office-gobmx/starmath/source/parse.cxx
2011-11-27 13:23:17 -06:00

2533 lines
79 KiB
C++

/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
/*************************************************************************
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* Copyright 2000, 2010 Oracle and/or its affiliates.
*
* OpenOffice.org - a multi-platform office productivity suite
*
* This file is part of OpenOffice.org.
*
* OpenOffice.org is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License version 3
* only, as published by the Free Software Foundation.
*
* OpenOffice.org 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 Lesser General Public License version 3 for more details
* (a copy is included in the LICENSE file that accompanied this code).
*
* You should have received a copy of the GNU Lesser General Public License
* version 3 along with OpenOffice.org. If not, see
* <http://www.openoffice.org/license.html>
* for a copy of the LGPLv3 License.
*
************************************************************************/
#include <stdio.h>
#define SMDLL 1
#include <com/sun/star/i18n/UnicodeType.hpp>
#include <i18npool/lang.h>
#include <unotools/charclass.hxx>
#include <editeng/unolingu.hxx>
#include <unotools/syslocale.hxx>
#include <sal/macros.h>
#include "parse.hxx"
#include "starmath.hrc"
#include "smdll.hxx"
#include "smmod.hxx"
#include "config.hxx"
#include "node.hxx"
using namespace ::com::sun::star;
using namespace ::com::sun::star::i18n;
///////////////////////////////////////////////////////////////////////////
static inline bool strnccmp(const String &u1, xub_StrLen nIdx,
const sal_Char *s2, xub_StrLen nLen)
{
return u1.EqualsIgnoreCaseAscii( s2, nIdx, nLen );
}
static const sal_Unicode aDelimiterTable[] =
{
' ', '\t', '\n', '\r', '+', '-', '*', '/', '=', '#',
'%', '\\', '"', '~', '`', '>', '<', '&', '|', '(',
')', '{', '}', '[', ']', '^', '_',
'\0' // end of list symbol
};
static inline bool IsDigit( sal_Unicode cChar )
{
return '0' <= cChar && cChar <= '9';
}
///////////////////////////////////////////////////////////////////////////
SmToken::SmToken() :
eType (TUNKNOWN),
cMathChar ('\0')
{
nGroup = nCol = nRow = nLevel = 0;
}
SmToken::SmToken(SmTokenType eTokenType,
sal_Unicode cMath,
const sal_Char* pText,
sal_uLong nTokenGroup,
sal_uInt16 nTokenLevel) {
eType = eTokenType;
cMathChar = cMath;
aText.AssignAscii(pText);
nGroup = nTokenGroup;
nLevel = nTokenLevel;
nCol = nRow = 0;
}
///////////////////////////////////////////////////////////////////////////
static const SmTokenTableEntry aTokenTable[] =
{
{ "Im" , TIM, MS_IM, TGSTANDALONE, 5 },
{ "MZ23", TDEBUG, '\0', TGATTRIBUT, 0 },
{ "Re" , TRE, MS_RE, TGSTANDALONE, 5 },
{ "abs", TABS, '\0', TGUNOPER, 13 },
{ "arcosh", TACOSH, '\0', TGFUNCTION, 5 },
{ "arcoth", TACOTH, '\0', TGFUNCTION, 5 },
{ "acute", TACUTE, MS_ACUTE, TGATTRIBUT, 5 },
{ "aleph" , TALEPH, MS_ALEPH, TGSTANDALONE, 5 },
{ "alignb", TALIGNC, '\0', TGALIGN | TGDISCARDED, 0},
{ "alignc", TALIGNC, '\0', TGALIGN, 0},
{ "alignl", TALIGNL, '\0', TGALIGN, 0},
{ "alignm", TALIGNC, '\0', TGALIGN | TGDISCARDED, 0},
{ "alignr", TALIGNR, '\0', TGALIGN, 0},
{ "alignt", TALIGNC, '\0', TGALIGN | TGDISCARDED, 0},
{ "and", TAND, MS_AND, TGPRODUCT, 0},
{ "approx", TAPPROX, MS_APPROX, TGRELATION, 0},
{ "arccos", TACOS, '\0', TGFUNCTION, 5},
{ "arccot", TACOT, '\0', TGFUNCTION, 5},
{ "arcsin", TASIN, '\0', TGFUNCTION, 5},
{ "arctan", TATAN, '\0', TGFUNCTION, 5},
{ "arsinh", TASINH, '\0', TGFUNCTION, 5},
{ "artanh", TATANH, '\0', TGFUNCTION, 5},
{ "backepsilon" , TBACKEPSILON, MS_BACKEPSILON, TGSTANDALONE, 5},
{ "bar", TBAR, MS_BAR, TGATTRIBUT, 5},
{ "binom", TBINOM, '\0', 0, 5 },
{ "black", TBLACK, '\0', TGCOLOR, 0},
{ "blue", TBLUE, '\0', TGCOLOR, 0},
{ "bold", TBOLD, '\0', TGFONTATTR, 5},
{ "boper", TBOPER, '\0', TGPRODUCT, 0},
{ "breve", TBREVE, MS_BREVE, TGATTRIBUT, 5},
{ "bslash", TBACKSLASH, MS_BACKSLASH, TGPRODUCT, 0 },
{ "cdot", TCDOT, MS_CDOT, TGPRODUCT, 0},
{ "check", TCHECK, MS_CHECK, TGATTRIBUT, 5},
{ "circ" , TCIRC, MS_CIRC, TGSTANDALONE, 5},
{ "circle", TCIRCLE, MS_CIRCLE, TGATTRIBUT, 5},
{ "color", TCOLOR, '\0', TGFONTATTR, 5},
{ "coprod", TCOPROD, MS_COPROD, TGOPER, 5},
{ "cos", TCOS, '\0', TGFUNCTION, 5},
{ "cosh", TCOSH, '\0', TGFUNCTION, 5},
{ "cot", TCOT, '\0', TGFUNCTION, 5},
{ "coth", TCOTH, '\0', TGFUNCTION, 5},
{ "csub", TCSUB, '\0', TGPOWER, 0},
{ "csup", TCSUP, '\0', TGPOWER, 0},
{ "cyan", TCYAN, '\0', TGCOLOR, 0},
{ "dddot", TDDDOT, MS_DDDOT, TGATTRIBUT, 5},
{ "ddot", TDDOT, MS_DDOT, TGATTRIBUT, 5},
{ "def", TDEF, MS_DEF, TGRELATION, 0},
{ "div", TDIV, MS_DIV, TGPRODUCT, 0},
{ "divides", TDIVIDES, MS_LINE, TGRELATION, 0},
{ "dlarrow" , TDLARROW, MS_DLARROW, TGSTANDALONE, 5},
{ "dlrarrow" , TDLRARROW, MS_DLRARROW, TGSTANDALONE, 5},
{ "dot", TDOT, MS_DOT, TGATTRIBUT, 5},
{ "dotsaxis", TDOTSAXIS, MS_DOTSAXIS, TGSTANDALONE, 5}, // 5 to continue expression
{ "dotsdiag", TDOTSDIAG, MS_DOTSUP, TGSTANDALONE, 5}, //
{ "dotsdown", TDOTSDOWN, MS_DOTSDOWN, TGSTANDALONE, 5}, //
{ "dotslow", TDOTSLOW, MS_DOTSLOW, TGSTANDALONE, 5}, //
{ "dotsup", TDOTSUP, MS_DOTSUP, TGSTANDALONE, 5}, //
{ "dotsvert", TDOTSVERT, MS_DOTSVERT, TGSTANDALONE, 5}, //
{ "downarrow" , TDOWNARROW, MS_DOWNARROW, TGSTANDALONE, 5},
{ "drarrow" , TDRARROW, MS_DRARROW, TGSTANDALONE, 5},
{ "emptyset" , TEMPTYSET, MS_EMPTYSET, TGSTANDALONE, 5},
{ "equiv", TEQUIV, MS_EQUIV, TGRELATION, 0},
{ "exists", TEXISTS, MS_EXISTS, TGSTANDALONE, 5},
{ "notexists", TNOTEXISTS, MS_NOTEXISTS, TGSTANDALONE, 5},
{ "exp", TEXP, '\0', TGFUNCTION, 5},
{ "fact", TFACT, MS_FACT, TGUNOPER, 5},
{ "fixed", TFIXED, '\0', TGFONT, 0},
{ "font", TFONT, '\0', TGFONTATTR, 5},
{ "forall", TFORALL, MS_FORALL, TGSTANDALONE, 5},
{ "from", TFROM, '\0', TGLIMIT, 0},
{ "func", TFUNC, '\0', TGFUNCTION, 5},
{ "ge", TGE, MS_GE, TGRELATION, 0},
{ "geslant", TGESLANT, MS_GESLANT, TGRELATION, 0 },
{ "gg", TGG, MS_GG, TGRELATION, 0},
{ "grave", TGRAVE, MS_GRAVE, TGATTRIBUT, 5},
{ "green", TGREEN, '\0', TGCOLOR, 0},
{ "gt", TGT, MS_GT, TGRELATION, 0},
{ "hat", THAT, MS_HAT, TGATTRIBUT, 5},
{ "hbar" , THBAR, MS_HBAR, TGSTANDALONE, 5},
{ "iiint", TIIINT, MS_IIINT, TGOPER, 5},
{ "iint", TIINT, MS_IINT, TGOPER, 5},
{ "in", TIN, MS_IN, TGRELATION, 0},
{ "infinity" , TINFINITY, MS_INFINITY, TGSTANDALONE, 5},
{ "infty" , TINFINITY, MS_INFINITY, TGSTANDALONE, 5},
{ "int", TINT, MS_INT, TGOPER, 5},
{ "intersection", TINTERSECT, MS_INTERSECT, TGPRODUCT, 0},
{ "ital", TITALIC, '\0', TGFONTATTR, 5},
{ "italic", TITALIC, '\0', TGFONTATTR, 5},
{ "lambdabar" , TLAMBDABAR, MS_LAMBDABAR, TGSTANDALONE, 5},
