2
0
Fork 0
OB-Xd/Modules/gin/3rdparty/muParser/muParserBase.cpp

1806 lines
63 KiB
C++
Executable File

/*
__________
_____ __ __\______ \_____ _______ ______ ____ _______
/ \ | | \| ___/\__ \ \_ __ \/ ___/_/ __ \\_ __ \
| Y Y \| | /| | / __ \_| | \/\___ \ \ ___/ | | \/
|__|_| /|____/ |____| (____ /|__| /____ > \___ >|__|
\/ \/ \/ \/
Copyright (C) 2011 Ingo Berg
Permission is hereby granted, free of charge, to any person obtaining a copy of this
software and associated documentation files (the "Software"), to deal in the Software
without restriction, including without limitation the rights to use, copy, modify,
merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or
substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#include "muParserBase.h"
#include "muParserTemplateMagic.h"
//--- Standard includes ------------------------------------------------------------------------
#include <cassert>
#include <algorithm>
#include <cmath>
#include <memory>
#include <vector>
#include <deque>
#include <sstream>
#include <locale>
#ifdef MUP_USE_OPENMP
#include <omp.h>
#endif
#if __clang__
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wconversion"
#endif
using namespace std;
/** \file
\brief This file contains the basic implementation of the muparser engine.
*/
namespace mu
{
std::locale ParserBase::s_locale = std::locale(std::locale::classic(), new change_dec_sep<char_type>('.'));
bool ParserBase::g_DbgDumpCmdCode = false;
bool ParserBase::g_DbgDumpStack = false;
//------------------------------------------------------------------------------
/** \brief Identifiers for built in binary operators.
When defining custom binary operators with #AddOprt(...) make sure not to choose
names conflicting with these definitions.
*/
const char_type* ParserBase::c_DefaultOprt[] =
{
_T("<="), _T(">="), _T("!="),
_T("=="), _T("<"), _T(">"),
_T("+"), _T("-"), _T("*"),
_T("/"), _T("^"), _T("&&"),
_T("||"), _T("="), _T("("),
_T(")"), _T("?"), _T(":"), nullptr
};
const int ParserBase::s_MaxNumOpenMPThreads = 16;
//------------------------------------------------------------------------------
/** \brief Constructor.
\param a_szFormula the formula to interpret.
\throw ParserException if a_szFormula is null.
*/
ParserBase::ParserBase()
:m_pParseFormula(&ParserBase::ParseString)
,m_vRPN()
,m_vStringBuf()
,m_pTokenReader()
,m_FunDef()
,m_PostOprtDef()
,m_InfixOprtDef()
,m_OprtDef()
,m_ConstDef()
,m_StrVarDef()
,m_VarDef()
,m_bBuiltInOp(true)
,m_sNameChars()
,m_sOprtChars()
,m_sInfixOprtChars()
,m_nIfElseCounter(0)
,m_vStackBuffer()
,m_nFinalResultIdx(0)
{
InitTokenReader();
}
//---------------------------------------------------------------------------
/** \brief Copy constructor.
The parser can be safely copy constructed but the bytecode is reset during
copy construction.
*/
ParserBase::ParserBase(const ParserBase &a_Parser)
:m_pParseFormula(&ParserBase::ParseString)
,m_vRPN()
,m_vStringBuf()
,m_pTokenReader()
,m_FunDef()
,m_PostOprtDef()
,m_InfixOprtDef()
,m_OprtDef()
,m_ConstDef()
,m_StrVarDef()
,m_VarDef()
,m_bBuiltInOp(true)
,m_sNameChars()
,m_sOprtChars()
,m_sInfixOprtChars()
,m_nIfElseCounter(0)
{
m_pTokenReader.reset(new token_reader_type(this));
Assign(a_Parser);
}
//---------------------------------------------------------------------------
ParserBase::~ParserBase()
{}
//---------------------------------------------------------------------------
/** \brief Assignment operator.
Implemented by calling Assign(a_Parser). Self assignment is suppressed.
\param a_Parser Object to copy to this.
\return *this
\throw nothrow
*/
ParserBase& ParserBase::operator=(const ParserBase &a_Parser)
{
Assign(a_Parser);
return *this;
}
//---------------------------------------------------------------------------
/** \brief Copy state of a parser object to this.
Clears Variables and Functions of this parser.
Copies the states of all internal variables.
Resets parse function to string parse mode.
\param a_Parser the source object.
*/
void ParserBase::Assign(const ParserBase &a_Parser)
{
if (&a_Parser==this)
return;
// Don't copy bytecode instead cause the parser to create new bytecode
// by resetting the parse function.
ReInit();
m_ConstDef = a_Parser.m_ConstDef; // Copy user define constants
m_VarDef = a_Parser.m_VarDef; // Copy user defined variables
m_bBuiltInOp = a_Parser.m_bBuiltInOp;
m_vStringBuf = a_Parser.m_vStringBuf;
m_vStackBuffer = a_Parser.m_vStackBuffer;
m_nFinalResultIdx = a_Parser.m_nFinalResultIdx;
m_StrVarDef = a_Parser.m_StrVarDef;
m_vStringVarBuf = a_Parser.m_vStringVarBuf;
m_nIfElseCounter = a_Parser.m_nIfElseCounter;
m_pTokenReader.reset(a_Parser.m_pTokenReader->Clone(this));
// Copy function and operator callbacks
m_FunDef = a_Parser.m_FunDef; // Copy function definitions
m_PostOprtDef = a_Parser.m_PostOprtDef; // post value unary operators
m_InfixOprtDef = a_Parser.m_InfixOprtDef; // unary operators for infix notation
m_OprtDef = a_Parser.m_OprtDef; // binary operators
m_sNameChars = a_Parser.m_sNameChars;
m_sOprtChars = a_Parser.m_sOprtChars;
m_sInfixOprtChars = a_Parser.m_sInfixOprtChars;
}
//---------------------------------------------------------------------------
/** \brief Set the decimal separator.
\param cDecSep Decimal separator as a character value.
\sa SetThousandsSep
By default muparser uses the "C" locale. The decimal separator of this
locale is overwritten by the one provided here.
*/
void ParserBase::SetDecSep(char_type cDecSep)
{
char_type cThousandsSep = std::use_facet< change_dec_sep<char_type> >(s_locale).thousands_sep();
s_locale = std::locale(std::locale("C"), new change_dec_sep<char_type>(cDecSep, cThousandsSep));
}
//---------------------------------------------------------------------------
/** \brief Sets the thousands operator.
\param cThousandsSep The thousands separator as a character
\sa SetDecSep
By default muparser uses the "C" locale. The thousands separator of this
locale is overwritten by the one provided here.
*/
void ParserBase::SetThousandsSep(char_type cThousandsSep)
{
char_type cDecSep = std::use_facet< change_dec_sep<char_type> >(s_locale).decimal_point();
s_locale = std::locale(std::locale("C"), new change_dec_sep<char_type>(cDecSep, cThousandsSep));
}
//---------------------------------------------------------------------------
/** \brief Resets the locale.
The default locale used "." as decimal separator, no thousands separator and
"," as function argument separator.
*/
void ParserBase::ResetLocale()
{
s_locale = std::locale(std::locale("C"), new change_dec_sep<char_type>('.'));
SetArgSep(',');
}
//---------------------------------------------------------------------------
/** \brief Initialize the token reader.
Create new token reader object and submit pointers to function, operator,
constant and variable definitions.
