/*
* Cppcheck - A tool for static C/C++ code analysis
* Copyright (C) 2007-2016 Cppcheck team.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program 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 General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see .
*/
#include "mathlib.h"
#include "errorlogger.h"
#include // VS2013: std::min and std::max
#include
#include
#include
#include
#if defined(_MSC_VER) && _MSC_VER <= 1700 // VS2012 doesn't have std::isinf and std::isnan
#define ISINF(x) (!_finite(x))
#define ISNAN(x) (_isnan(x))
#elif defined(__INTEL_COMPILER)
#define ISINF(x) (isinf(x))
#define ISNAN(x) (isnan(x))
#else // Use C++11 functions
#define ISINF(x) (std::isinf(x))
#define ISNAN(x) (std::isnan(x))
#endif
MathLib::value::value(const std::string &s) :
intValue(0), doubleValue(0), isUnsigned(false)
{
if (MathLib::isFloat(s)) {
type = MathLib::value::FLOAT;
doubleValue = MathLib::toDoubleNumber(s);
return;
}
if (!MathLib::isInt(s))
throw InternalError(0, "Invalid value");
type = MathLib::value::INT;
intValue = MathLib::toLongNumber(s);
if (isIntHex(s) && intValue < 0)
isUnsigned = true;
// read suffix
if (s.size() >= 2U) {
for (std::size_t i = s.size() - 1U; i > 0U; --i) {
char c = s[i];
if (c == 'u' || c == 'U')
isUnsigned = true;
else if (c == 'l' || c == 'L') {
if (type == MathLib::value::INT)
type = MathLib::value::LONG;
else if (type == MathLib::value::LONG)
type = MathLib::value::LONGLONG;
}
}
}
}
std::string MathLib::value::str() const
{
std::ostringstream ostr;
if (type == MathLib::value::FLOAT) {
if (ISNAN(doubleValue))
return "nan.0";
if (ISINF(doubleValue))
return (doubleValue > 0) ? "inf.0" : "-inf.0";
ostr.precision(9);
ostr << std::fixed << doubleValue;
// remove trailing zeros
std::string ret(ostr.str());
std::string::size_type pos = ret.size() - 1U;
while (ret[pos] == '0')
pos--;
if (ret[pos] == '.')
++pos;
return ret.substr(0, pos+1);
}
if (isUnsigned)
ostr << static_cast(intValue) << "U";
else
ostr << intValue;
if (type == MathLib::value::LONG)
ostr << "L";
else if (type == MathLib::value::LONGLONG)
ostr << "LL";
return ostr.str();
}
void MathLib::value::promote(const MathLib::value &v)
{
if (isInt() && v.isInt()) {
if (type < v.type) {
type = v.type;
isUnsigned = v.isUnsigned;
} else if (type == v.type) {
isUnsigned |= v.isUnsigned;
}
} else if (!isFloat()) {
isUnsigned = false;
doubleValue = intValue;
type = MathLib::value::FLOAT;
}
}
MathLib::value MathLib::value::calc(char op, const MathLib::value &v1, const MathLib::value &v2)
{
value temp(v1);
temp.promote(v2);
if (temp.isFloat()) {
switch (op) {
case '+':
temp.doubleValue += v2.getDoubleValue();
break;
case '-':
temp.doubleValue -= v2.getDoubleValue();
break;
case '*':
temp.doubleValue *= v2.getDoubleValue();
break;
case '/':
temp.doubleValue /= v2.getDoubleValue();
break;
case '%':
case '&':
case '|':
case '^':
throw InternalError(0, "Invalid calculation");
default:
throw InternalError(0, "Unhandled calculation");
}
} else if (temp.isUnsigned) {
switch (op) {
case '+':
temp.intValue += (unsigned long long)v2.intValue;
break;
case '-':
temp.intValue -= (unsigned long long)v2.intValue;
break;
case '*':
temp.intValue *= (unsigned long long)v2.intValue;
break;
case '/':
if (v2.intValue == 0)
throw InternalError(0, "Internal Error: Division by zero");
if (v1.intValue == std::numeric_limits::min() && std::abs(v2.intValue)<=1)
throw InternalError(0, "Internal Error: Division overflow");
temp.intValue /= (unsigned long long)v2.intValue;
break;
case '%':
if (v2.intValue == 0)
throw InternalError(0, "Internal Error: Division by zero");
temp.intValue %= (unsigned long long)v2.intValue;