{ "langle", TLANGLE, MS_LANGLE, TGLBRACES, 5},
{ "lbrace", TLBRACE, MS_LBRACE, TGLBRACES, 5},
{ "lceil", TLCEIL, MS_LCEIL, TGLBRACES, 5},
{ "ldbracket", TLDBRACKET, MS_LDBRACKET, TGLBRACES, 5},
{ "ldline", TLDLINE, MS_DLINE, TGLBRACES, 5},
{ "le", TLE, MS_LE, TGRELATION, 0},
{ "left", TLEFT, '\0', 0, 5},
{ "leftarrow" , TLEFTARROW, MS_LEFTARROW, TGSTANDALONE, 5},
{ "leslant", TLESLANT, MS_LESLANT, TGRELATION, 0 },
{ "lfloor", TLFLOOR, MS_LFLOOR, TGLBRACES, 5},
{ "lim", TLIM, '\0', TGOPER, 5},
{ "liminf", TLIMINF, '\0', TGOPER, 5},
{ "limsup", TLIMSUP, '\0', TGOPER, 5},
{ "lint", TLINT, MS_LINT, TGOPER, 5},
{ "ll", TLL, MS_LL, TGRELATION, 0},
{ "lline", TLLINE, MS_LINE, TGLBRACES, 5},
{ "llint", TLLINT, MS_LLINT, TGOPER, 5},
{ "lllint", TLLLINT, MS_LLLINT, TGOPER, 5},
{ "ln", TLN, '\0', TGFUNCTION, 5},
{ "log", TLOG, '\0', TGFUNCTION, 5},
{ "lsub", TLSUB, '\0', TGPOWER, 0},
{ "lsup", TLSUP, '\0', TGPOWER, 0},
{ "lt", TLT, MS_LT, TGRELATION, 0},
{ "magenta", TMAGENTA, '\0', TGCOLOR, 0},
{ "matrix", TMATRIX, '\0', 0, 5},
{ "minusplus", TMINUSPLUS, MS_MINUSPLUS, TGUNOPER | TGSUM, 5},
{ "mline", TMLINE, MS_LINE, 0, 0}, //! nicht in TGRBRACES, Level 0
{ "nabla", TNABLA, MS_NABLA, TGSTANDALONE, 5},
{ "nbold", TNBOLD, '\0', TGFONTATTR, 5},
{ "ndivides", TNDIVIDES, MS_NDIVIDES, TGRELATION, 0},
{ "neg", TNEG, MS_NEG, TGUNOPER, 5 },
{ "neq", TNEQ, MS_NEQ, TGRELATION, 0},
{ "newline", TNEWLINE, '\0', 0, 0},
{ "ni", TNI, MS_NI, TGRELATION, 0},
{ "nitalic", TNITALIC, '\0', TGFONTATTR, 5},
{ "none", TNONE, '\0', TGLBRACES | TGRBRACES, 0},
{ "nospace", TNOSPACE, '\0', TGSTANDALONE, 5},
{ "notin", TNOTIN, MS_NOTIN, TGRELATION, 0},
{ "nroot", TNROOT, MS_SQRT, TGUNOPER, 5},
{ "nsubset", TNSUBSET, MS_NSUBSET, TGRELATION, 0 },
{ "nsupset", TNSUPSET, MS_NSUPSET, TGRELATION, 0 },
{ "nsubseteq", TNSUBSETEQ, MS_NSUBSETEQ, TGRELATION, 0 },
{ "nsupseteq", TNSUPSETEQ, MS_NSUPSETEQ, TGRELATION, 0 },
{ "odivide", TODIVIDE, MS_ODIVIDE, TGPRODUCT, 0},
{ "odot", TODOT, MS_ODOT, TGPRODUCT, 0},
{ "ominus", TOMINUS, MS_OMINUS, TGSUM, 0},
{ "oper", TOPER, '\0', TGOPER, 5},
{ "oplus", TOPLUS, MS_OPLUS, TGSUM, 0},
{ "or", TOR, MS_OR, TGSUM, 0},
{ "ortho", TORTHO, MS_ORTHO, TGRELATION, 0},
{ "otimes", TOTIMES, MS_OTIMES, TGPRODUCT, 0},
{ "over", TOVER, '\0', TGPRODUCT, 0},
{ "overbrace", TOVERBRACE, MS_OVERBRACE, TGPRODUCT, 5},
{ "overline", TOVERLINE, '\0', TGATTRIBUT, 5},
{ "overstrike", TOVERSTRIKE, '\0', TGATTRIBUT, 5},
{ "owns", TNI, MS_NI, TGRELATION, 0},
{ "parallel", TPARALLEL, MS_DLINE, TGRELATION, 0},
{ "partial", TPARTIAL, MS_PARTIAL, TGSTANDALONE, 5 },
{ "phantom", TPHANTOM, '\0', TGFONTATTR, 5},
{ "plusminus", TPLUSMINUS, MS_PLUSMINUS, TGUNOPER | TGSUM, 5},
{ "prec", TPRECEDES, MS_PRECEDES, TGRELATION, 0 },
{ "preccurlyeq", TPRECEDESEQUAL, MS_PRECEDESEQUAL, TGRELATION, 0 },
{ "precsim", TPRECEDESEQUIV, MS_PRECEDESEQUIV, TGRELATION, 0 },
{ "nprec", TNOTPRECEDES, MS_NOTPRECEDES, TGRELATION, 0 },
{ "prod", TPROD, MS_PROD, TGOPER, 5},
{ "prop", TPROP, MS_PROP, TGRELATION, 0},
{ "rangle", TRANGLE, MS_RANGLE, TGRBRACES, 0}, //! 0 to terminate expression
{ "rbrace", TRBRACE, MS_RBRACE, TGRBRACES, 0}, //
{ "rceil", TRCEIL, MS_RCEIL, TGRBRACES, 0}, //
{ "rdbracket", TRDBRACKET, MS_RDBRACKET, TGRBRACES, 0}, //
{ "rdline", TRDLINE, MS_DLINE, TGRBRACES, 0}, //
{ "red", TRED, '\0', TGCOLOR, 0},
{ "rfloor", TRFLOOR, MS_RFLOOR, TGRBRACES, 0}, //! 0 to terminate expression
{ "right", TRIGHT, '\0', 0, 0},
{ "rightarrow" , TRIGHTARROW, MS_RIGHTARROW, TGSTANDALONE, 5},
{ "rline", TRLINE, MS_LINE, TGRBRACES, 0}, //! 0 to terminate expression
{ "rsub", TRSUB, '\0', TGPOWER, 0},
{ "rsup", TRSUP, '\0', TGPOWER, 0},
{ "sans", TSANS, '\0', TGFONT, 0},
{ "serif", TSERIF, '\0', TGFONT, 0},
{ "setC" , TSETC, MS_SETC, TGSTANDALONE, 5},
{ "setN" , TSETN, MS_SETN, TGSTANDALONE, 5},
{ "setQ" , TSETQ, MS_SETQ, TGSTANDALONE, 5},
{ "setR" , TSETR, MS_SETR, TGSTANDALONE, 5},
{ "setZ" , TSETZ, MS_SETZ, TGSTANDALONE, 5},
{ "setminus", TBACKSLASH, MS_BACKSLASH, TGPRODUCT, 0 },
{ "sim", TSIM, MS_SIM, TGRELATION, 0},
{ "simeq", TSIMEQ, MS_SIMEQ, TGRELATION, 0},
{ "sin", TSIN, '\0', TGFUNCTION, 5},
{ "sinh", TSINH, '\0', TGFUNCTION, 5},
{ "size", TSIZE, '\0', TGFONTATTR, 5},
{ "slash", TSLASH, MS_SLASH, TGPRODUCT, 0 },
{ "sqrt", TSQRT, MS_SQRT, TGUNOPER, 5},
{ "stack", TSTACK, '\0', 0, 5},
{ "sub", TRSUB, '\0', TGPOWER, 0},
{ "subset", TSUBSET, MS_SUBSET, TGRELATION, 0},
{ "succ", TSUCCEEDS, MS_SUCCEEDS, TGRELATION, 0 },
{ "succcurlyeq", TSUCCEEDSEQUAL, MS_SUCCEEDSEQUAL, TGRELATION, 0 },
{ "succsim", TSUCCEEDSEQUIV, MS_SUCCEEDSEQUIV, TGRELATION, 0 },
{ "nsucc", TNOTSUCCEEDS, MS_NOTSUCCEEDS, TGRELATION, 0 },
{ "subseteq", TSUBSETEQ, MS_SUBSETEQ, TGRELATION, 0},
{ "sum", TSUM, MS_SUM, TGOPER, 5},
{ "sup", TRSUP, '\0', TGPOWER, 0},
{ "supset", TSUPSET, MS_SUPSET, TGRELATION, 0},
{ "supseteq", TSUPSETEQ, MS_SUPSETEQ, TGRELATION, 0},
{ "tan", TTAN, '\0', TGFUNCTION, 5},
{ "tanh", TTANH, '\0', TGFUNCTION, 5},
{ "tilde", TTILDE, MS_TILDE, TGATTRIBUT, 5},
{ "times", TTIMES, MS_TIMES, TGPRODUCT, 0},
{ "to", TTO, '\0', TGLIMIT, 0},
{ "toward", TTOWARD, MS_RIGHTARROW, TGRELATION, 0},
{ "transl", TTRANSL, MS_TRANSL, TGRELATION, 0},
{ "transr", TTRANSR, MS_TRANSR, TGRELATION, 0},
{ "underbrace", TUNDERBRACE, MS_UNDERBRACE, TGPRODUCT, 5},
{ "underline", TUNDERLINE, '\0', TGATTRIBUT, 5},
{ "union", TUNION, MS_UNION, TGSUM, 0},
{ "uoper", TUOPER, '\0', TGUNOPER, 5},
{ "uparrow" , TUPARROW, MS_UPARROW, TGSTANDALONE, 5},
{ "vec", TVEC, MS_VEC, TGATTRIBUT, 5},
{ "white", TWHITE, '\0', TGCOLOR, 0},
{ "widebslash", TWIDEBACKSLASH, MS_BACKSLASH, TGPRODUCT, 0 },
{ "widehat", TWIDEHAT, MS_HAT, TGATTRIBUT, 5},
{ "widetilde", TWIDETILDE, MS_TILDE, TGATTRIBUT, 5},
{ "wideslash", TWIDESLASH, MS_SLASH, TGPRODUCT, 0 },
{ "widevec", TWIDEVEC, MS_VEC, TGATTRIBUT, 5},
{ "wp" , TWP, MS_WP, TGSTANDALONE, 5},
{ "yellow", TYELLOW, '\0', TGCOLOR, 0},
{ "", TEND, '\0', 0, 0}
};
const SmTokenTableEntry * SmParser::GetTokenTableEntry( const String &rName )
{
const SmTokenTableEntry * pRes = 0;
if (rName.Len())
{
sal_Int32 nEntries = SAL_N_ELEMENTS(aTokenTable);
for (sal_Int32 i = 0; i < nEntries; ++i)
{
if (rName.EqualsIgnoreCaseAscii( aTokenTable[i].pIdent ))
{
pRes = &aTokenTable[i];
break;
}
}
}
return pRes;
}
///////////////////////////////////////////////////////////////////////////
#if OSL_DEBUG_LEVEL > 1
bool SmParser::IsDelimiter( const String &rTxt, xub_StrLen nPos )
// returns 'true' iff cChar is '\0' or a delimeter
{
OSL_ENSURE( nPos <= rTxt.Len(), "index out of range" );
sal_Unicode cChar = rTxt.GetChar( nPos );
if(!cChar)
return true;
// check if 'cChar' is in the delimeter table