\post m_pTokenReader.get()!=0
\throw nothrow
*/
void ParserBase::InitTokenReader()
{
m_pTokenReader.reset(new token_reader_type(this));
}
//---------------------------------------------------------------------------
/** \brief Reset parser to string parsing mode and clear internal buffers.
Clear bytecode, reset the token reader.
\throw nothrow
*/
void ParserBase::ReInit() const
{
m_pParseFormula = &ParserBase::ParseString;
m_vStringBuf.clear();
m_vRPN.clear();
m_pTokenReader->ReInit();
m_nIfElseCounter = 0;
}
//---------------------------------------------------------------------------
void ParserBase::OnDetectVar(string_type * /*pExpr*/, int & /*nStart*/, int & /*nEnd*/)
{}
//---------------------------------------------------------------------------
/** \brief Returns the version of muparser.
\param eInfo A flag indicating whether the full version info should be
returned or not.
Format is as follows: "MAJOR.MINOR (COMPILER_FLAGS)" The COMPILER_FLAGS
are returned only if eInfo==pviFULL.
*/
string_type ParserBase::GetVersion(EParserVersionInfo eInfo) const
{
stringstream_type ss;
ss << MUP_VERSION;
if (eInfo==pviFULL)
{
ss << _T(" (") << MUP_VERSION_DATE;
ss << std::dec << _T("; ") << sizeof(void*)*8 << _T("BIT");
#ifdef _DEBUG
ss << _T("; DEBUG");
#else
ss << _T("; RELEASE");
#endif
#ifdef _UNICODE
ss << _T("; UNICODE");
#else
#ifdef _MBCS
ss << _T("; MBCS");
#else
ss << _T("; ASCII");
#endif
#endif
#ifdef MUP_USE_OPENMP
ss << _T("; OPENMP");
//#else
// ss << _T("; NO_OPENMP");
#endif
#if defined(MUP_MATH_EXCEPTIONS)
ss << _T("; MATHEXC");
//#else
// ss << _T("; NO_MATHEXC");
#endif
ss << _T(")");
}
return ss.str();
}
//---------------------------------------------------------------------------
/** \brief Add a value parsing function.
When parsing an expression muParser tries to detect values in the expression
string using different valident callbacks. Thus it's possible to parse
for hex values, binary values and floating point values.
*/
void ParserBase::AddValIdent(identfun_type a_pCallback)
{
m_pTokenReader->AddValIdent(a_pCallback);
}
//---------------------------------------------------------------------------
/** \brief Set a function that can create variable pointer for unknown expression variables.
\param a_pFactory A pointer to the variable factory.
\param pUserData A user defined context pointer.
*/
void ParserBase::SetVarFactory(facfun_type a_pFactory, void *pUserData)
{
m_pTokenReader->SetVarCreator(a_pFactory, pUserData);
}
//---------------------------------------------------------------------------
/** \brief Add a function or operator callback to the parser. */
void ParserBase::AddCallback( const string_type &a_strName,
const ParserCallback &a_Callback,
funmap_type &a_Storage,
const char_type *a_szCharSet )
{
if (a_Callback.GetAddr()==nullptr)
Error(ecINVALID_FUN_PTR);
const funmap_type *pFunMap = &a_Storage;
// Check for conflicting operator or function names
if ( pFunMap!=&m_FunDef && m_FunDef.find(a_strName)!=m_FunDef.end() )
Error(ecNAME_CONFLICT, -1, a_strName);
if ( pFunMap!=&m_PostOprtDef && m_PostOprtDef.find(a_strName)!=m_PostOprtDef.end() )
Error(ecNAME_CONFLICT, -1, a_strName);
if ( pFunMap!=&m_InfixOprtDef && pFunMap!=&m_OprtDef && m_InfixOprtDef.find(a_strName)!=m_InfixOprtDef.end() )
Error(ecNAME_CONFLICT, -1, a_strName);
if ( pFunMap!=&m_InfixOprtDef && pFunMap!=&m_OprtDef && m_OprtDef.find(a_strName)!=m_OprtDef.end() )
Error(ecNAME_CONFLICT, -1, a_strName);
CheckOprt(a_strName, a_Callback, a_szCharSet);
a_Storage[a_strName] = a_Callback;
ReInit();
}
//---------------------------------------------------------------------------
/** \brief Check if a name contains invalid characters.
\throw ParserException if the name contains invalid characters.
*/
void ParserBase::CheckOprt(const string_type &a_sName,
const ParserCallback &a_Callback,
const string_type &a_szCharSet) const
{
if ( !a_sName.length() ||
(a_sName.find_first_not_of(a_szCharSet)!=string_type::npos) ||
(a_sName[0]>='0' && a_sName[0]<='9'))
{
switch(a_Callback.GetCode())
{
case cmOPRT_POSTFIX: Error(ecINVALID_POSTFIX_IDENT, -1, a_sName); break;
case cmOPRT_INFIX: Error(ecINVALID_INFIX_IDENT, -1, a_sName); break;
default: Error(ecINVALID_NAME, -1, a_sName);
}
}
}
//---------------------------------------------------------------------------
/** \brief Check if a name contains invalid characters.
\throw ParserException if the name contains invalid characters.
*/
void ParserBase::CheckName(const string_type &a_sName,
const string_type &a_szCharSet) const
{
if ( !a_sName.length() ||
(a_sName.find_first_not_of(a_szCharSet)!=string_type::npos) ||
(a_sName[0]>='0' && a_sName[0]<='9'))
{
Error(ecINVALID_NAME);
}
}
//---------------------------------------------------------------------------
/** \brief Set the formula.
\param a_strFormula Formula as string_type
\throw ParserException in case of syntax errors.
Triggers first time calculation thus the creation of the bytecode and
scanning of used variables.
*/
void ParserBase::SetExpr(const string_type &a_sExpr)
{
// Check locale compatibility
std::locale loc;
if (m_pTokenReader->GetArgSep()==std::use_facet<numpunct<char_type> >(loc).decimal_point())
Error(ecLOCALE);
// <ibg> 20060222: Bugfix for Borland-Kylix:
// adding a space to the expression will keep Borlands KYLIX from going wild
// when calling tellg on a stringstream created from the expression after
// reading a value at the end of an expression. (mu::Parser::IsVal function)
// (tellg returns -1 otherwise causing the parser to ignore the value)
string_type sBuf(a_sExpr + _T(" ") );
m_pTokenReader->SetFormula(sBuf);
ReInit();
}
//---------------------------------------------------------------------------
/** \brief Get the default symbols used for the built in operators.
\sa c_DefaultOprt
*/
const char_type** ParserBase::GetOprtDef() const
{
return (const char_type **)(&c_DefaultOprt[0]);
}
//---------------------------------------------------------------------------
/** \brief Define the set of valid characters to be used in names of
functions, variables, constants.
*/
void ParserBase::DefineNameChars(const char_type *a_szCharset)
{
m_sNameChars = a_szCharset;
}
//---------------------------------------------------------------------------
/** \brief Define the set of valid characters to be used in names of
binary operators and postfix operators.
*/
void ParserBase::DefineOprtChars(const char_type *a_szCharset)
{
m_sOprtChars = a_szCharset;
}
//---------------------------------------------------------------------------
/** \brief Define the set of valid characters to be used in names of
infix operators.
*/
void ParserBase::DefineInfixOprtChars(const char_type *a_szCharset)
{
m_sInfixOprtChars = a_szCharset;
}
//---------------------------------------------------------------------------
/** \brief Virtual function that defines the characters allowed in name identifiers.