break;
case '&':
temp.intValue &= (unsigned long long)v2.intValue;
break;
case '|':
temp.intValue |= (unsigned long long)v2.intValue;
break;
case '^':
temp.intValue ^= (unsigned long long)v2.intValue;
break;
default:
throw InternalError(0, "Unhandled calculation");
}
} else {
switch (op) {
case '+':
temp.intValue += v2.intValue;
break;
case '-':
temp.intValue -= v2.intValue;
break;
case '*':
temp.intValue *= v2.intValue;
break;
case '/':
if (v2.intValue == 0)
throw InternalError(0, "Internal Error: Division by zero");
if (v1.intValue == std::numeric_limits::min() && std::abs(v2.intValue)<=1)
throw InternalError(0, "Internal Error: Division overflow");
temp.intValue /= v2.intValue;
break;
case '%':
if (v2.intValue == 0)
throw InternalError(0, "Internal Error: Division by zero");
temp.intValue %= v2.intValue;
break;
case '&':
temp.intValue &= v2.intValue;
break;
case '|':
temp.intValue |= v2.intValue;
break;
case '^':
temp.intValue ^= v2.intValue;
break;
default:
throw InternalError(0, "Unhandled calculation");
}
}
return temp;
}
int MathLib::value::compare(const MathLib::value &v) const
{
value temp(*this);
temp.promote(v);
if (temp.isFloat()) {
if (temp.doubleValue < v.getDoubleValue())
return -1;
if (temp.doubleValue > v.getDoubleValue())
return 1;
return 0;
}
if (temp.isUnsigned) {
if ((unsigned long long)intValue < (unsigned long long)v.intValue)
return -1;
if ((unsigned long long)intValue > (unsigned long long)v.intValue)
return 1;
return 0;
}
if (intValue < v.intValue)
return -1;
if (intValue > v.intValue)
return 1;
return 0;
}
MathLib::value MathLib::value::add(int v) const
{
MathLib::value temp(*this);
if (temp.isInt())
temp.intValue += v;
else
temp.doubleValue += v;
return temp;
}
MathLib::value MathLib::value::shiftLeft(const MathLib::value &v) const
{
if (!isInt() || !v.isInt())
throw InternalError(0, "Shift operand is not integer");
MathLib::value ret(*this);
ret.intValue <<= v.intValue;
return ret;
}
MathLib::value MathLib::value::shiftRight(const MathLib::value &v) const
{
if (!isInt() || !v.isInt())
throw InternalError(0, "Shift operand is not integer");
MathLib::value ret(*this);
ret.intValue >>= v.intValue;
return ret;
}
MathLib::biguint MathLib::toULongNumber(const std::string & str)
{
// hexadecimal numbers:
if (isIntHex(str)) {
if (str[0] == '-') {
biguint ret = 0;
std::istringstream istr(str);
istr >> std::hex >> ret;
return ret;
} else {
unsigned long long ret = 0;
std::istringstream istr(str);
istr >> std::hex >> ret;
return (biguint)ret;
}
}
// octal numbers:
if (isOct(str)) {
biguint ret = 0;
std::istringstream istr(str);
istr >> std::oct >> ret;
return ret;
}
// binary numbers:
if (isBin(str)) {
biguint ret = 0;
for (std::string::size_type i = str[0] == '0'?2:3; i < str.length(); i++) {
ret <<= 1;
if (str[i] == '1')
ret |= 1;
}
/* if (str[0] == '-')
ret = -ret; */
return ret;
}
if (isFloat(str)) {
// Things are going to be less precise now: the value can't b represented in the biguint type.
// Use min/max values as an approximation. See #5843
const double doubleval = std::atof(str.c_str());
if (doubleval > (double)std::numeric_limits::max())
return std::numeric_limits::max();
else
return static_cast(doubleval);
}
biguint ret = 0;
std::istringstream istr(str);
istr >> ret;
return ret;
}
static unsigned int encodeMultiChar(const std::string& str)
{
unsigned int retval(str.front());
for (std::string::const_iterator it=str.begin()+1; it!=str.end(); ++it) {
retval = retval<<8 | *it;
}
return retval;
}
MathLib::bigint MathLib::characterLiteralToLongNumber(const std::string& str)
{
if (str.empty())
return 0; // for unit-testing...