const sal_Unicode *pDelim = &aDelimiterTable[0];
for ( ; *pDelim != 0; pDelim++)
if (*pDelim == cChar)
break;
sal_Int16 nTypJp = SM_MOD()->GetSysLocale().GetCharClass().getType( rTxt, nPos );
bool bIsDelim = (*pDelim != 0 ||
nTypJp == com::sun::star::i18n::UnicodeType::SPACE_SEPARATOR ||
nTypJp == com::sun::star::i18n::UnicodeType::CONTROL);
return bIsDelim;
}
#endif
void SmParser::Insert(const String &rText, sal_uInt16 nPos)
{
m_aBufferString.Insert(rText, nPos);
xub_StrLen nLen = rText.Len();
m_nBufferIndex = m_nBufferIndex + nLen;
m_nTokenIndex = m_nTokenIndex + nLen;
}
void SmParser::Replace( sal_uInt16 nPos, sal_uInt16 nLen, const String &rText )
{
OSL_ENSURE( nPos + nLen <= m_aBufferString.Len(), "argument mismatch" );
m_aBufferString.Replace( nPos, nLen, rText );
sal_Int16 nChg = rText.Len() - nLen;
m_nBufferIndex = m_nBufferIndex + nChg;
m_nTokenIndex = m_nTokenIndex + nChg;
}
// First character may be any alphabetic
const sal_Int32 coStartFlags =
KParseTokens::ANY_LETTER_OR_NUMBER |
KParseTokens::IGNORE_LEADING_WS;
// Continuing characters may be any alphanumeric or dot.
const sal_Int32 coContFlags =
((coStartFlags | KParseTokens::ASC_DOT) & ~KParseTokens::IGNORE_LEADING_WS)
| KParseTokens::TWO_DOUBLE_QUOTES_BREAK_STRING;
// First character for numbers, may be any numeric or dot
const sal_Int32 coNumStartFlags =
KParseTokens::ASC_DIGIT |
KParseTokens::ASC_DOT |
KParseTokens::IGNORE_LEADING_WS;
// Continuing characters for numbers, may be any numeric or dot.
const sal_Int32 coNumContFlags =
(coNumStartFlags | KParseTokens::ASC_DOT) & ~KParseTokens::IGNORE_LEADING_WS;
void SmParser::NextToken()
{
static const String aEmptyStr;
xub_StrLen nBufLen = m_aBufferString.Len();
ParseResult aRes;
xub_StrLen nRealStart;
bool bCont;
bool bNumStart = false;
CharClass aCC(SM_MOD()->GetSysLocale().GetCharClass().getLocale());
do
{
// skip white spaces
while (UnicodeType::SPACE_SEPARATOR ==
aCC.getType( m_aBufferString, m_nBufferIndex ))
++m_nBufferIndex;
sal_Int32 nStartFlags = coStartFlags;
sal_Int32 nContFlags = coContFlags;
sal_Unicode cFirstChar = m_aBufferString.GetChar( m_nBufferIndex );
aRes = aCC.parseAnyToken( m_aBufferString, m_nBufferIndex,
nStartFlags, aEmptyStr,
nContFlags, aEmptyStr );
// #i45779# parse numbers correctly
// i.e. independent from the locale setting.
// (note that #i11752# remains fixed)
if ((aRes.TokenType & KParseType::IDENTNAME) && IsDigit( cFirstChar ))
{
ParseResult aTmpRes;
lang::Locale aOldLoc( aCC.getLocale() );
aCC.setLocale( m_aDotLoc );
aTmpRes = aCC.parsePredefinedToken(
KParseType::ASC_NUMBER,
m_aBufferString, m_nBufferIndex,
KParseTokens::ASC_DIGIT, aEmptyStr,
KParseTokens::ASC_DIGIT | KParseTokens::ASC_DOT, aEmptyStr );
aCC.setLocale( aOldLoc );
if (aTmpRes.TokenType & KParseType::ASC_NUMBER)
aRes.TokenType = aTmpRes.TokenType;
}
nRealStart = m_nBufferIndex + sal::static_int_cast< xub_StrLen >(aRes.LeadingWhiteSpace);
m_nBufferIndex = nRealStart;
bCont = false;
if ( aRes.TokenType == 0 &&
nRealStart < nBufLen &&
'\n' == m_aBufferString.GetChar( nRealStart ) )
{
// keep data needed for tokens row and col entry up to date
++m_Row;
m_nBufferIndex = m_nColOff = nRealStart + 1;
bCont = true;
}
else if (aRes.TokenType & KParseType::ONE_SINGLE_CHAR)
{
String aName( m_aBufferString.Copy( nRealStart, 2 ));
if ( aName.EqualsAscii( "%%" ))
{
//SkipComment
m_nBufferIndex = nRealStart + 2;
while (m_nBufferIndex < nBufLen &&
'\n' != m_aBufferString.GetChar( m_nBufferIndex ))
++m_nBufferIndex;
bCont = true;
}
}
} while (bCont);
// set index of current token
m_nTokenIndex = m_nBufferIndex;
m_aCurToken.nRow = m_Row;
m_aCurToken.nCol = nRealStart - m_nColOff + 1;
bool bHandled = true;
if (nRealStart >= nBufLen)
{
m_aCurToken.eType = TEND;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.Erase();
}
else if ((aRes.TokenType & (KParseType::ASC_NUMBER | KParseType::UNI_NUMBER))
|| (bNumStart && (aRes.TokenType & KParseType::IDENTNAME)))
{
sal_Int32 n = aRes.EndPos - nRealStart;
OSL_ENSURE( n >= 0, "length < 0" );
m_aCurToken.eType = TNUMBER;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText = m_aBufferString.Copy( nRealStart, sal::static_int_cast< xub_StrLen >(n) );
#if OSL_DEBUG_LEVEL > 1
if (!IsDelimiter( m_aBufferString, static_cast< xub_StrLen >(aRes.EndPos) ))
{
OSL_FAIL( "identifier really finished? (compatibility!)" );
}
#endif
}
else if (aRes.TokenType & KParseType::DOUBLE_QUOTE_STRING)
{
m_aCurToken.eType = TTEXT;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText = aRes.DequotedNameOrString;
m_aCurToken.nRow = m_Row;
m_aCurToken.nCol = nRealStart - m_nColOff + 2;
}
else if (aRes.TokenType & KParseType::IDENTNAME)
{
sal_Int32 n = aRes.EndPos - nRealStart;
OSL_ENSURE( n >= 0, "length < 0" );
String aName( m_aBufferString.Copy( nRealStart, sal::static_int_cast< xub_StrLen >(n) ) );
const SmTokenTableEntry *pEntry = GetTokenTableEntry( aName );
if (pEntry)
{
m_aCurToken.eType = pEntry->eType;
m_aCurToken.cMathChar = pEntry->cMathChar;
m_aCurToken.nGroup = pEntry->nGroup;
m_aCurToken.nLevel = pEntry->nLevel;
m_aCurToken.aText.AssignAscii( pEntry->pIdent );
}
else
{
m_aCurToken.eType = TIDENT;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText = aName;
#if OSL_DEBUG_LEVEL > 1
if (!IsDelimiter( m_aBufferString, static_cast< xub_StrLen >(aRes.EndPos) ))
{
OSL_FAIL( "identifier really finished? (compatibility!)" );
}
#endif
}
}
else if (aRes.TokenType == 0 && '_' == m_aBufferString.GetChar( nRealStart ))
{
m_aCurToken.eType = TRSUB;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = TGPOWER;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "_" );
aRes.EndPos = nRealStart + 1;
}
else if (aRes.TokenType & KParseType::BOOLEAN)
{
sal_Int32 &rnEndPos = aRes.EndPos;
String aName( m_aBufferString.Copy( nRealStart,
sal::static_int_cast< xub_StrLen >(rnEndPos - nRealStart) ));
if (2 >= aName.Len())
{
sal_Unicode ch = aName.GetChar( 0 );
switch (ch)
{
case '<':
{
if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( "<<" ))
{
m_aCurToken.eType = TLL;
m_aCurToken.cMathChar = MS_LL;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "<<" );
rnEndPos = nRealStart + 2;
}
else if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( "<=" ))
{
m_aCurToken.eType = TLE;
m_aCurToken.cMathChar = MS_LE;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "<=" );
rnEndPos = nRealStart + 2;
}
else if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( "<>" ))
{
m_aCurToken.eType = TNEQ;
m_aCurToken.cMathChar = MS_NEQ;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "<>" );
rnEndPos = nRealStart + 2;
}
else if (m_aBufferString.Copy( nRealStart, 3 ).