\sa #ValidOprtChars, #ValidPrefixOprtChars
*/
const char_type* ParserBase::ValidNameChars() const
{
assert(m_sNameChars.size());
return m_sNameChars.c_str();
}
//---------------------------------------------------------------------------
/** \brief Virtual function that defines the characters allowed in operator definitions.
\sa #ValidNameChars, #ValidPrefixOprtChars
*/
const char_type* ParserBase::ValidOprtChars() const
{
assert(m_sOprtChars.size());
return m_sOprtChars.c_str();
}
//---------------------------------------------------------------------------
/** \brief Virtual function that defines the characters allowed in infix operator definitions.
\sa #ValidNameChars, #ValidOprtChars
*/
const char_type* ParserBase::ValidInfixOprtChars() const
{
assert(m_sInfixOprtChars.size());
return m_sInfixOprtChars.c_str();
}
//---------------------------------------------------------------------------
/** \brief Add a user defined operator.
\post Will reset the Parser to string parsing mode.
*/
void ParserBase::DefinePostfixOprt(const string_type &a_sName,
fun_type1 a_pFun,
bool a_bAllowOpt)
{
AddCallback(a_sName,
ParserCallback(a_pFun, nullptr, a_bAllowOpt, prPOSTFIX, cmOPRT_POSTFIX),
m_PostOprtDef,
ValidOprtChars() );
}
//---------------------------------------------------------------------------
/** \brief Initialize user defined functions.
Calls the virtual functions InitFun(), InitConst() and InitOprt().
*/
void ParserBase::Init()
{
InitCharSets();
InitFun();
InitConst();
InitOprt();
}
//---------------------------------------------------------------------------
/** \brief Add a user defined operator.
\post Will reset the Parser to string parsing mode.
\param [in] a_sName operator Identifier
\param [in] a_pFun Operator callback function
\param [in] a_iPrec Operator Precedence (default=prSIGN)
\param [in] a_bAllowOpt True if operator is volatile (default=false)
\sa EPrec
*/
void ParserBase::DefineInfixOprt(const string_type &a_sName,
fun_type1 a_pFun,
int a_iPrec,
bool a_bAllowOpt)
{
AddCallback(a_sName,
ParserCallback(a_pFun, nullptr, a_bAllowOpt, a_iPrec, cmOPRT_INFIX),
m_InfixOprtDef,
ValidInfixOprtChars() );
}
//---------------------------------------------------------------------------
/** \brief Define a binary operator.
\param [in] a_sName The identifier of the operator.
\param [in] a_pFun Pointer to the callback function.
\param [in] a_iPrec Precedence of the operator.
\param [in] a_eAssociativity The associativity of the operator.
\param [in] a_bAllowOpt If this is true the operator may be optimized away.
Adds a new Binary operator the the parser instance.
*/
void ParserBase::DefineOprt( const string_type &a_sName,
fun_type2 a_pFun,
unsigned a_iPrec,
EOprtAssociativity a_eAssociativity,
bool a_bAllowOpt )
{
// Check for conflicts with built in operator names
for (int i=0; m_bBuiltInOp && i<cmENDIF; ++i)
if (a_sName == string_type(c_DefaultOprt[i]))
Error(ecBUILTIN_OVERLOAD, -1, a_sName);
AddCallback(a_sName,
ParserCallback(a_pFun, nullptr, a_bAllowOpt, a_iPrec, a_eAssociativity),
m_OprtDef,
ValidOprtChars() );
}
//---------------------------------------------------------------------------
/** \brief Define a new string constant.
\param [in] a_strName The name of the constant.
\param [in] a_strVal the value of the constant.
*/
void ParserBase::DefineStrConst(const string_type &a_strName, const string_type &a_strVal)
{
// Test if a constant with that names already exists
if (m_StrVarDef.find(a_strName)!=m_StrVarDef.end())
Error(ecNAME_CONFLICT);
CheckName(a_strName, ValidNameChars());
m_vStringVarBuf.push_back(a_strVal); // Store variable string in internal buffer
m_StrVarDef[a_strName] = m_vStringVarBuf.size()-1; // bind buffer index to variable name
ReInit();
}
//---------------------------------------------------------------------------
/** \brief Add a user defined variable.
\param [in] a_sName the variable name
\param [in] a_pVar A pointer to the variable value.
\post Will reset the Parser to string parsing mode.
\throw ParserException in case the name contains invalid signs or a_pVar is NULL.
*/
void ParserBase::DefineVar(const string_type &a_sName, value_type *a_pVar)
{
if (a_pVar==nullptr)
Error(ecINVALID_VAR_PTR);
// Test if a constant with that names already exists
if (m_ConstDef.find(a_sName)!=m_ConstDef.end())
Error(ecNAME_CONFLICT);
CheckName(a_sName, ValidNameChars());
m_VarDef[a_sName] = a_pVar;
ReInit();
}
//---------------------------------------------------------------------------
/** \brief Add a user defined constant.
\param [in] a_sName The name of the constant.
\param [in] a_fVal the value of the constant.
\post Will reset the Parser to string parsing mode.
\throw ParserException in case the name contains invalid signs.
*/
void ParserBase::DefineConst(const string_type &a_sName, value_type a_fVal)
{
CheckName(a_sName, ValidNameChars());
m_ConstDef[a_sName] = a_fVal;
ReInit();
}
//---------------------------------------------------------------------------
/** \brief Get operator priority.
\throw ParserException if a_Oprt is no operator code
*/
int ParserBase::GetOprtPrecedence(const token_type &a_Tok) const
{
switch (a_Tok.GetCode())
{
// built in operators
case cmEND: return -5;
case cmARG_SEP: return -4;
case cmASSIGN: return -1;
case cmELSE:
case cmIF: return 0;
case cmLAND: return prLAND;
case cmLOR: return prLOR;
case cmLT:
case cmGT:
case cmLE:
case cmGE:
case cmNEQ:
case cmEQ: return prCMP;
case cmADD:
case cmSUB: return prADD_SUB;
case cmMUL:
case cmDIV: return prMUL_DIV;
case cmPOW: return prPOW;
// user defined binary operators
case cmOPRT_INFIX:
case cmOPRT_BIN: return a_Tok.GetPri();
default: Error(ecINTERNAL_ERROR, 5);
return 999;
}
}
//---------------------------------------------------------------------------
/** \brief Get operator priority.
\throw ParserException if a_Oprt is no operator code
*/
EOprtAssociativity ParserBase::GetOprtAssociativity(const token_type &a_Tok) const
{
switch (a_Tok.GetCode())
{
case cmASSIGN:
case cmLAND:
case cmLOR:
case cmLT:
case cmGT:
case cmLE:
case cmGE:
case cmNEQ:
case cmEQ:
case cmADD:
case cmSUB:
case cmMUL:
case cmDIV: return oaLEFT;
case cmPOW: return oaRIGHT;
case cmOPRT_BIN: return a_Tok.GetAssociativity();
default: return oaNONE;
}
}
//---------------------------------------------------------------------------
/** \brief Return a map containing the used variables only. */
const varmap_type& ParserBase::GetUsedVar() const
{
try
{
m_pTokenReader->IgnoreUndefVar(true);
CreateRPN(); // try to create bytecode, but don't use it for any further calculations since it
// may contain references to nonexisting variables.
m_pParseFormula = &ParserBase::ParseString;
m_pTokenReader->IgnoreUndefVar(false);
}
catch(exception_type & /*e*/)
{
// Make sure to stay in string parse mode, don't call ReInit()
// because it deletes the array with the used variables
m_pParseFormula = &ParserBase::ParseString;
m_pTokenReader->IgnoreUndefVar(false);
throw;
}
return m_pTokenReader->GetUsedVar();
}
//---------------------------------------------------------------------------
/** \brief Return a map containing the used variables only. */
const varmap_type& ParserBase::GetVar() const
{
return m_VarDef;
}
//---------------------------------------------------------------------------
/** \brief Return a map containing all parser constants. */
const valmap_type& ParserBase::GetConst() const
{
return m_ConstDef;
}
//---------------------------------------------------------------------------
/** \brief Return prototypes of all parser functions.