// C99 6.4.4.4
// The value of an integer character constant containing more than one character (e.g., 'ab'),
// or containing a character or escape sequence that does not map to a single-byte execution character,
// is implementation-defined.
// clang and gcc seem to use the following encoding: 'AB' as (('A' << 8) | 'B')
const std::string& normStr = normalizeCharacterLiteral(str);
return encodeMultiChar(normStr);
}
std::string MathLib::normalizeCharacterLiteral(const std::string& iLiteral)
{
std::string normalizedLiteral;
const std::string::size_type iLiteralLen = iLiteral.size();
for (std::string::size_type idx = 0; idx < iLiteralLen ; ++idx) {
if (iLiteral[idx] != '\\') {
normalizedLiteral.push_back(iLiteral[idx]);
continue;
}
++idx;
if (idx == iLiteralLen) {
throw InternalError(0, "Internal Error. MathLib::normalizeCharacterLiteral: Unhandled char constant '" + iLiteral + "'.");
}
switch (iLiteral[idx]) {
case 'x':
// Hexa-decimal number: skip \x and interpret the next two characters
{
if (++idx == iLiteralLen)
throw InternalError(0, "Internal Error. MathLib::normalizeCharacterLiteral: Unhandled char constant '" + iLiteral + "'.");
std::string tempBuf;
tempBuf.push_back(iLiteral[idx]);
if (++idx != iLiteralLen)
tempBuf.push_back(iLiteral[idx]);
normalizedLiteral.push_back(static_cast(MathLib::toULongNumber("0x" + tempBuf)));
continue;
}
case 'u':
case 'U':
// Unicode string; just skip the \u or \U
if (idx + 1 == iLiteralLen)
throw InternalError(0, "Internal Error. MathLib::characterLiteralToLongNumber: Unhandled char constant '" + iLiteral + "'.");
continue;
}
// Single digit octal number
if (1 == iLiteralLen - idx) {
switch (iLiteral[idx]) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
normalizedLiteral.push_back(iLiteral[idx]-'0');
break;
case 'a':
normalizedLiteral.push_back('\a');
break;
case 'b':
normalizedLiteral.push_back('\b');
break;
case 'e':
normalizedLiteral.push_back(0x1B); // clang, gcc, tcc interpnormalizedLiteral this as 0x1B - escape character
break;
case 'f':
normalizedLiteral.push_back('\f');
break;
case 'n':
normalizedLiteral.push_back('\n');
break;
case 'r':
normalizedLiteral.push_back('\r');
break;
case 't':
normalizedLiteral.push_back('\t');
break;
case 'v':
normalizedLiteral.push_back('\v');
break;
case '\\':
case '\?':
case '\'':
case '\"':
normalizedLiteral.push_back(iLiteral[idx]);
break;
default:
throw InternalError(0, "Internal Error. MathLib::normalizeCharacterLiteral: Unhandled char constant '" + iLiteral + "'.");
}
continue;
}
// 2-3 digit octal number
if (!MathLib::isOctalDigit(iLiteral[idx]))
throw InternalError(0, "Internal Error. MathLib::normalizeCharacterLiteral: Unhandled char constant '" + iLiteral + "'.");
std::string tempBuf;
tempBuf.push_back(iLiteral[idx]);
++idx;
if (MathLib::isOctalDigit(iLiteral[idx])) {
tempBuf.push_back(iLiteral[idx]);
++idx;
if (MathLib::isOctalDigit(iLiteral[idx])) {
tempBuf.push_back(iLiteral[idx]);
}
}
normalizedLiteral.push_back(static_cast(MathLib::toLongNumber("0" + tempBuf)));
}
return normalizedLiteral;
}
MathLib::bigint MathLib::toLongNumber(const std::string & str)
{
// hexadecimal numbers:
if (isIntHex(str)) {
if (str[0] == '-') {
bigint ret = 0;
std::istringstream istr(str);
istr >> std::hex >> ret;
return ret;
} else {
unsigned long long ret = 0;
std::istringstream istr(str);
istr >> std::hex >> ret;
return (bigint)ret;
}
}
// octal numbers:
if (isOct(str)) {
bigint ret = 0;
std::istringstream istr(str);
istr >> std::oct >> ret;
return ret;
}
// binary numbers:
if (isBin(str)) {
bigint ret = 0;
for (std::string::size_type i = str[0] == '0'?2:3; i < str.length(); i++) {
ret <<= 1;
if (str[i] == '1')
ret |= 1;
}
if (str[0] == '-')
ret = -ret;
return ret;
}
if (isFloat(str)) {
// Things are going to be less precise now: the value can't be represented in the bigint type.