EqualsAscii( "<?>" ))
{
m_aCurToken.eType = TPLACE;
m_aCurToken.cMathChar = MS_PLACE;
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "<?>" );
rnEndPos = nRealStart + 3;
}
else
{
m_aCurToken.eType = TLT;
m_aCurToken.cMathChar = MS_LT;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "<" );
}
}
break;
case '>':
{
if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( ">=" ))
{
m_aCurToken.eType = TGE;
m_aCurToken.cMathChar = MS_GE;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( ">=" );
rnEndPos = nRealStart + 2;
}
else if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( ">>" ))
{
m_aCurToken.eType = TGG;
m_aCurToken.cMathChar = MS_GG;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( ">>" );
rnEndPos = nRealStart + 2;
}
else
{
m_aCurToken.eType = TGT;
m_aCurToken.cMathChar = MS_GT;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( ">" );
}
}
break;
default:
bHandled = false;
}
}
}
else if (aRes.TokenType & KParseType::ONE_SINGLE_CHAR)
{
sal_Int32 &rnEndPos = aRes.EndPos;
String aName( m_aBufferString.Copy( nRealStart,
sal::static_int_cast< xub_StrLen >(rnEndPos - nRealStart) ) );
if (1 == aName.Len())
{
sal_Unicode ch = aName.GetChar( 0 );
switch (ch)
{
case '%':
{
//! modifies aRes.EndPos
OSL_ENSURE( rnEndPos >= nBufLen ||
'%' != m_aBufferString.GetChar( sal::static_int_cast< xub_StrLen >(rnEndPos) ),
"unexpected comment start" );
// get identifier of user-defined character
ParseResult aTmpRes = aCC.parseAnyToken(
m_aBufferString, rnEndPos,
KParseTokens::ANY_LETTER,
aEmptyStr,
coContFlags,
aEmptyStr );
xub_StrLen nTmpStart = sal::static_int_cast< xub_StrLen >(rnEndPos +
aTmpRes.LeadingWhiteSpace);
// default setting for the case that no identifier
// i.e. a valid symbol-name is following the '%'
// character
m_aCurToken.eType = TTEXT;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText = String();
m_aCurToken.nRow = sal::static_int_cast< xub_StrLen >(m_Row);
m_aCurToken.nCol = nTmpStart - m_nColOff;
if (aTmpRes.TokenType & KParseType::IDENTNAME)
{
xub_StrLen n = sal::static_int_cast< xub_StrLen >(aTmpRes.EndPos - nTmpStart);
m_aCurToken.eType = TSPECIAL;
m_aCurToken.aText = m_aBufferString.Copy( sal::static_int_cast< xub_StrLen >(nTmpStart-1), n+1 );
OSL_ENSURE( aTmpRes.EndPos > rnEndPos,
"empty identifier" );
if (aTmpRes.EndPos > rnEndPos)
rnEndPos = aTmpRes.EndPos;
else
++rnEndPos;
}
// if no symbol-name was found we start-over with
// finding the next token right afer the '%' sign.
// I.e. we leave rnEndPos unmodified.
}
break;
case '[':
{
m_aCurToken.eType = TLBRACKET;
m_aCurToken.cMathChar = MS_LBRACKET;
m_aCurToken.nGroup = TGLBRACES;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "[" );
}
break;
case '\\':
{
m_aCurToken.eType = TESCAPE;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "\\" );
}
break;
case ']':
{
m_aCurToken.eType = TRBRACKET;
m_aCurToken.cMathChar = MS_RBRACKET;
m_aCurToken.nGroup = TGRBRACES;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "]" );
}
break;
case '^':
{
m_aCurToken.eType = TRSUP;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = TGPOWER;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "^" );
}
break;
case '`':
{
m_aCurToken.eType = TSBLANK;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = TGBLANK;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "`" );
}
break;
case '{':
{
m_aCurToken.eType = TLGROUP;
m_aCurToken.cMathChar = MS_LBRACE;
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "{" );
}
break;
case '|':
{
m_aCurToken.eType = TOR;
m_aCurToken.cMathChar = MS_OR;
m_aCurToken.nGroup = TGSUM;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "|" );
}
break;
case '}':
{
m_aCurToken.eType = TRGROUP;
m_aCurToken.cMathChar = MS_RBRACE;
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "}" );
}
break;
case '~':
{
m_aCurToken.eType = TBLANK;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = TGBLANK;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "~" );
}
break;
case '#':
{
if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( "##" ))
{
m_aCurToken.eType = TDPOUND;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "##" );
rnEndPos = nRealStart + 2;
}
else
{
m_aCurToken.eType = TPOUND;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "#" );
}
}
break;
case '&':
{
m_aCurToken.eType = TAND;
m_aCurToken.cMathChar = MS_AND;
m_aCurToken.nGroup = TGPRODUCT;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "&" );
}
break;
case '(':
{
m_aCurToken.eType = TLPARENT;
m_aCurToken.cMathChar = MS_LPARENT;
m_aCurToken.nGroup = TGLBRACES;
m_aCurToken.nLevel = 5; //! 0 to continue expression
m_aCurToken.aText.AssignAscii( "(" );
}
break;
case ')':
{
m_aCurToken.eType = TRPARENT;
m_aCurToken.cMathChar = MS_RPARENT;
m_aCurToken.nGroup = TGRBRACES;
m_aCurToken.nLevel = 0; //! 0 to terminate expression
m_aCurToken.aText.AssignAscii( ")" );
}
break;
case '*':
{
m_aCurToken.eType = TMULTIPLY;
m_aCurToken.cMathChar = MS_MULTIPLY;
m_aCurToken.nGroup = TGPRODUCT;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "*" );
}
break;
case '+':
{
if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( "+-" ))
{
m_aCurToken.eType = TPLUSMINUS;
m_aCurToken.cMathChar = MS_PLUSMINUS;
m_aCurToken.nGroup = TGUNOPER | TGSUM;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "+-" );
rnEndPos = nRealStart + 2;
}
else
{
m_aCurToken.eType = TPLUS;
m_aCurToken.cMathChar = MS_PLUS;
m_aCurToken.nGroup = TGUNOPER | TGSUM;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "+" );
}
}
break;
case '-':
{
if (m_aBufferString.Copy( nRealStart, 2 ).
EqualsAscii( "-+" ))
{
m_aCurToken.eType = TMINUSPLUS;
m_aCurToken.cMathChar = MS_MINUSPLUS;
m_aCurToken.nGroup = TGUNOPER | TGSUM;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "-+" );
rnEndPos = nRealStart + 2;
}
else
{
m_aCurToken.eType = TMINUS;
m_aCurToken.cMathChar = MS_MINUS;
m_aCurToken.nGroup = TGUNOPER | TGSUM;
m_aCurToken.nLevel = 5;
m_aCurToken.aText.AssignAscii( "-" );
}
}
break;
case '.':
{
// for compatibility with SO5.2
// texts like .34 ...56 ... h ...78..90
// will be treated as numbers
m_aCurToken.eType = TNUMBER;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
xub_StrLen nTxtStart = m_nBufferIndex;
sal_Unicode cChar;
do
{
cChar = m_aBufferString.GetChar( ++m_nBufferIndex );
}
while ( cChar == '.' || IsDigit( cChar ) );
m_aCurToken.aText = m_aBufferString.Copy( sal::static_int_cast< xub_StrLen >(nTxtStart),
sal::static_int_cast< xub_StrLen >(m_nBufferIndex - nTxtStart) );
aRes.EndPos = m_nBufferIndex;
}
break;
case '/':
{
m_aCurToken.eType = TDIVIDEBY;
m_aCurToken.cMathChar = MS_SLASH;
m_aCurToken.nGroup = TGPRODUCT;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "/" );
}
break;
case '=':
{
m_aCurToken.eType = TASSIGN;
m_aCurToken.cMathChar = MS_ASSIGN;
m_aCurToken.nGroup = TGRELATION;
m_aCurToken.nLevel = 0;
m_aCurToken.aText.AssignAscii( "=" );
}
break;
default:
bHandled = false;
}
}
}
else
bHandled = false;
if (!bHandled)
{
m_aCurToken.eType = TCHARACTER;
m_aCurToken.cMathChar = '\0';
m_aCurToken.nGroup = 0;
m_aCurToken.nLevel = 5;
m_aCurToken.aText = m_aBufferString.Copy( nRealStart, 1 );
aRes.EndPos = nRealStart + 1;
}
if (TEND != m_aCurToken.eType)
m_nBufferIndex = sal::static_int_cast< xub_StrLen >(aRes.EndPos);
}
////////////////////////////////////////
// grammar
//
void SmParser::Table()
{
SmNodeArray LineArray;
Line();
while (m_aCurToken.eType == TNEWLINE)
{
NextToken();
Line();
}
if (m_aCurToken.eType != TEND)
Error(PE_UNEXPECTED_CHAR);
sal_uLong n = m_aNodeStack.Count();
LineArray.resize(n);
for (sal_uLong i = 0; i < n; i++)
LineArray[n - (i + 1)] = m_aNodeStack.Pop();
SmStructureNode *pSNode = new SmTableNode(m_aCurToken);
pSNode->SetSubNodes(LineArray);
m_aNodeStack.Push(pSNode);
}
void SmParser::Align()
// parse alignment info (if any), then go on with rest of expression
{
SmStructureNode *pSNode = 0;
bool bNeedGroupClose = false;
if (TokenInGroup(TGALIGN))
{
if (CONVERT_40_TO_50 == GetConversion())
// encapsulate expression to be aligned in group braces
// (here group-open brace)
{ Insert('{', GetTokenIndex());
bNeedGroupClose = true;
// get first valid align statement in sequence
// (the dominant one in 4.0) and erase all others (especially old
// discarded tokens) from command string.