\return #m_FunDef
\sa FunProt
\throw nothrow
The return type is a map of the public type #funmap_type containing the prototype
definitions for all numerical parser functions. String functions are not part of
this map. The Prototype definition is encapsulated in objects of the class FunProt
one per parser function each associated with function names via a map construct.
*/
const funmap_type& ParserBase::GetFunDef() const
{
return m_FunDef;
}
//---------------------------------------------------------------------------
/** \brief Retrieve the formula. */
const string_type& ParserBase::GetExpr() const
{
return m_pTokenReader->GetExpr();
}
//---------------------------------------------------------------------------
/** \brief Execute a function that takes a single string argument.
\param a_FunTok Function token.
\throw exception_type If the function token is not a string function
*/
ParserBase::token_type ParserBase::ApplyStrFunc(const token_type &a_FunTok,
const std::vector<token_type> &a_vArg) const
{
if (a_vArg.back().GetCode()!=cmSTRING)
Error(ecSTRING_EXPECTED, m_pTokenReader->GetPos(), a_FunTok.GetAsString());
token_type valTok;
generic_fun_type pFunc = a_FunTok.GetFuncAddr();
assert(pFunc);
try
{
// Check function arguments; write dummy value into valtok to represent the result
switch(a_FunTok.GetArgCount())
{
case 0: valTok.SetVal(1); a_vArg[0].GetAsString(); break;
case 1: valTok.SetVal(1); a_vArg[1].GetAsString(); a_vArg[0].GetVal(); break;
case 2: valTok.SetVal(1); a_vArg[2].GetAsString(); a_vArg[1].GetVal(); a_vArg[0].GetVal(); break;
default: Error(ecINTERNAL_ERROR);
}
}
catch(ParserError& )
{
Error(ecVAL_EXPECTED, m_pTokenReader->GetPos(), a_FunTok.GetAsString());
}
// string functions won't be optimized
m_vRPN.AddStrFun(pFunc, a_FunTok.GetParam(), a_FunTok.GetArgCount(), a_vArg.back().GetIdx());
// Push dummy value representing the function result to the stack
return valTok;
}
//---------------------------------------------------------------------------
/** \brief Apply a function token.
\param iArgCount Number of Arguments actually gathered used only for multiarg functions.
\post The result is pushed to the value stack
\post The function token is removed from the stack
\throw exception_type if Argument count does not match function requirements.
*/
void ParserBase::ApplyFunc( ParserStack<token_type> &a_stOpt,
ParserStack<token_type> &a_stVal,
int a_iArgCount) const
{
assert(m_pTokenReader.get());
// Operator stack empty or does not contain tokens with callback functions
if (a_stOpt.empty() || a_stOpt.top().GetFuncAddr()==nullptr )
return;
token_type funTok = a_stOpt.pop();
assert(funTok.GetFuncAddr());
// Binary operators must rely on their internal operator number
// since counting of operators relies on commas for function arguments
// binary operators do not have commas in their expression
int iArgCount = (funTok.GetCode()==cmOPRT_BIN) ? funTok.GetArgCount() : a_iArgCount;
// determine how many parameters the function needs. To remember iArgCount includes the
// string parameter whilst GetArgCount() counts only numeric parameters.
int iArgRequired = funTok.GetArgCount() + ((funTok.GetType()==tpSTR) ? 1 : 0);
// That's the number of numerical parameters
int iArgNumerical = iArgCount - ((funTok.GetType()==tpSTR) ? 1 : 0);
if (funTok.GetCode()==cmFUNC_STR && iArgCount-iArgNumerical>1)
Error(ecINTERNAL_ERROR);
if (funTok.GetArgCount()>=0 && iArgCount>iArgRequired)
Error(ecTOO_MANY_PARAMS, m_pTokenReader->GetPos()-1, funTok.GetAsString());
if (funTok.GetCode()!=cmOPRT_BIN && iArgCount<iArgRequired )
Error(ecTOO_FEW_PARAMS, m_pTokenReader->GetPos()-1, funTok.GetAsString());
if (funTok.GetCode()==cmFUNC_STR && iArgCount>iArgRequired )
Error(ecTOO_MANY_PARAMS, m_pTokenReader->GetPos()-1, funTok.GetAsString());
// Collect the numeric function arguments from the value stack and store them
// in a vector
std::vector<token_type> stArg;
for (int i=0; i<iArgNumerical; ++i)
{
stArg.push_back( a_stVal.pop() );
if ( stArg.back().GetType()==tpSTR && funTok.GetType()!=tpSTR )
Error(ecVAL_EXPECTED, m_pTokenReader->GetPos(), funTok.GetAsString());
}
switch(funTok.GetCode())
{
case cmFUNC_STR:
stArg.push_back(a_stVal.pop());
if ( stArg.back().GetType()==tpSTR && funTok.GetType()!=tpSTR )
Error(ecVAL_EXPECTED, m_pTokenReader->GetPos(), funTok.GetAsString());
ApplyStrFunc(funTok, stArg);
break;
case cmFUNC_BULK:
m_vRPN.AddBulkFun(funTok.GetFuncAddr(), funTok.GetParam(), (int)stArg.size());
break;
case cmOPRT_BIN:
case cmOPRT_POSTFIX:
case cmOPRT_INFIX:
case cmFUNC:
if (funTok.GetArgCount()==-1 && iArgCount==0)
Error(ecTOO_FEW_PARAMS, m_pTokenReader->GetPos(), funTok.GetAsString());
m_vRPN.AddFun(funTok.GetFuncAddr(), funTok.GetParam(), (funTok.GetArgCount()==-1) ? -iArgNumerical : iArgNumerical);
break;
default:
break;
}
// Push dummy value representing the function result to the stack
token_type token;
token.SetVal(1);
a_stVal.push(token);
}
//---------------------------------------------------------------------------
void ParserBase::ApplyIfElse(ParserStack<token_type> &a_stOpt,
ParserStack<token_type> &a_stVal) const
{
// Check if there is an if Else clause to be calculated
while (a_stOpt.size() && a_stOpt.top().GetCode()==cmELSE)
{
token_type opElse = a_stOpt.pop();
MUP_ASSERT(a_stOpt.size()>0);
// Take the value associated with the else branch from the value stack
token_type vVal2 = a_stVal.pop();
MUP_ASSERT(a_stOpt.size()>0);
MUP_ASSERT(a_stVal.size()>=2);
// it then else is a ternary operator Pop all three values from the value s
// tack and just return the right value
token_type vVal1 = a_stVal.pop();
token_type vExpr = a_stVal.pop();
a_stVal.push( (vExpr.GetVal()!=0) ? vVal1 : vVal2);
token_type opIf = a_stOpt.pop();
MUP_ASSERT(opElse.GetCode()==cmELSE);
MUP_ASSERT(opIf.GetCode()==cmIF);
m_vRPN.AddIfElse(cmENDIF);
} // while pending if-else-clause found
}
//---------------------------------------------------------------------------
/** \brief Performs the necessary steps to write code for
the execution of binary operators into the bytecode.
