeftPar )
          if( token.asOperator() != Token::Caret )
          if( opPrecedence( op.asOperator() ) >= opPrecedence( token.asOperator() ) )
          {
            ruleFound = true;
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.pop();
            syntaxStack.push( b );
            switch( op.asOperator() )
            {
              // simple binary operations
              case Token::Plus:         d->codes.append( Opcode::Add ); break;
              case Token::Minus:        d->codes.append( Opcode::Sub ); break;
              case Token::Asterisk:     d->codes.append( Opcode::Mul ); break;
              case Token::Slash:        d->codes.append( Opcode::Div ); break;
              case Token::Caret:        d->codes.append( Opcode::Pow ); break;
              case Token::Ampersand:    d->codes.append( Opcode::Concat ); break;

              // simple value comparisons
              case Token::Equal:        d->codes.append( Opcode::Equal ); break;
              case Token::Less:         d->codes.append( Opcode::Less ); break;
              case Token::Greater:      d->codes.append( Opcode::Greater ); break;

              // NotEqual is Equal, followed by Not
              case Token::NotEqual:
                d->codes.append( Opcode::Equal );
                d->codes.append( Opcode::Not );
                break;

              // LessOrEqual is Greater, followed by Not
              case Token::LessEqual:
                d->codes.append( Opcode::Greater );
                d->codes.append( Opcode::Not );
                break;

              // GreaterOrEqual is Less, followed by Not
              case Token::GreaterEqual:
                d->codes.append( Opcode::Less );
                d->codes.append( Opcode::Not );
                break;
              default: break;
            };
          }
         }

         // rule for unary operator:  (op1) (op2) X -> (op1) X
         // conditions: op2 is unary, token is not '('
         // action: push (op2) to result
         // e.g.  "* - 2" becomes "*"
         if( !ruleFound )
         if( token.asOperator() != Token::LeftPar )
         if( syntaxStack.itemCount() >= 3 )
         {
            Token x = syntaxStack.top();
            Token op2 = syntaxStack.top( 1 );
            Token op1 = syntaxStack.top( 2 );
            if( !x.isOperator() )
            if( op1.isOperator() )
            if( op2.isOperator() )
            if( ( op2.asOperator() == Token::Plus ) ||
               ( op2.asOperator() == Token::Minus ) )
            {
              ruleFound = true;
              syntaxStack.pop();
              syntaxStack.pop();
              syntaxStack.push( x );
              if( op2.asOperator() == Token::Minus )
                d->codes.append( Opcode( Opcode::Neg ) );
            }
          }

         // auxilary rule for unary operator:  (op) X -> X
         // conditions: op is unary, op is first in syntax stack, token is not '('
         // action: push (op) to result
         if( !ruleFound )
         if( token.asOperator() != Token::LeftPar )
         if( syntaxStack.itemCount() == 2 )
         {
            Token x = syntaxStack.top();
            Token op = syntaxStack.top( 1 );
            if( !x.isOperator() )
            if( op.isOperator() )
            if( ( op.asOperator() == Token::Plus ) ||
               ( op.asOperator() == Token::Minus ) )
            {
              ruleFound = true;
              syntaxStack.pop();
              syntaxStack.pop();
              syntaxStack.push( x );
              if( op.asOperator() == Token::Minus )
                d->codes.append( Opcode( Opcode::Neg ) );
            }
          }

        if( !ruleFound ) break;
      }

      // can't apply rules anymore, push the token
      if( token.asOperator() != Token::Percent )
        syntaxStack.push( token );
    }
  }

  // syntaxStack must left only one operand and end-of-formula (i.e. InvalidOp)
  d->valid = false;
  if( syntaxStack.itemCount() == 2 )
  if( syntaxStack.top().isOperator() )
  if( syntaxStack.top().asOperator() == Token::InvalidOp )
  if( !syntaxStack.top(1).isOperator() )
    d->valid = true;

  // bad parsing ? clean-up everything
  if( !d->valid )
  {
    d->constants.clear();
    d->codes.clear();
  }
}

bool Formula::isNamedArea( const TQString& expr ) const
{
    TQString tokenText( expr );
    // check for named areas ...
    if (d->sheet) {
        const TQValueList<Reference> areas = d->sheet->doc()->listArea();
        TQValueList<Reference>::const_iterator it;
        for (it = areas.begin(); it != areas.end(); ++it) {
            if ((*it).ref_name.lower() == tokenText.lower()) {
                 // we got a named area
                return true;
            }
        }
    }
    return false;
}


// Evaluates the formula, returns the result.

struct stackEntry {
  void reset () { row1 = col1 = row2 = col2 = -1; };
  Value val;
  int row1, col1, row2, col2;
};

Value Formula::eval() const
{
  TQValueStack<stackEntry> stack;
  stackEntry entry;
  unsigned index;
  Value val1, val2;
  TQString c;
  TQValueVector<Value> args;