while (TokenInGroup(TGALIGN))
{
if (TokenInGroup(TGDISCARDED) || pSNode)
{
m_nBufferIndex = GetTokenIndex();
m_aBufferString.Erase(m_nBufferIndex, m_aCurToken.aText.Len());
}
else
pSNode = new SmAlignNode(m_aCurToken);
NextToken();
}
}
else
{
pSNode = new SmAlignNode(m_aCurToken);
NextToken();
// allow for just one align statement in 5.0
if (CONVERT_40_TO_50 != GetConversion() && TokenInGroup(TGALIGN))
{ Error(PE_DOUBLE_ALIGN);
return;
}
}
}
Expression();
if (bNeedGroupClose)
Insert('}', GetTokenIndex());
if (pSNode)
{ pSNode->SetSubNodes(m_aNodeStack.Pop(), 0);
m_aNodeStack.Push(pSNode);
}
}
void SmParser::Line()
{
sal_uInt16 n = 0;
SmNodeArray ExpressionArray;
ExpressionArray.resize(n);
// start with single expression that may have an alignment statement
// (and go on with expressions that must not have alignment
// statements in 'while' loop below. See also 'Expression()'.)
if (m_aCurToken.eType != TEND && m_aCurToken.eType != TNEWLINE)
{ Align();
ExpressionArray.resize(++n);
ExpressionArray[n - 1] = m_aNodeStack.Pop();
}
while (m_aCurToken.eType != TEND && m_aCurToken.eType != TNEWLINE)
{ if (CONVERT_40_TO_50 != GetConversion())
Expression();
else
Align();
ExpressionArray.resize(++n);
ExpressionArray[n - 1] = m_aNodeStack.Pop();
}
//If there's no expression, add an empty one.
//this is to avoid a formula tree without any caret
//positions, in visual formula editor.
if(ExpressionArray.size() == 0)
ExpressionArray.push_back(new SmExpressionNode(SmToken()));
SmStructureNode *pSNode = new SmLineNode(m_aCurToken);
pSNode->SetSubNodes(ExpressionArray);
m_aNodeStack.Push(pSNode);
}
void SmParser::Expression()
{
bool bUseExtraSpaces = true;
SmNode *pNode = m_aNodeStack.Pop();
if (pNode)
{
if (pNode->GetToken().eType == TNOSPACE)
bUseExtraSpaces = false;
else
m_aNodeStack.Push(pNode); // push the node from above again (now to be used as argument to this current 'nospace' node)
}
sal_uInt16 n = 0;
SmNodeArray RelationArray;
RelationArray.resize(n);
Relation();
RelationArray.resize(++n);
RelationArray[n - 1] = m_aNodeStack.Pop();
while (m_aCurToken.nLevel >= 4)
{ Relation();
RelationArray.resize(++n);
RelationArray[n - 1] = m_aNodeStack.Pop();
}
SmExpressionNode *pSNode = new SmExpressionNode(m_aCurToken);
pSNode->SetSubNodes(RelationArray);
pSNode->SetUseExtraSpaces(bUseExtraSpaces);
m_aNodeStack.Push(pSNode);
}
void SmParser::Relation()
{
Sum();
while (TokenInGroup(TGRELATION))
{
SmStructureNode *pSNode = new SmBinHorNode(m_aCurToken);
SmNode *pFirst = m_aNodeStack.Pop();
OpSubSup();
SmNode *pSecond = m_aNodeStack.Pop();
Sum();
pSNode->SetSubNodes(pFirst, pSecond, m_aNodeStack.Pop());
m_aNodeStack.Push(pSNode);
}
}
void SmParser::Sum()
{
Product();
while (TokenInGroup(TGSUM))
{
SmStructureNode *pSNode = new SmBinHorNode(m_aCurToken);
SmNode *pFirst = m_aNodeStack.Pop();
OpSubSup();
SmNode *pSecond = m_aNodeStack.Pop();
Product();
pSNode->SetSubNodes(pFirst, pSecond, m_aNodeStack.Pop());
m_aNodeStack.Push(pSNode);
}
}
void SmParser::Product()
{
Power();
while (TokenInGroup(TGPRODUCT))
{ SmStructureNode *pSNode;
SmNode *pFirst = m_aNodeStack.Pop(),
*pOper;
bool bSwitchArgs = false;
SmTokenType eType = m_aCurToken.eType;
switch (eType)
{
case TOVER:
pSNode = new SmBinVerNode(m_aCurToken);
pOper = new SmRectangleNode(m_aCurToken);
NextToken();
break;
case TBOPER:
pSNode = new SmBinHorNode(m_aCurToken);
NextToken();
//Let the glyph node know it's a binary operation
m_aCurToken.eType = TBOPER;
m_aCurToken.nGroup = TGPRODUCT;
GlyphSpecial();
pOper = m_aNodeStack.Pop();
break;
case TOVERBRACE :
case TUNDERBRACE :
pSNode = new SmVerticalBraceNode(m_aCurToken);
pOper = new SmMathSymbolNode(m_aCurToken);
NextToken();
break;
case TWIDEBACKSLASH:
case TWIDESLASH:
{
SmBinDiagonalNode *pSTmp = new SmBinDiagonalNode(m_aCurToken);
pSTmp->SetAscending(eType == TWIDESLASH);
pSNode = pSTmp;
pOper = new SmPolyLineNode(m_aCurToken);
NextToken();
bSwitchArgs = true;
break;
}
default:
pSNode = new SmBinHorNode(m_aCurToken);
OpSubSup();
pOper = m_aNodeStack.Pop();
}
Power();
if (bSwitchArgs)
//! vgl siehe SmBinDiagonalNode::Arrange
pSNode->SetSubNodes(pFirst, m_aNodeStack.Pop(), pOper);
else
pSNode->SetSubNodes(pFirst, pOper, m_aNodeStack.Pop());
m_aNodeStack.Push(pSNode);
}
}
void SmParser::SubSup(sal_uLong nActiveGroup)
{
OSL_ENSURE(nActiveGroup == TGPOWER || nActiveGroup == TGLIMIT,
"Sm: wrong token group");
if (!TokenInGroup(nActiveGroup))
// already finish
return;
SmSubSupNode *pNode = new SmSubSupNode(m_aCurToken);
//! Of course 'm_aCurToken' is just the first sub-/supscript token.
//! It should be of no further interest. The positions of the
//! sub-/supscripts will be identified by the corresponding subnodes
//! index in the 'aSubNodes' array (enum value from 'SmSubSup').
pNode->SetUseLimits(nActiveGroup == TGLIMIT);
// initialize subnodes array
SmNodeArray aSubNodes;
aSubNodes.resize(1 + SUBSUP_NUM_ENTRIES);
aSubNodes[0] = m_aNodeStack.Pop();
for (sal_uInt16 i = 1; i < aSubNodes.size(); i++)
aSubNodes[i] = NULL;
// process all sub-/supscripts
int nIndex = 0;
while (TokenInGroup(nActiveGroup))
{ SmTokenType eType (m_aCurToken.eType);
// skip sub-/supscript token
NextToken();
// get sub-/supscript node on top of stack
if (eType == TFROM || eType == TTO)
{
// parse limits in old 4.0 and 5.0 style
Relation();
}
else
Term();
switch (eType)
{ case TRSUB : nIndex = (int) RSUB; break;
case TRSUP : nIndex = (int) RSUP; break;
case TFROM :
case TCSUB : nIndex = (int) CSUB; break;
case TTO :
case TCSUP : nIndex = (int) CSUP; break;
case TLSUB : nIndex = (int) LSUB; break;
case TLSUP : nIndex = (int) LSUP; break;
default :
OSL_FAIL("Sm: unknown case");
}
nIndex++;
OSL_ENSURE(1 <= nIndex && nIndex <= 1 + SUBSUP_NUM_ENTRIES,
"SmParser::Power() : sub-/supscript index falsch");
// set sub-/supscript if not already done
if (aSubNodes[nIndex] != NULL)
Error(PE_DOUBLE_SUBSUPSCRIPT);
aSubNodes[nIndex] = m_aNodeStack.Pop();
}
pNode->SetSubNodes(aSubNodes);
m_aNodeStack.Push(pNode);
}
void SmParser::OpSubSup()
{
// push operator symbol
m_aNodeStack.Push(new SmMathSymbolNode(m_aCurToken));
// skip operator token
NextToken();
// get sub- supscripts if any
if (TokenInGroup(TGPOWER))
SubSup(TGPOWER);
}
void SmParser::Power()
{
// get body for sub- supscripts on top of stack
Term();
SubSup(TGPOWER);
}
void SmParser::Blank()
{
OSL_ENSURE(TokenInGroup(TGBLANK), "Sm : wrong token");
SmBlankNode *pBlankNode = new SmBlankNode(m_aCurToken);
while (TokenInGroup(TGBLANK))
{
pBlankNode->IncreaseBy(m_aCurToken);
NextToken();
}
// Blanks am Zeilenende ignorieren wenn die entsprechende Option gesetzt ist
if ( m_aCurToken.eType == TNEWLINE ||
(m_aCurToken.eType == TEND && SM_MOD()->GetConfig()->IsIgnoreSpacesRight()) )
{
pBlankNode->Clear();
}
m_aNodeStack.Push(pBlankNode);
}
void SmParser::Term()
{
switch (m_aCurToken.eType)
{
case TESCAPE :
Escape();
break;
case TNOSPACE :
case TLGROUP :
{
bool bNoSpace = m_aCurToken.eType == TNOSPACE;
if (bNoSpace) // push 'no space' node and continue to parse expression
{
m_aNodeStack.Push(new SmExpressionNode(m_aCurToken));
NextToken();
}
if (m_aCurToken.eType != TLGROUP)
{
m_aNodeStack.Pop(); // get rid of the 'no space' node pushed above
Term();
}
else
{
NextToken();
// allow for empty group
if (m_aCurToken.eType == TRGROUP)
{
if (bNoSpace) // get rid of the 'no space' node pushed above
m_aNodeStack.Pop();
SmStructureNode *pSNode = new SmExpressionNode(m_aCurToken);
pSNode->SetSubNodes(NULL, NULL);
m_aNodeStack.Push(pSNode);
NextToken();
}
else // go as usual
{
Align();
if (m_aCurToken.eType != TRGROUP)
Error(PE_RGROUP_EXPECTED);
else
NextToken();
}
}
}
break;
case TLEFT :
Brace();
break;
case TBLANK :
case TSBLANK :
Blank();
break;
case TTEXT :
m_aNodeStack.Push(new SmTextNode(m_aCurToken, FNT_TEXT));