*/
void ParserBase::ApplyBinOprt(ParserStack<token_type> &a_stOpt,
ParserStack<token_type> &a_stVal) const
{
// is it a user defined binary operator?
if (a_stOpt.top().GetCode()==cmOPRT_BIN)
{
ApplyFunc(a_stOpt, a_stVal, 2);
}
else
{
MUP_ASSERT(a_stVal.size()>=2);
token_type valTok1 = a_stVal.pop(),
valTok2 = a_stVal.pop(),
optTok = a_stOpt.pop(),
resTok;
if ( valTok1.GetType()!=valTok2.GetType() ||
(valTok1.GetType()==tpSTR && valTok2.GetType()==tpSTR) )
Error(ecOPRT_TYPE_CONFLICT, m_pTokenReader->GetPos(), optTok.GetAsString());
if (optTok.GetCode()==cmASSIGN)
{
if (valTok2.GetCode()!=cmVAR)
Error(ecUNEXPECTED_OPERATOR, -1, _T("="));
m_vRPN.AddAssignOp(valTok2.GetVar());
}
else
m_vRPN.AddOp(optTok.GetCode());
resTok.SetVal(1);
a_stVal.push(resTok);
}
}
//---------------------------------------------------------------------------
/** \brief Apply a binary operator.
\param a_stOpt The operator stack
\param a_stVal The value stack
*/
void ParserBase::ApplyRemainingOprt(ParserStack<token_type> &stOpt,
ParserStack<token_type> &stVal) const
{
while (stOpt.size() &&
stOpt.top().GetCode() != cmBO &&
stOpt.top().GetCode() != cmIF)
{
token_type tok = stOpt.top();
switch (tok.GetCode())
{
case cmOPRT_INFIX:
case cmOPRT_BIN:
case cmLE:
case cmGE:
case cmNEQ:
case cmEQ:
case cmLT:
case cmGT:
case cmADD:
case cmSUB:
case cmMUL:
case cmDIV:
case cmPOW:
case cmLAND:
case cmLOR:
case cmASSIGN:
if (stOpt.top().GetCode()==cmOPRT_INFIX)
ApplyFunc(stOpt, stVal, 1);
else
ApplyBinOprt(stOpt, stVal);
break;
case cmELSE:
ApplyIfElse(stOpt, stVal);
break;
default:
Error(ecINTERNAL_ERROR);
}
}
}
//---------------------------------------------------------------------------
/** \brief Parse the command code.
\sa ParseString(...)
Command code contains precalculated stack positions of the values and the
associated operators. The Stack is filled beginning from index one the
value at index zero is not used at all.
*/
value_type ParserBase::ParseCmdCode() const
{
return ParseCmdCodeBulk(0, 0);
}
//---------------------------------------------------------------------------
/** \brief Evaluate the RPN.
\param nOffset The offset added to variable addresses (for bulk mode)
\param nThreadID OpenMP Thread id of the calling thread
*/
value_type ParserBase::ParseCmdCodeBulk(int nOffset, int nThreadID) const
{
assert(nThreadID<=s_MaxNumOpenMPThreads);
// Note: The check for nOffset==0 and nThreadID here is not necessary but
// brings a minor performance gain when not in bulk mode.
value_type *Stack = ((nOffset==0) && (nThreadID==0)) ? &m_vStackBuffer[0] : &m_vStackBuffer[nThreadID * (m_vStackBuffer.size() / s_MaxNumOpenMPThreads)];
value_type buf;
int sidx(0);
for (const SToken *pTok = m_vRPN.GetBase(); pTok->Cmd!=cmEND ; ++pTok)
{
switch (pTok->Cmd)
{
// built in binary operators
case cmLE: --sidx; Stack[sidx] = Stack[sidx] <= Stack[sidx+1]; continue;
case cmGE: --sidx; Stack[sidx] = Stack[sidx] >= Stack[sidx+1]; continue;
case cmNEQ: --sidx; Stack[sidx] = Stack[sidx] != Stack[sidx+1]; continue;
case cmEQ: --sidx; Stack[sidx] = Stack[sidx] == Stack[sidx+1]; continue;
case cmLT: --sidx; Stack[sidx] = Stack[sidx] < Stack[sidx+1]; continue;
case cmGT: --sidx; Stack[sidx] = Stack[sidx] > Stack[sidx+1]; continue;
case cmADD: --sidx; Stack[sidx] += Stack[1+sidx]; continue;
case cmSUB: --sidx; Stack[sidx] -= Stack[1+sidx]; continue;
case cmMUL: --sidx; Stack[sidx] *= Stack[1+sidx]; continue;
case cmDIV: --sidx;
#if defined(MUP_MATH_EXCEPTIONS)
if (Stack[1+sidx]==0)
Error(ecDIV_BY_ZERO);
#endif
Stack[sidx] /= Stack[1+sidx];
continue;
case cmPOW:
--sidx; Stack[sidx] = MathImpl<value_type>::Pow(Stack[sidx], Stack[1+sidx]);
continue;
case cmLAND: --sidx; Stack[sidx] = Stack[sidx] && Stack[sidx+1]; continue;
case cmLOR: --sidx; Stack[sidx] = Stack[sidx] || Stack[sidx+1]; continue;
case cmASSIGN:
// Bugfix for Bulkmode:
// for details see:
// https://groups.google.com/forum/embed/?place=forum/muparser-dev&showsearch=true&showpopout=true&showtabs=false&parenturl=http://muparser.beltoforion.de/mup_forum.html&afterlogin&pli=1#!topic/muparser-dev/szgatgoHTws
--sidx; Stack[sidx] = *(pTok->Oprt.ptr + nOffset) = Stack[sidx + 1]; continue;
// original code:
//--sidx; Stack[sidx] = *pTok->Oprt.ptr = Stack[sidx+1]; continue;
//case cmBO: // unused, listed for compiler optimization purposes
//case cmBC:
// MUP_FAIL(INVALID_CODE_IN_BYTECODE);
// continue;
case cmIF:
if (Stack[sidx--]==0)
pTok += pTok->Oprt.offset;
continue;
case cmELSE:
pTok += pTok->Oprt.offset;
continue;
case cmENDIF:
continue;
//case cmARG_SEP:
// MUP_FAIL(INVALID_CODE_IN_BYTECODE);
// continue;
// value and variable tokens
case cmVAR: Stack[++sidx] = *(pTok->Val.ptr + nOffset); continue;
case cmVAL: Stack[++sidx] = pTok->Val.data2; continue;
case cmVARPOW2: buf = *(pTok->Val.ptr + nOffset);
Stack[++sidx] = buf*buf;
continue;
case cmVARPOW3: buf = *(pTok->Val.ptr + nOffset);
Stack[++sidx] = buf*buf*buf;
continue;
case cmVARPOW4: buf = *(pTok->Val.ptr + nOffset);
Stack[++sidx] = buf*buf*buf*buf;
continue;
case cmVARMUL: Stack[++sidx] = *(pTok->Val.ptr + nOffset) * pTok->Val.data + pTok->Val.data2;
continue;
// Next is treatment of numeric functions
case cmFUNC:
{
int iArgCount = pTok->Fun.argc;
SParam p;
p.id = pTok->Fun.id;
p.param = pTok->Fun.param;
// switch according to argument count
switch(iArgCount)
{
case 0: sidx += 1; Stack[sidx] = (*(fun_type0)pTok->Fun.ptr)(p); continue;
case 1: Stack[sidx] = (*(fun_type1)pTok->Fun.ptr)(p, Stack[sidx]); continue;
case 2: sidx -= 1; Stack[sidx] = (*(fun_type2)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1]); continue;
case 3: sidx -= 2; Stack[sidx] = (*(fun_type3)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2]); continue;
case 4: sidx -= 3; Stack[sidx] = (*(fun_type4)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3]); continue;
case 5: sidx -= 4; Stack[sidx] = (*(fun_type5)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4]); continue;
case 6: sidx -= 5; Stack[sidx] = (*(fun_type6)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5]); continue;
case 7: sidx -= 6; Stack[sidx] = (*(fun_type7)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6]); continue;