  Sheet *sheet = 0;
  ValueParser* parser = 0;
  ValueConverter* converter = 0;
  ValueCalc* calc = 0;

  if (d->sheet)
  {
    sheet = d->sheet;
    converter = sheet->doc()->converter();
    calc = sheet->doc()->calc();
  }
  else
  {
    parser = new ValueParser( KGlobal::locale() );
    converter = new ValueConverter( parser );
    calc = new ValueCalc( converter );
  }

  Function* function;
  FuncExtra fe;
  fe.mycol = fe.myrow = 0;
  if (d->cell) {
    fe.mycol = d->cell->column();
    fe.myrow = d->cell->row();
  }

  if( d->dirty )
  {
    Tokens tokens = scan( d->expression );
    d->valid = tokens.valid();
    if( tokens.valid() )
      compile( tokens );
  }

  if( !d->valid )
    return Value::errorVALUE();

  for( unsigned pc = 0; pc < d->codes.count(); pc++ )
  {
    Value ret;   // for the function caller
    Opcode& opcode = d->codes[pc];
    index = opcode.index;
    switch( opcode.type )
    {
      // no operation
      case Opcode::Nop:
        break;

      // load a constant, push to stack
      case Opcode::Load:
        entry.reset();
        entry.val = d->constants[index];
        stack.push (entry);
        break;

      // unary operation
      case Opcode::Neg:
        entry.reset();
        entry.val = stack.pop().val;
        if (!entry.val.isError()) // do nothing if we got an error
          entry.val = calc->mul (entry.val, -1);
        stack.push (entry);
        break;

      // binary operation: take two values from stack, do the operation,
      // push the result to stack
      case Opcode::Add:
        entry.reset();
        val2 = stack.pop().val;
        val1 = stack.pop().val;
        val2 = calc->add( val1, val2 );
        entry.reset();
        entry.val = val2;
        stack.push (entry);
        break;

      case Opcode::Sub:
        val2 = stack.pop().val;
        val1 = stack.pop().val;
        val2 = calc->sub( val1, val2 );
        entry.reset();
        entry.val = val2;
        stack.push (entry);
        break;

      case Opcode::Mul:
        val2 = stack.pop().val;
        val1 = stack.pop().val;
        val2 = calc->mul( val1, val2 );
        entry.reset();
        entry.val = val2;
        stack.push (entry);
        break;

      case Opcode::Div:
        val2 = stack.pop().val;
        val1 = stack.pop().val;
        val2 = calc->div( val1, val2 );
        entry.reset();
        entry.val = val2;
        stack.push (entry);
        break;

      case Opcode::Pow:
        val2 = stack.pop().val;
        val1 = stack.pop().val;
        val2 = calc->pow( val1, val2 );
        entry.reset();
        entry.val = val2;
        stack.push (entry);
        break;

      // string concatenation
      case Opcode::Concat:
        val1 = converter->asString (stack.pop().val);
	val2 = converter->asString (stack.pop().val);
        if (val1.isError() || val2.isError())
	  val1 = Value::errorVALUE();
	else
          val1.setValue( val2.asString().append( val1.asString() ) );
        entry.reset();
        entry.val = val1;
        stack.push (entry);
        break;

      // logical not
      case Opcode::Not:
        val1 = converter->asBoolean (stack.pop().val);
        if( val1.isError() )
	  val1 = Value::errorVALUE();
	else
	  val1.setValue( !val1.asBoolean() );
        entry.reset();
        entry.val = val1;
        stack.push (entry);
        break;

      // comparison
      case Opcode::Equal:
        val1 = stack.pop().val;
        val2 = stack.pop().val;
        if( !val1.allowComparison( val2 ) )
          val1 = Value::errorNA();
	else if( val2.compare( val1 ) == 0 )
          val1 = Value (true);
        else
          val1 = Value (false);
        entry.reset();
        entry.val = val1;
        stack.push (entry);
        break;

      // less than
      case Opcode::Less:
        val1 = stack.pop().val;
        val2 = stack.pop().val;
        if( !val1.allowComparison( val2 ) )
          val1 = Value::errorNA();
	else if( val2.compare( val1 ) < 0 )
          val1 = Value (true);
        else
          val1 = Value (false);
        entry.reset();
        entry.val = val1;
        stack.push (entry);
        break;