NextToken();
break;
case TIDENT :
case TCHARACTER :
m_aNodeStack.Push(new SmTextNode(m_aCurToken, FNT_VARIABLE));
NextToken();
break;
case TNUMBER :
m_aNodeStack.Push(new SmTextNode(m_aCurToken, FNT_NUMBER));
NextToken();
break;
case TLEFTARROW :
case TRIGHTARROW :
case TUPARROW :
case TDOWNARROW :
case TSETN :
case TSETZ :
case TSETQ :
case TSETR :
case TSETC :
case THBAR :
case TLAMBDABAR :
case TCIRC :
case TDRARROW :
case TDLARROW :
case TDLRARROW :
case TBACKEPSILON :
case TALEPH :
case TIM :
case TRE :
case TWP :
case TEMPTYSET :
case TINFINITY :
case TEXISTS :
case TNOTEXISTS :
case TFORALL :
case TPARTIAL :
case TNABLA :
case TTOWARD :
case TDOTSAXIS :
case TDOTSDIAG :
case TDOTSDOWN :
case TDOTSLOW :
case TDOTSUP :
case TDOTSVERT :
m_aNodeStack.Push(new SmMathSymbolNode(m_aCurToken));
NextToken();
break;
case TPLACE:
m_aNodeStack.Push(new SmPlaceNode(m_aCurToken));
NextToken();
break;
case TSPECIAL:
Special();
break;
case TBINOM:
Binom();
break;
case TSTACK:
Stack();
break;
case TMATRIX:
Matrix();
break;
default:
if (TokenInGroup(TGLBRACES))
{ Brace();
}
else if (TokenInGroup(TGOPER))
{ Operator();
}
else if (TokenInGroup(TGUNOPER))
{ UnOper();
}
else if ( TokenInGroup(TGATTRIBUT)
|| TokenInGroup(TGFONTATTR))
{ SmStructureNodeArray aArray;
bool bIsAttr;
sal_uInt16 n = 0;
while (true == (bIsAttr = TokenInGroup(TGATTRIBUT))
|| TokenInGroup(TGFONTATTR))
{ aArray.resize(n + 1);
if (bIsAttr)
Attribut();
else
FontAttribut();
// check if casting in following line is ok
OSL_ENSURE(!m_aNodeStack.Top()->IsVisible(), "Sm : Ooops...");
aArray[n] = (SmStructureNode *) m_aNodeStack.Pop();
n++;
}
Power();
SmNode *pFirstNode = m_aNodeStack.Pop();
while (n > 0)
{ aArray[n - 1]->SetSubNodes(0, pFirstNode);
pFirstNode = aArray[n - 1];
n--;
}
m_aNodeStack.Push(pFirstNode);
}
else if (TokenInGroup(TGFUNCTION))
{ if (CONVERT_40_TO_50 != GetConversion())
{ Function();
}
else // encapsulate old 4.0 style parsing in braces
{
// insert opening brace
Insert('{', GetTokenIndex());
//
// parse in 4.0 style
//
Function();
SmNode *pFunc = m_aNodeStack.Pop();
if (m_aCurToken.eType == TLPARENT)
{ Term();
}
else
{ Align();
}
// insert closing brace
Insert('}', GetTokenIndex());
SmStructureNode *pSNode = new SmExpressionNode(pFunc->GetToken());
pSNode->SetSubNodes(pFunc, m_aNodeStack.Pop());
m_aNodeStack.Push(pSNode);
}
}
else
Error(PE_UNEXPECTED_CHAR);
}
}
void SmParser::Escape()
{
NextToken();
switch (m_aCurToken.eType)
{
case TLPARENT :
case TRPARENT :
case TLBRACKET :
case TRBRACKET :
case TLDBRACKET :
case TRDBRACKET :
case TLBRACE :
case TLGROUP :
case TRBRACE :
case TRGROUP :
case TLANGLE :
case TRANGLE :
case TLCEIL :
case TRCEIL :
case TLFLOOR :
case TRFLOOR :
case TLLINE :
case TRLINE :
case TLDLINE :
case TRDLINE :
break;
default:
Error(PE_UNEXPECTED_TOKEN);
}
SmNode *pNode = new SmMathSymbolNode(m_aCurToken);
m_aNodeStack.Push(pNode);
NextToken();
}
void SmParser::Operator()
{
if (TokenInGroup(TGOPER))
{ SmStructureNode *pSNode = new SmOperNode(m_aCurToken);
// put operator on top of stack
Oper();
if (TokenInGroup(TGLIMIT) || TokenInGroup(TGPOWER))
SubSup(m_aCurToken.nGroup);
SmNode *pOperator = m_aNodeStack.Pop();
// get argument
Power();
pSNode->SetSubNodes(pOperator, m_aNodeStack.Pop());
m_aNodeStack.Push(pSNode);
}
}
void SmParser::Oper()
{
SmTokenType eType (m_aCurToken.eType);
SmNode *pNode = NULL;
switch (eType)
{
case TSUM :
case TPROD :
case TCOPROD :
case TINT :
case TIINT :
case TIIINT :
case TLINT :
case TLLINT :
case TLLLINT :
pNode = new SmMathSymbolNode(m_aCurToken);
break;
case TLIM :
case TLIMSUP :
case TLIMINF :
{
const sal_Char* pLim = 0;
switch (eType)
{
case TLIM : pLim = "lim"; break;
case TLIMSUP : pLim = "lim sup"; break;
case TLIMINF : pLim = "lim inf"; break;
default:
break;
}
if( pLim )
m_aCurToken.aText.AssignAscii( pLim );
pNode = new SmTextNode(m_aCurToken, FNT_TEXT);
}
break;
case TOVERBRACE :
case TUNDERBRACE :
pNode = new SmMathSymbolNode(m_aCurToken);
break;
case TOPER :
NextToken();
OSL_ENSURE(m_aCurToken.eType == TSPECIAL, "Sm: wrong token");
pNode = new SmGlyphSpecialNode(m_aCurToken);
break;
default :
OSL_FAIL("Sm: unknown case");
}
m_aNodeStack.Push(pNode);
NextToken();
}
void SmParser::UnOper()
{
OSL_ENSURE(TokenInGroup(TGUNOPER), "Sm: wrong token");
SmToken aNodeToken = m_aCurToken;
SmTokenType eType = m_aCurToken.eType;
bool bIsPostfix = eType == TFACT;
SmStructureNode *pSNode;
SmNode *pOper = 0,
*pExtra = 0,
*pArg;
switch (eType)
{
case TABS :
case TSQRT :
NextToken();
break;
case TNROOT :
NextToken();
Power();
pExtra = m_aNodeStack.Pop();
break;
case TUOPER :
NextToken();
//Let the glyph know what it is...
m_aCurToken.eType = TUOPER;
m_aCurToken.nGroup = TGUNOPER;
GlyphSpecial();
pOper = m_aNodeStack.Pop();
break;
case TPLUS :
case TMINUS :
case TPLUSMINUS :
case TMINUSPLUS :
case TNEG :
case TFACT :
OpSubSup();
pOper = m_aNodeStack.Pop();
break;
default :
Error(PE_UNOPER_EXPECTED);
}
// get argument
Power();
pArg = m_aNodeStack.Pop();
if (eType == TABS)
{ pSNode = new SmBraceNode(aNodeToken);
pSNode->SetScaleMode(SCALE_HEIGHT);
// build nodes for left & right lines
// (text, group, level of the used token are of no interrest here)
// we'll use row & column of the keyword for abs
aNodeToken.eType = TABS;
//
aNodeToken.cMathChar = MS_LINE;
SmNode* pLeft = new SmMathSymbolNode(aNodeToken);
//
aNodeToken.cMathChar = MS_LINE;
SmNode* pRight = new SmMathSymbolNode(aNodeToken);
pSNode->SetSubNodes(pLeft, pArg, pRight);
}
else if (eType == TSQRT || eType == TNROOT)
{ pSNode = new SmRootNode(aNodeToken);
pOper = new SmRootSymbolNode(aNodeToken);
pSNode->SetSubNodes(pExtra, pOper, pArg);
}
else
{ pSNode = new SmUnHorNode(aNodeToken);
if (bIsPostfix)
pSNode->SetSubNodes(pArg, pOper);
else
// prefix operator
pSNode->SetSubNodes(pOper, pArg);
}
m_aNodeStack.Push(pSNode);
}
void SmParser::Attribut()
{
OSL_ENSURE(TokenInGroup(TGATTRIBUT), "Sm: wrong token group");
SmStructureNode *pSNode = new SmAttributNode(m_aCurToken);
SmNode *pAttr;
SmScaleMode eScaleMode = SCALE_NONE;
// get appropriate node for the attribut itself
switch (m_aCurToken.eType)
{ case TUNDERLINE :
case TOVERLINE :
case TOVERSTRIKE :
pAttr = new SmRectangleNode(m_aCurToken);
eScaleMode = SCALE_WIDTH;
break;
case TWIDEVEC :
case TWIDEHAT :
case TWIDETILDE :
pAttr = new SmMathSymbolNode(m_aCurToken);
eScaleMode = SCALE_WIDTH;
break;
default :
pAttr = new SmMathSymbolNode(m_aCurToken);
}
NextToken();
pSNode->SetSubNodes(pAttr, 0);
pSNode->SetScaleMode(eScaleMode);
m_aNodeStack.Push(pSNode);
}
void SmParser::FontAttribut()
{
OSL_ENSURE(TokenInGroup(TGFONTATTR), "Sm: wrong token group");
switch (m_aCurToken.eType)
{
case TITALIC :
case TNITALIC :
case TBOLD :
case TNBOLD :
case TPHANTOM :
m_aNodeStack.Push(new SmFontNode(m_aCurToken));
NextToken();
break;
case TSIZE :
FontSize();
break;
case TFONT :
Font();
break;
case TCOLOR :
Color();
break;
default :
OSL_FAIL("Sm: unknown case");
}
}
void SmParser::Color()
{
OSL_ENSURE(m_aCurToken.eType == TCOLOR, "Sm : Ooops...");
// last color rules, get that one
SmToken aToken;
do
{ NextToken();
if (TokenInGroup(TGCOLOR))
{ aToken = m_aCurToken;
NextToken();
}
else
Error(PE_COLOR_EXPECTED);
} while (m_aCurToken.eType == TCOLOR);
m_aNodeStack.Push(new SmFontNode(aToken));
}
void SmParser::Font()
{
OSL_ENSURE(m_aCurToken.eType == TFONT, "Sm : Ooops...");
// last font rules, get that one
SmToken aToken;
do
{ NextToken();
if (TokenInGroup(TGFONT))
{ aToken = m_aCurToken;
NextToken();
}
else
Error(PE_FONT_EXPECTED);
} while (m_aCurToken.eType == TFONT);
m_aNodeStack.Push(new SmFontNode(aToken));
}
// gets number used as arguments in Math formulas (e.g. 'size' command)
// Format: no negative numbers, must start with a digit, no exponent notation, ...