case 8: sidx -= 7; Stack[sidx] = (*(fun_type8)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6], Stack[sidx+7]); continue;
case 9: sidx -= 8; Stack[sidx] = (*(fun_type9)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6], Stack[sidx+7], Stack[sidx+8]); continue;
case 10:sidx -= 9; Stack[sidx] = (*(fun_type10)pTok->Fun.ptr)(p, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6], Stack[sidx+7], Stack[sidx+8], Stack[sidx+9]); continue;
default:
if (iArgCount>0) // function with variable arguments store the number as a negative value
Error(ecINTERNAL_ERROR, 1);
sidx -= -iArgCount - 1;
Stack[sidx] =(*(multfun_type)pTok->Fun.ptr)(p, &Stack[sidx], -iArgCount);
continue;
}
}
// Next is treatment of string functions
case cmFUNC_STR:
{
sidx -= pTok->Fun.argc -1;
// The index of the string argument in the string table
int iIdxStack = pTok->Fun.idx;
MUP_ASSERT( iIdxStack>=0 && iIdxStack<(int)m_vStringBuf.size() );
SParam p;
p.id = pTok->Fun.id;
p.param = pTok->Fun.param;
switch(pTok->Fun.argc) // switch according to argument count
{
case 0: Stack[sidx] = (*(strfun_type1)pTok->Fun.ptr)(p, m_vStringBuf[iIdxStack].c_str()); continue;
case 1: Stack[sidx] = (*(strfun_type2)pTok->Fun.ptr)(p, m_vStringBuf[iIdxStack].c_str(), Stack[sidx]); continue;
case 2: Stack[sidx] = (*(strfun_type3)pTok->Fun.ptr)(p, m_vStringBuf[iIdxStack].c_str(), Stack[sidx], Stack[sidx+1]); continue;
}
continue;
}
case cmFUNC_BULK:
{
int iArgCount = pTok->Fun.argc;
SParam p;
p.id = pTok->Fun.id;
p.param = pTok->Fun.param;
// switch according to argument count
switch(iArgCount)
{
case 0: sidx += 1; Stack[sidx] = (*(bulkfun_type0 )pTok->Fun.ptr)(p, nOffset, nThreadID); continue;
case 1: Stack[sidx] = (*(bulkfun_type1 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx]); continue;
case 2: sidx -= 1; Stack[sidx] = (*(bulkfun_type2 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1]); continue;
case 3: sidx -= 2; Stack[sidx] = (*(bulkfun_type3 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2]); continue;
case 4: sidx -= 3; Stack[sidx] = (*(bulkfun_type4 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3]); continue;
case 5: sidx -= 4; Stack[sidx] = (*(bulkfun_type5 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4]); continue;
case 6: sidx -= 5; Stack[sidx] = (*(bulkfun_type6 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5]); continue;
case 7: sidx -= 6; Stack[sidx] = (*(bulkfun_type7 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6]); continue;
case 8: sidx -= 7; Stack[sidx] = (*(bulkfun_type8 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6], Stack[sidx+7]); continue;
case 9: sidx -= 8; Stack[sidx] = (*(bulkfun_type9 )pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6], Stack[sidx+7], Stack[sidx+8]); continue;
case 10:sidx -= 9; Stack[sidx] = (*(bulkfun_type10)pTok->Fun.ptr)(p, nOffset, nThreadID, Stack[sidx], Stack[sidx+1], Stack[sidx+2], Stack[sidx+3], Stack[sidx+4], Stack[sidx+5], Stack[sidx+6], Stack[sidx+7], Stack[sidx+8], Stack[sidx+9]); continue;
default:
Error(ecINTERNAL_ERROR, 2);
continue;
}
}
default:
Error(ecINTERNAL_ERROR, 3);
return 0;
} // switch CmdCode
} // for all bytecode tokens
return Stack[m_nFinalResultIdx];
}
//---------------------------------------------------------------------------
void ParserBase::CreateRPN() const
{
if (!m_pTokenReader->GetExpr().length())
Error(ecUNEXPECTED_EOF, 0);
ParserStack<token_type> stOpt, stVal;
ParserStack<int> stArgCount;
token_type opta, opt; // for storing operators
token_type val, tval; // for storing value
ReInit();
// The outermost counter counts the number of separated items
// such as in "a=10,b=20,c=c+a"
stArgCount.push(1);
for(;;)
{
opt = m_pTokenReader->ReadNextToken();
switch (opt.GetCode())
{
//
// Next three are different kind of value entries
//
case cmSTRING:
opt.SetIdx((int)m_vStringBuf.size()); // Assign buffer index to token
stVal.push(opt);
m_vStringBuf.push_back(opt.GetAsString()); // Store string in internal buffer
break;
case cmVAR:
stVal.push(opt);
m_vRPN.AddVar( static_cast<value_type*>(opt.GetVar()) );
break;
case cmVAL:
stVal.push(opt);
m_vRPN.AddVal( opt.GetVal() );
break;
case cmELSE:
m_nIfElseCounter--;
if (m_nIfElseCounter<0)
Error(ecMISPLACED_COLON, m_pTokenReader->GetPos());
ApplyRemainingOprt(stOpt, stVal);
m_vRPN.AddIfElse(cmELSE);
stOpt.push(opt);
break;
case cmARG_SEP:
if (stArgCount.empty())
Error(ecUNEXPECTED_ARG_SEP, m_pTokenReader->GetPos());
++stArgCount.top();
// Falls through.
// intentional (no break!)
case cmEND:
ApplyRemainingOprt(stOpt, stVal);
break;
case cmBC:
{
// The argument count for parameterless functions is zero
// by default an opening bracket sets parameter count to 1
// in preparation of arguments to come. If the last token
// was an opening bracket we know better...
if (opta.GetCode()==cmBO)
--stArgCount.top();
ApplyRemainingOprt(stOpt, stVal);
// Check if the bracket content has been evaluated completely
if (stOpt.size() && stOpt.top().GetCode()==cmBO)
{
// if opt is ")" and opta is "(" the bracket has been evaluated, now its time to check
// if there is either a function or a sign pending
// neither the opening nor the closing bracket will be pushed back to
// the operator stack
// Check if a function is standing in front of the opening bracket,
// if yes evaluate it afterwards check for infix operators
assert(stArgCount.size());
int iArgCount = stArgCount.pop();
stOpt.pop(); // Take opening bracket from stack
if (iArgCount>1 && ( stOpt.size()==0 ||
(stOpt.top().GetCode()!=cmFUNC &&
stOpt.top().GetCode()!=cmFUNC_BULK &&
stOpt.top().GetCode()!=cmFUNC_STR) ) )
Error(ecUNEXPECTED_ARG, m_pTokenReader->GetPos());
// The opening bracket was popped from the stack now check if there
// was a function before this bracket
if (stOpt.size() &&
stOpt.top().GetCode()!=cmOPRT_INFIX &&
stOpt.top().GetCode()!=cmOPRT_BIN &&
stOpt.top().GetFuncAddr()!=nullptr)
{
ApplyFunc(stOpt, stVal, iArgCount);
}
}
} // if bracket content is evaluated
break;
//
// Next are the binary operator entries
//
//case cmAND: // built in binary operators
//case cmOR:
//case cmXOR:
case cmIF:
m_nIfElseCounter++;
// Falls through.