      // greater than
      case Opcode::Greater:
        val1 = stack.pop().val;
        val2 = stack.pop().val;
        if( !val1.allowComparison( val2 ) )
          val1 = Value::errorNA();
	else if( val2.compare( val1 ) > 0 )
          val1 = Value (true);
        else
          val1 = Value (false);
        entry.reset();
        entry.val = val1;
        stack.push (entry);
        break;


      case Opcode::Cell:
        c = d->constants[index].asString();
        val1 = Value::empty();
        entry.reset();
        if (sheet)
        {
          Point cell (c, sheet->workbook(), sheet);
          if (cell.isValid())
          {
            val1 = cell.sheet()->value (cell.column(), cell.row());
            // store the reference, so we can use it within functions
            entry.col1 = entry.col2 = cell.column();
            entry.row1 = entry.row2 = cell.row();
          }
        }
        entry.val = val1;
        stack.push (entry);
        break;

      case Opcode::Range:
        c = d->constants[index].asString();
        val1 = Value::empty();
        entry.reset();
        if (sheet)
        {
          Range range (c, sheet->workbook(), sheet);
          if (range.isValid())
          {
            val1 = range.sheet()->valueRange (range.startCol(), range.startRow(),
                range.endCol(), range.endRow());
            // store the reference, so we can use it within functions
            entry.col1 = range.startCol();
            entry.row1 = range.startRow();
            entry.col2 = range.endCol();
            entry.row2 = range.endRow();
          }
        }
        entry.val = val1;
        stack.push (entry);
        break;

      case Opcode::Ref:
        val1 = d->constants[index];
        entry.reset();
        entry.val = val1;
        stack.push (entry);
        break;

      // calling function
      case Opcode::Function:
        if( stack.count() < index )
          // (Tomas) umm, how could that be ? I mean, the index value
          //  is computed from the stack *confused*
          return Value::errorVALUE(); // not enough arguments

        args.clear();
        fe.ranges.clear ();
        fe.ranges.reserve (index);
        fe.sheet = sheet;
        for( ; index; index-- )
        {
          stackEntry e = stack.pop();
          args.insert (args.begin(), e.val);
          // TODO: create and fill a FunctionExtra object, if needed
          // problem: we don't know if we need it, as we don't have the
          // fuction name yet ...
          fe.ranges[index - 1].col1 = e.col1;
          fe.ranges[index - 1].row1 = e.row1;
          fe.ranges[index - 1].col2 = e.col2;
          fe.ranges[index - 1].row2 = e.row2;
        }

        // function name as string value
        val1 = converter->asString (stack.pop().val);
        if( val1.isError() )
          return Value::errorVALUE();
        function = FunctionRepository::self()->function ( val1.asString() );
        if( !function )
          return Value::errorVALUE(); // no such function

        ret = function->exec (args, calc, &fe);
        entry.reset();
        entry.val = ret;
        stack.push (entry);

        break;

      default:
        break;
    }
  }

  if (!d->sheet) {
    delete parser;
    delete converter;
    delete calc;
  }

  // more than one value in stack ? unsuccesful execution...
  if( stack.count() != 1 )
    return Value::errorVALUE();

  return stack.pop().val;

}

// Debugging aid

TQString Formula::dump() const
{
  TQString result;

  if( d->dirty )
  {
    Tokens tokens = scan( d->expression );
    compile( tokens );
  }

  result = TQString("Expression: [%1]\n").tqarg( d->expression );
#if 0
  Value value = eval();
  result.append( TQString("Result: %1\n").tqarg(
      converter->asString(value).asString() ) );
#endif

  result.append("  Constants:\n");
  for( unsigned c = 0; c < d->constants.count(); c++ )
  {
    TQString vtext;
    Value val = d->constants[c];
    if( val.isString() ) vtext = TQString("[%1]").tqarg( val.asString() );
    else if( val.isNumber() ) vtext = TQString("%1").tqarg( val.asFloat() );
    else if( val.isBoolean() ) vtext = TQString("%1").tqarg( val.asBoolean() ? "True":"False");
    else if( val.isError() ) vtext = "error";
    else vtext = "???";
    result += TQString("    #%1 = %2\n").tqarg(c).tqarg( vtext );
  }

  result.append("\n");
  result.append("  Code:\n");
  for( unsigned i = 0; i < d->codes.count(); i++ )
  {
    TQString ctext;
    switch( d->codes[i].type )
    {
      case Opcode::Load:     ctext = TQString("Load #%1").tqarg( d->codes[i].index ); break;
      case Opcode::Ref:      ctext = TQString("Ref #%1").tqarg( d->codes[i].index ); break;
      case Opcode::Function: ctext = TQString("Function (%1)").tqarg( d->codes[i].index ); break;
      case Opcode::Add:      ctext = "Add"; break;
      case Opcode::Sub:      ctext = "Sub"; break;
      case Opcode::Mul:      ctext = "Mul"; break;
      case Opcode::Div:      ctext = "Div"; break;
      case Opcode::Neg:      ctext = "Neg"; break;
      case Opcode::Concat:   ctext = "Concat"; break;
      case Opcode::Pow:      ctext = "Pow"; break;
      case Opcode::Equal:    ctext = "Equal"; break;
      case Opcode::Not:      ctext = "Not"; break;
      case Opcode::Less:     ctext = "Less"; break;
      case Opcode::Greater:  ctext = "Greater"; break;
      default: ctext = "Unknown"; break;
    }
    result.append( "   " ).append( ctext ).append("\n");
  }

  return result;
}

TQTextStream& operator<<( TQTextStream& ts, Formula formula )
{
  ts << formula.dump();
  return ts;
}