bool lcl_IsNumber(const UniString& rText)
{
bool bPoint = false;
const sal_Unicode* pBuffer = rText.GetBuffer();
for(xub_StrLen nPos = 0; nPos < rText.Len(); nPos++, pBuffer++)
{
const sal_Unicode cChar = *pBuffer;
if(cChar == '.')
{
if(bPoint)
return false;
else
bPoint = true;
}
else if ( !IsDigit( cChar ) )
return false;
}
return true;
}
void SmParser::FontSize()
{
OSL_ENSURE(m_aCurToken.eType == TSIZE, "Sm : Ooops...");
sal_uInt16 Type;
SmFontNode *pFontNode = new SmFontNode(m_aCurToken);
NextToken();
switch (m_aCurToken.eType)
{
case TNUMBER: Type = FNTSIZ_ABSOLUT; break;
case TPLUS: Type = FNTSIZ_PLUS; break;
case TMINUS: Type = FNTSIZ_MINUS; break;
case TMULTIPLY: Type = FNTSIZ_MULTIPLY; break;
case TDIVIDEBY: Type = FNTSIZ_DIVIDE; break;
default:
delete pFontNode;
Error(PE_SIZE_EXPECTED);
return;
}
if (Type != FNTSIZ_ABSOLUT)
{
NextToken();
if (m_aCurToken.eType != TNUMBER)
{
delete pFontNode;
Error(PE_SIZE_EXPECTED);
return;
}
}
// get number argument
Fraction aValue( 1L );
if (lcl_IsNumber( m_aCurToken.aText ))
{
double fTmp;
if ((fTmp = m_aCurToken.aText.ToDouble()) != 0.0)
{
aValue = fTmp;
//!! keep the numerator and denominator from being to large
//!! otherwise ongoing multiplications may result in overflows
//!! (for example in SmNode::SetFontSize the font size calculated
//!! may become 0 because of this!!! Happens e.g. for ftmp = 2.9 with Linux
//!! or ftmp = 1.11111111111111111... (11/9) on every platform.)
if (aValue.GetDenominator() > 1000)
{
long nNum = aValue.GetNumerator();
long nDenom = aValue.GetDenominator();
while (nDenom > 1000)
{
nNum /= 10;
nDenom /= 10;
}
aValue = Fraction( nNum, nDenom );
}
}
}
NextToken();
pFontNode->SetSizeParameter(aValue, Type);
m_aNodeStack.Push(pFontNode);
}
void SmParser::Brace()
{
OSL_ENSURE(m_aCurToken.eType == TLEFT || TokenInGroup(TGLBRACES),
"Sm: kein Klammer Ausdruck");
SmStructureNode *pSNode = new SmBraceNode(m_aCurToken);
SmNode *pBody = 0,
*pLeft = 0,
*pRight = 0;
SmScaleMode eScaleMode = SCALE_NONE;
SmParseError eError = PE_NONE;
if (m_aCurToken.eType == TLEFT)
{ NextToken();
eScaleMode = SCALE_HEIGHT;
// check for left bracket
if (TokenInGroup(TGLBRACES) || TokenInGroup(TGRBRACES))
{
pLeft = new SmMathSymbolNode(m_aCurToken);
NextToken();
Bracebody(true);
pBody = m_aNodeStack.Pop();
if (m_aCurToken.eType == TRIGHT)
{ NextToken();
// check for right bracket
if (TokenInGroup(TGLBRACES) || TokenInGroup(TGRBRACES))
{
pRight = new SmMathSymbolNode(m_aCurToken);
NextToken();
}
else
eError = PE_RBRACE_EXPECTED;
}
else
eError = PE_RIGHT_EXPECTED;
}
else
eError = PE_LBRACE_EXPECTED;
}
else
{
if (TokenInGroup(TGLBRACES))
{
pLeft = new SmMathSymbolNode(m_aCurToken);
NextToken();
Bracebody(false);
pBody = m_aNodeStack.Pop();
SmTokenType eExpectedType = TUNKNOWN;
switch (pLeft->GetToken().eType)
{ case TLPARENT : eExpectedType = TRPARENT; break;
case TLBRACKET : eExpectedType = TRBRACKET; break;
case TLBRACE : eExpectedType = TRBRACE; break;
case TLDBRACKET : eExpectedType = TRDBRACKET; break;
case TLLINE : eExpectedType = TRLINE; break;
case TLDLINE : eExpectedType = TRDLINE; break;
case TLANGLE : eExpectedType = TRANGLE; break;
case TLFLOOR : eExpectedType = TRFLOOR; break;
case TLCEIL : eExpectedType = TRCEIL; break;
default :
OSL_FAIL("Sm: unknown case");
}
if (m_aCurToken.eType == eExpectedType)
{
pRight = new SmMathSymbolNode(m_aCurToken);
NextToken();
}
else
eError = PE_PARENT_MISMATCH;
}
else
eError = PE_LBRACE_EXPECTED;
}
if (eError == PE_NONE)
{ OSL_ENSURE(pLeft, "Sm: NULL pointer");
OSL_ENSURE(pRight, "Sm: NULL pointer");
pSNode->SetSubNodes(pLeft, pBody, pRight);
pSNode->SetScaleMode(eScaleMode);
m_aNodeStack.Push(pSNode);
}
else
{ delete pSNode;
delete pBody;
delete pLeft;
delete pRight;
Error(eError);
}
}
void SmParser::Bracebody(bool bIsLeftRight)
{
SmStructureNode *pBody = new SmBracebodyNode(m_aCurToken);
SmNodeArray aNodes;
sal_uInt16 nNum = 0;
// get body if any
if (bIsLeftRight)
{
do
{
if (m_aCurToken.eType == TMLINE)
{
m_aNodeStack.Push(new SmMathSymbolNode(m_aCurToken));
NextToken();
nNum++;
}
else if (m_aCurToken.eType != TRIGHT)
{ Align();
nNum++;
if (m_aCurToken.eType != TMLINE && m_aCurToken.eType != TRIGHT)
Error(PE_RIGHT_EXPECTED);
}
} while (m_aCurToken.eType != TEND && m_aCurToken.eType != TRIGHT);
}
else
{
do
{
if (m_aCurToken.eType == TMLINE)
{
m_aNodeStack.Push(new SmMathSymbolNode(m_aCurToken));
NextToken();
nNum++;
}
else if (!TokenInGroup(TGRBRACES))
{ Align();
nNum++;
if (m_aCurToken.eType != TMLINE && !TokenInGroup(TGRBRACES))
Error(PE_RBRACE_EXPECTED);
}
} while (m_aCurToken.eType != TEND && !TokenInGroup(TGRBRACES));
}
// build argument vector in parsing order
aNodes.resize(nNum);
for (sal_uInt16 i = 0; i < nNum; i++)
aNodes[nNum - 1 - i] = m_aNodeStack.Pop();
pBody->SetSubNodes(aNodes);
pBody->SetScaleMode(bIsLeftRight ? SCALE_HEIGHT : SCALE_NONE);
m_aNodeStack.Push(pBody);
}
void SmParser::Function()
{
switch (m_aCurToken.eType)
{
case TFUNC:
NextToken(); // skip "FUNC"-statement
// fall through
case TSIN :
case TCOS :
case TTAN :
case TCOT :
case TASIN :
case TACOS :
case TATAN :
case TACOT :
case TSINH :
case TCOSH :
case TTANH :
case TCOTH :
case TASINH :
case TACOSH :
case TATANH :
case TACOTH :
case TLN :
case TLOG :
case TEXP :
m_aNodeStack.Push(new SmTextNode(m_aCurToken, FNT_FUNCTION));
NextToken();
break;
default:
Error(PE_FUNC_EXPECTED);
}
}
void SmParser::Binom()
{
SmNodeArray ExpressionArray;
SmStructureNode *pSNode = new SmTableNode(m_aCurToken);
NextToken();
Sum();
Sum();
ExpressionArray.resize(2);
for (int i = 0; i < 2; i++)
ExpressionArray[2 - (i + 1)] = m_aNodeStack.Pop();
pSNode->SetSubNodes(ExpressionArray);
m_aNodeStack.Push(pSNode);
}
void SmParser::Stack()
{
SmNodeArray ExpressionArray;
NextToken();
if (m_aCurToken.eType == TLGROUP)
{
sal_uInt16 n = 0;
do
{
NextToken();
Align();
n++;
}
while (m_aCurToken.eType == TPOUND);
ExpressionArray.resize(n);
for (sal_uInt16 i = 0; i < n; i++)
ExpressionArray[n - (i + 1)] = m_aNodeStack.Pop();
if (m_aCurToken.eType != TRGROUP)
Error(PE_RGROUP_EXPECTED);
NextToken();
//We need to let the table node know it context
//it's used in SmNodeToTextVisitor
SmToken aTok = m_aCurToken;
aTok.eType = TSTACK;
SmStructureNode *pSNode = new SmTableNode(aTok);
pSNode->SetSubNodes(ExpressionArray);
m_aNodeStack.Push(pSNode);
}
else
Error(PE_LGROUP_EXPECTED);
}
void SmParser::Matrix()
{
SmNodeArray ExpressionArray;
NextToken();
if (m_aCurToken.eType == TLGROUP)
{
sal_uInt16 c = 0;
do
{
NextToken();
Align();
c++;
}
while (m_aCurToken.eType == TPOUND);
sal_uInt16 r = 1;
while (m_aCurToken.eType == TDPOUND)
{
NextToken();
for (sal_uInt16 i = 0; i < c; i++)
{
Align();
if (i < (c - 1))
{
if (m_aCurToken.eType == TPOUND)
{
NextToken();
}
else
Error(PE_POUND_EXPECTED);
}
}
r++;
}
long nRC = r * c;
ExpressionArray.resize(nRC);
for (sal_uInt16 i = 0; i < (nRC); i++)
ExpressionArray[(nRC) - (i + 1)] = m_aNodeStack.Pop();
if (m_aCurToken.eType != TRGROUP)
Error(PE_RGROUP_EXPECTED);
NextToken();
SmMatrixNode *pMNode = new SmMatrixNode(m_aCurToken);
pMNode->SetSubNodes(ExpressionArray);
pMNode->SetRowCol(r, c);
m_aNodeStack.Push(pMNode);
}
else
Error(PE_LGROUP_EXPECTED);
}
void SmParser::Special()
{
bool bReplace = false;
String &rName = m_aCurToken.aText;
String aNewName;
if (CONVERT_NONE == GetConversion())
{
// conversion of symbol names for 6.0 (XML) file format
// (name change on import / export.