// intentional (no break!)
case cmLAND:
case cmLOR:
case cmLT:
case cmGT:
case cmLE:
case cmGE:
case cmNEQ:
case cmEQ:
case cmADD:
case cmSUB:
case cmMUL:
case cmDIV:
case cmPOW:
case cmASSIGN:
case cmOPRT_BIN:
// A binary operator (user defined or built in) has been found.
while ( stOpt.size() &&
stOpt.top().GetCode() != cmBO &&
stOpt.top().GetCode() != cmELSE &&
stOpt.top().GetCode() != cmIF)
{
int nPrec1 = GetOprtPrecedence(stOpt.top()),
nPrec2 = GetOprtPrecedence(opt);
if (stOpt.top().GetCode()==opt.GetCode())
{
// Deal with operator associativity
EOprtAssociativity eOprtAsct = GetOprtAssociativity(opt);
if ( (eOprtAsct==oaRIGHT && (nPrec1 <= nPrec2)) ||
(eOprtAsct==oaLEFT && (nPrec1 < nPrec2)) )
{
break;
}
}
else if (nPrec1 < nPrec2)
{
// In case the operators are not equal the precedence decides alone...
break;
}
if (stOpt.top().GetCode()==cmOPRT_INFIX)
ApplyFunc(stOpt, stVal, 1);
else
ApplyBinOprt(stOpt, stVal);
} // while ( ... )
if (opt.GetCode()==cmIF)
m_vRPN.AddIfElse(opt.GetCode());
// The operator can't be evaluated right now, push back to the operator stack
stOpt.push(opt);
break;
//
// Last section contains functions and operators implicitly mapped to functions
//
case cmBO:
stArgCount.push(1);
stOpt.push(opt);
break;
case cmOPRT_INFIX:
case cmFUNC:
case cmFUNC_BULK:
case cmFUNC_STR:
stOpt.push(opt);
break;
case cmOPRT_POSTFIX:
stOpt.push(opt);
ApplyFunc(stOpt, stVal, 1); // this is the postfix operator
break;
default: Error(ecINTERNAL_ERROR, 3);
} // end of switch operator-token
opta = opt;
if ( opt.GetCode() == cmEND )
{
m_vRPN.Finalize();
break;
}
if (ParserBase::g_DbgDumpStack)
{
StackDump(stVal, stOpt);
m_vRPN.AsciiDump();
}
} // while (true)
if (ParserBase::g_DbgDumpCmdCode)
m_vRPN.AsciiDump();
if (m_nIfElseCounter>0)
Error(ecMISSING_ELSE_CLAUSE);
// get the last value (= final result) from the stack
MUP_ASSERT(stArgCount.size()==1);
m_nFinalResultIdx = stArgCount.top();
if (m_nFinalResultIdx==0)
Error(ecINTERNAL_ERROR, 9);
if (stVal.size()==0)
Error(ecEMPTY_EXPRESSION);
if (stVal.top().GetType()!=tpDBL)
Error(ecSTR_RESULT);
m_vStackBuffer.resize(m_vRPN.GetMaxStackSize() * s_MaxNumOpenMPThreads);
}
//---------------------------------------------------------------------------
/** \brief One of the two main parse functions.
\sa ParseCmdCode(...)
Parse expression from input string. Perform syntax checking and create
bytecode. After parsing the string and creating the bytecode the function
pointer #m_pParseFormula will be changed to the second parse routine the
uses bytecode instead of string parsing.
*/
value_type ParserBase::ParseString() const
{
try
{
CreateRPN();
m_pParseFormula = &ParserBase::ParseCmdCode;
return (this->*m_pParseFormula)();
}
catch(ParserError &exc)
{
exc.SetFormula(m_pTokenReader->GetExpr());
throw;
}
}
//---------------------------------------------------------------------------
/** \brief Create an error containing the parse error position.
This function will create an Parser Exception object containing the error text and
its position.
\param a_iErrc [in] The error code of type #EErrorCodes.
\param a_iPos [in] The position where the error was detected.
\param a_strTok [in] The token string representation associated with the error.
\throw ParserException always throws that's the only purpose of this function.
*/
void ParserBase::Error(EErrorCodes a_iErrc, int a_iPos, const string_type &a_sTok) const
{
throw exception_type(a_iErrc, a_sTok, m_pTokenReader->GetExpr(), a_iPos);
}
//------------------------------------------------------------------------------
/** \brief Clear all user defined variables.
\throw nothrow
Resets the parser to string parsing mode by calling #ReInit.
*/
void ParserBase::ClearVar()
{
m_VarDef.clear();
ReInit();
}
//------------------------------------------------------------------------------
/** \brief Remove a variable from internal storage.
\throw nothrow
Removes a variable if it exists. If the Variable does not exist nothing will be done.
*/
void ParserBase::RemoveVar(const string_type &a_strVarName)
{
varmap_type::iterator item = m_VarDef.find(a_strVarName);
if (item!=m_VarDef.end())
{
m_VarDef.erase(item);
ReInit();
}
}
//------------------------------------------------------------------------------
/** \brief Clear all functions.
\post Resets the parser to string parsing mode.
\throw nothrow
*/
void ParserBase::ClearFun()
{
m_FunDef.clear();
ReInit();
}
//------------------------------------------------------------------------------
/** \brief Clear all user defined constants.
Both numeric and string constants will be removed from the internal storage.
\post Resets the parser to string parsing mode.
\throw nothrow
*/
void ParserBase::ClearConst()
{
m_ConstDef.clear();
m_StrVarDef.clear();
ReInit();
}
//------------------------------------------------------------------------------
/** \brief Clear all user defined postfix operators.
\post Resets the parser to string parsing mode.
\throw nothrow
*/
void ParserBase::ClearPostfixOprt()
{
m_PostOprtDef.clear();
ReInit();
}
//------------------------------------------------------------------------------
/** \brief Clear all user defined binary operators.
\post Resets the parser to string parsing mode.
\throw nothrow
*/
void ParserBase::ClearOprt()
{
m_OprtDef.clear();
ReInit();
}
//------------------------------------------------------------------------------
/** \brief Clear the user defined Prefix operators.
\post Resets the parser to string parser mode.
\throw nothrow
*/
void ParserBase::ClearInfixOprt()
{
m_InfixOprtDef.clear();
ReInit();
}
//------------------------------------------------------------------------------
/** \brief Enable or disable the formula optimization feature.
\post Resets the parser to string parser mode.
\throw nothrow
*/
void ParserBase::EnableOptimizer(bool a_bIsOn)
{
m_vRPN.EnableOptimizer(a_bIsOn);
ReInit();
}
//---------------------------------------------------------------------------
/** \brief Enable the dumping of bytecode and stack content on the console.
\param bDumpCmd Flag to enable dumping of the current bytecode to the console.