// UI uses localized names XML file format does not.)
if( rName.Len() && rName.GetChar( 0 ) == sal_Unicode( '%' ) )
{
if (IsImportSymbolNames())
{
const SmLocalizedSymbolData &rLSD = SM_MOD()->GetLocSymbolData();
aNewName = rLSD.GetUiSymbolName( rName.Copy( 1 ) );
bReplace = true;
}
else if (IsExportSymbolNames())
{
const SmLocalizedSymbolData &rLSD = SM_MOD()->GetLocSymbolData();
aNewName = rLSD.GetExportSymbolName( rName.Copy( 1 ) );
bReplace = true;
}
}
if( aNewName.Len() )
aNewName.Insert( '%', 0 );
}
else // 5.0 <-> 6.0 formula text (symbol name) conversion
{
LanguageType nLanguage = GetLanguage();
SmLocalizedSymbolData &rData = SM_MOD()->GetLocSymbolData();
const ResStringArray *pFrom = 0;
const ResStringArray *pTo = 0;
if (CONVERT_50_TO_60 == GetConversion())
{
pFrom = rData.Get50NamesArray( nLanguage );
pTo = rData.Get60NamesArray( nLanguage );
}
else if (CONVERT_60_TO_50 == GetConversion())
{
pFrom = rData.Get60NamesArray( nLanguage );
pTo = rData.Get50NamesArray( nLanguage );
}
if (pFrom && pTo)
{
OSL_ENSURE( pFrom->Count() == pTo->Count(),
"array length mismatch" );
sal_uInt16 nCount = sal::static_int_cast< sal_uInt16 >(pFrom->Count());
for (sal_uInt16 i = 0; i < nCount; ++i)
{
if (pFrom->GetString(i) == rName)
{
aNewName = pTo->GetString(i);
bReplace = true;
}
}
}
// else:
// conversion arrays not found or (usually)
// conversion not necessary
}
if (bReplace && aNewName.Len() && rName != aNewName)
{
Replace( GetTokenIndex(), rName.Len(), aNewName );
rName = aNewName;
}
// add symbol name to list of used symbols
const String aSymbolName( m_aCurToken.aText.Copy( 1 ) );
if (aSymbolName.Len() > 0 )
AddToUsedSymbols( aSymbolName );
m_aNodeStack.Push(new SmSpecialNode(m_aCurToken));
NextToken();
}
void SmParser::GlyphSpecial()
{
m_aNodeStack.Push(new SmGlyphSpecialNode(m_aCurToken));
NextToken();
}
void SmParser::Error(SmParseError eError)
{
SmStructureNode *pSNode = new SmExpressionNode(m_aCurToken);
SmErrorNode *pErr = new SmErrorNode(eError, m_aCurToken);
pSNode->SetSubNodes(pErr, 0);
//! put a structure node on the stack (instead of the error node itself)
//! because sometimes such a node is expected in order to attach some
//! subnodes
m_aNodeStack.Push(pSNode);
AddError(eError, pSNode);
NextToken();
}
// end gramar
SmParser::SmParser()
: m_aDotLoc( SvxCreateLocale( LANGUAGE_ENGLISH_US ) )
{
m_eConversion = CONVERT_NONE;
bImportSymNames = m_bExportSymNames = false;
m_nLang = Application::GetSettings().GetUILanguage();
}
SmNode *SmParser::Parse(const String &rBuffer)
{
ClearUsedSymbols();
m_aBufferString = rBuffer;
m_aBufferString.ConvertLineEnd( LINEEND_LF );
m_nBufferIndex = 0;
m_nTokenIndex = 0;
m_Row = 1;
m_nColOff = 0;
m_nCurError = -1;
for ( size_t i = 0, n = m_aErrDescList.size(); i < n; ++i )
delete m_aErrDescList[ i ];
m_aErrDescList.clear();
m_aNodeStack.Clear();
SetLanguage( Application::GetSettings().GetUILanguage() );
NextToken();
Table();
return m_aNodeStack.Pop();
}
SmNode *SmParser::ParseExpression(const String &rBuffer)
{
m_aBufferString = rBuffer;
m_aBufferString.ConvertLineEnd( LINEEND_LF );
m_nBufferIndex = 0;
m_nTokenIndex = 0;
m_Row = 1;
m_nColOff = 0;
m_nCurError = -1;
for ( size_t i = 0, n = m_aErrDescList.size(); i < n; ++i )
delete m_aErrDescList[ i ];
m_aErrDescList.clear();
m_aNodeStack.Clear();
SetLanguage( Application::GetSettings().GetUILanguage() );
NextToken();
Expression();
return m_aNodeStack.Pop();
}
size_t SmParser::AddError(SmParseError Type, SmNode *pNode)
{
SmErrorDesc *pErrDesc = new SmErrorDesc;
pErrDesc->Type = Type;
pErrDesc->pNode = pNode;
pErrDesc->Text = String(SmResId(RID_ERR_IDENT));
sal_uInt16 nRID;
switch (Type)
{
case PE_UNEXPECTED_CHAR: nRID = RID_ERR_UNEXPECTEDCHARACTER; break;
case PE_LGROUP_EXPECTED: nRID = RID_ERR_LGROUPEXPECTED; break;
case PE_RGROUP_EXPECTED: nRID = RID_ERR_RGROUPEXPECTED; break;
case PE_LBRACE_EXPECTED: nRID = RID_ERR_LBRACEEXPECTED; break;
case PE_RBRACE_EXPECTED: nRID = RID_ERR_RBRACEEXPECTED; break;
case PE_FUNC_EXPECTED: nRID = RID_ERR_FUNCEXPECTED; break;
case PE_UNOPER_EXPECTED: nRID = RID_ERR_UNOPEREXPECTED; break;
case PE_BINOPER_EXPECTED: nRID = RID_ERR_BINOPEREXPECTED; break;
case PE_SYMBOL_EXPECTED: nRID = RID_ERR_SYMBOLEXPECTED; break;
case PE_IDENTIFIER_EXPECTED: nRID = RID_ERR_IDENTEXPECTED; break;
case PE_POUND_EXPECTED: nRID = RID_ERR_POUNDEXPECTED; break;
case PE_COLOR_EXPECTED: nRID = RID_ERR_COLOREXPECTED; break;
case PE_RIGHT_EXPECTED: nRID = RID_ERR_RIGHTEXPECTED; break;
default:
nRID = RID_ERR_UNKOWN;
}
pErrDesc->Text += SmResId(nRID);
m_aErrDescList.push_back( pErrDesc );
return m_aErrDescList.size()-1;
}
const SmErrorDesc *SmParser::NextError()
{
if ( !m_aErrDescList.empty() )
if (m_nCurError > 0) return m_aErrDescList[ --m_nCurError ];
else
{
m_nCurError = 0;
return m_aErrDescList[ m_nCurError ];
}
else return NULL;
}
const SmErrorDesc *SmParser::PrevError()
{
if ( !m_aErrDescList.empty() )
if (m_nCurError < (int) (m_aErrDescList.size() - 1)) return m_aErrDescList[ ++m_nCurError ];
else
{
m_nCurError = (int) (m_aErrDescList.size() - 1);
return m_aErrDescList[ m_nCurError ];
}
else return NULL;
}
const SmErrorDesc *SmParser::GetError(size_t i)
{
if ( i < m_aErrDescList.size() )
return m_aErrDescList[ i ];
if ( (size_t)m_nCurError < m_aErrDescList.size() )
return m_aErrDescList[ m_nCurError ];
return NULL;
}
/* vim:set shiftwidth=4 softtabstop=4 expandtab: */