\param bDumpStack Flag to enable dumping of the stack content is written to the console.
This function is for debug purposes only!
*/
void ParserBase::EnableDebugDump(bool bDumpCmd, bool bDumpStack)
{
ParserBase::g_DbgDumpCmdCode = bDumpCmd;
ParserBase::g_DbgDumpStack = bDumpStack;
}
//------------------------------------------------------------------------------
/** \brief Enable or disable the built in binary operators.
\throw nothrow
\sa m_bBuiltInOp, ReInit()
If you disable the built in binary operators there will be no binary operators
defined. Thus you must add them manually one by one. It is not possible to
disable built in operators selectively. This function will Reinitialize the
parser by calling ReInit().
*/
void ParserBase::EnableBuiltInOprt(bool a_bIsOn)
{
m_bBuiltInOp = a_bIsOn;
ReInit();
}
//------------------------------------------------------------------------------
/** \brief Query status of built in variables.
\return #m_bBuiltInOp; true if built in operators are enabled.
\throw nothrow
*/
bool ParserBase::HasBuiltInOprt() const
{
return m_bBuiltInOp;
}
//------------------------------------------------------------------------------
/** \brief Get the argument separator character.
*/
char_type ParserBase::GetArgSep() const
{
return m_pTokenReader->GetArgSep();
}
//------------------------------------------------------------------------------
/** \brief Set argument separator.
\param cArgSep the argument separator character.
*/
void ParserBase::SetArgSep(char_type cArgSep)
{
m_pTokenReader->SetArgSep(cArgSep);
}
//------------------------------------------------------------------------------
/** \brief Dump stack content.
This function is used for debugging only.
*/
void ParserBase::StackDump(const ParserStack<token_type> &a_stVal,
const ParserStack<token_type> &a_stOprt) const
{
ParserStack<token_type> stOprt(a_stOprt),
stVal(a_stVal);
mu::console() << _T("\nValue stack:\n");
while ( !stVal.empty() )
{
token_type val = stVal.pop();
if (val.GetType()==tpSTR)
mu::console() << _T(" \"") << val.GetAsString() << _T("\" ");
else
mu::console() << _T(" ") << val.GetVal() << _T(" ");
}
mu::console() << "\nOperator stack:\n";
while ( !stOprt.empty() )
{
if (stOprt.top().GetCode()<=cmASSIGN)
{
mu::console() << _T("OPRT_INTRNL \"")
<< ParserBase::c_DefaultOprt[stOprt.top().GetCode()]
<< _T("\" \n");
}
else
{
switch(stOprt.top().GetCode())
{
case cmVAR: mu::console() << _T("VAR\n"); break;
case cmVAL: mu::console() << _T("VAL\n"); break;
case cmFUNC: mu::console() << _T("FUNC \"")
<< stOprt.top().GetAsString()
<< _T("\"\n"); break;
case cmFUNC_BULK: mu::console() << _T("FUNC_BULK \"")
<< stOprt.top().GetAsString()
<< _T("\"\n"); break;
case cmOPRT_INFIX: mu::console() << _T("OPRT_INFIX \"")
<< stOprt.top().GetAsString()
<< _T("\"\n"); break;
case cmOPRT_BIN: mu::console() << _T("OPRT_BIN \"")
<< stOprt.top().GetAsString()
<< _T("\"\n"); break;
case cmFUNC_STR: mu::console() << _T("FUNC_STR\n"); break;
case cmEND: mu::console() << _T("END\n"); break;
case cmUNKNOWN: mu::console() << _T("UNKNOWN\n"); break;
case cmBO: mu::console() << _T("BRACKET \"(\"\n"); break;
case cmBC: mu::console() << _T("BRACKET \")\"\n"); break;
case cmIF: mu::console() << _T("IF\n"); break;
case cmELSE: mu::console() << _T("ELSE\n"); break;
case cmENDIF: mu::console() << _T("ENDIF\n"); break;
default: mu::console() << stOprt.top().GetCode() << _T(" "); break;
}
}
stOprt.pop();
}
mu::console() << dec << endl;
}
//------------------------------------------------------------------------------
/** \brief Evaluate an expression containing comma separated subexpressions
\param [out] nStackSize The total number of results available
\return Pointer to the array containing all expression results
This member function can be used to retrieve all results of an expression
made up of multiple comma separated subexpressions (i.e. "x+y,sin(x),cos(y)")
*/
value_type* ParserBase::Eval(int &nStackSize) const
{
(this->*m_pParseFormula)();
nStackSize = m_nFinalResultIdx;
// (for historic reasons the stack starts at position 1)
return &m_vStackBuffer[1];
}
//---------------------------------------------------------------------------
/** \brief Return the number of results on the calculation stack.
If the expression contains comma separated subexpressions (i.e. "sin(y), x+y").
There may be more than one return value. This function returns the number of
available results.
*/
int ParserBase::GetNumResults() const
{
return m_nFinalResultIdx;
}
//---------------------------------------------------------------------------
/** \brief Calculate the result.
A note on const correctness:
I consider it important that Calc is a const function.
Due to caching operations Calc changes only the state of internal variables with one exception
m_UsedVar this is reset during string parsing and accessible from the outside. Instead of making
Calc non const GetUsedVar is non const because it explicitly calls Eval() forcing this update.
\pre A formula must be set.
\pre Variables must have been set (if needed)
\sa #m_pParseFormula
\return The evaluation result
\throw ParseException if no Formula is set or in case of any other error related to the formula.
*/
value_type ParserBase::Eval() const
{
return (this->*m_pParseFormula)();
}
//---------------------------------------------------------------------------
void ParserBase::Eval(value_type *results, int nBulkSize)
{
/* <ibg 2014-09-24/> Commented because it is making a unit test impossible
// Parallelization does not make sense for fewer than 10000 computations
// due to thread creation overhead. If the bulk size is below 2000
// computation is refused.
if (nBulkSize<2000)
{
throw ParserError(ecUNREASONABLE_NUMBER_OF_COMPUTATIONS);
}
*/
CreateRPN();
int i = 0;
#ifdef MUP_USE_OPENMP
//#define DEBUG_OMP_STUFF
#ifdef DEBUG_OMP_STUFF
int *pThread = new int[nBulkSize];
int *pIdx = new int[nBulkSize];
#endif
int nMaxThreads = std::min(omp_get_max_threads(), s_MaxNumOpenMPThreads);
int nThreadID = 0, ct = 0;
omp_set_num_threads(nMaxThreads);
#pragma omp parallel for schedule(static, nBulkSize/nMaxThreads) private(nThreadID)
for (i=0; i<nBulkSize; ++i)
{
nThreadID = omp_get_thread_num();
results[i] = ParseCmdCodeBulk(i, nThreadID);
#ifdef DEBUG_OMP_STUFF
#pragma omp critical
{
pThread[ct] = nThreadID;
pIdx[ct] = i;
ct++;
}
#endif
}
#ifdef DEBUG_OMP_STUFF
FILE *pFile = fopen("bulk_dbg.txt", "w");
for (i=0; i<nBulkSize; ++i)
{
fprintf(pFile, "idx: %d thread: %d \n", pIdx[i], pThread[i]);
}
delete [] pIdx;
delete [] pThread;
fclose(pFile);
#endif
#else
for (i=0; i<nBulkSize; ++i)
{
results[i] = ParseCmdCodeBulk(i, 0);
}
#endif
}
} // namespace mu
#if __clang__
#pragma clang diagnostic pop
#endif