// Use min/max values as an approximation. See #5843
const double doubleval = toDoubleNumber(str);
if (doubleval > (double)std::numeric_limits::max())
return std::numeric_limits::max();
else if (doubleval < (double)std::numeric_limits::min())
return std::numeric_limits::min();
else
return static_cast(doubleval);
}
if (str[0] == '\'' && str.size() >= 3U && str[str.size()-1U] == '\'') {
return characterLiteralToLongNumber(str.substr(1,str.size()-2));
}
bigint ret = 0;
std::istringstream istr(str);
istr >> ret;
return ret;
}
double MathLib::toDoubleNumber(const std::string &str)
{
if (isIntHex(str))
return static_cast(toLongNumber(str));
// nullcheck
else if (isNullValue(str))
return 0.0;
#ifdef __clang__
else if (isFloat(str)) // Workaround libc++ bug at http://llvm.org/bugs/show_bug.cgi?id=17782
// TODO : handle locale
return std::strtod(str.c_str(), 0);
#endif
// otherwise, convert to double
std::istringstream istr(str);
istr.imbue(std::locale::classic());
double ret;
istr >> ret;
return ret;
}
template<> std::string MathLib::toString(double value)
{
std::ostringstream result;
result.precision(12);
result << value;
if (result.str() == "-0")
return "0.0";
if (result.str().find('.') == std::string::npos)
return result.str() + ".0";
return result.str();
}
bool MathLib::isFloat(const std::string &s)
{
return isDecimalFloat(s) || isFloatHex(s);
}
bool MathLib::isDecimalFloat(const std::string &s)
{
if (s.empty())
return false;
enum State {
START, BASE_PLUSMINUS, BASE_DIGITS1, LEADING_DECIMAL, TRAILING_DECIMAL, BASE_DIGITS2, E, MANTISSA_PLUSMINUS, MANTISSA_DIGITS, SUFFIX_F, SUFFIX_L
} state = START;
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it) {
switch (state) {
case START:
if (*it=='+' || *it=='-')
state=BASE_PLUSMINUS;
else if (*it=='.')
state=LEADING_DECIMAL;
else if (std::isdigit(static_cast(*it)))
state=BASE_DIGITS1;
else
return false;
break;
case BASE_PLUSMINUS:
if (*it=='.')
state=LEADING_DECIMAL;
else if (std::isdigit(static_cast(*it)))
state=BASE_DIGITS1;
else if (*it=='e' || *it=='E')
state=E;
else
return false;
break;
case LEADING_DECIMAL:
if (std::isdigit(static_cast(*it)))
state=BASE_DIGITS2;
else if (*it=='e' || *it=='E')
state=E;
else
return false;
break;
case BASE_DIGITS1:
if (*it=='e' || *it=='E')
state=E;
else if (*it=='.')
state=TRAILING_DECIMAL;
else if (!std::isdigit(static_cast(*it)))
return false;
break;
case TRAILING_DECIMAL:
if (*it=='e' || *it=='E')
state=E;
else if (*it=='f' || *it=='F')
state=SUFFIX_F;
else if (*it=='l' || *it=='L')
state=SUFFIX_L;
else if (std::isdigit(static_cast(*it)))
state=BASE_DIGITS2;
else
return false;
break;
case BASE_DIGITS2:
if (*it=='e' || *it=='E')
state=E;
else if (*it=='f' || *it=='F')
state=SUFFIX_F;
else if (*it=='l' || *it=='L')
state=SUFFIX_L;
else if (!std::isdigit(static_cast(*it)))
return false;
break;
case E:
if (*it=='+' || *it=='-')
state=MANTISSA_PLUSMINUS;
else if (std::isdigit(static_cast(*it)))
state=MANTISSA_DIGITS;
else
return false;
break;
case MANTISSA_PLUSMINUS:
if (!std::isdigit(static_cast(*it)))
return false;
else
state=MANTISSA_DIGITS;
break;
case MANTISSA_DIGITS:
if (*it=='f' || *it=='F')
state=SUFFIX_F;
else if (*it=='l' || *it=='L')
state=SUFFIX_L;
else if (!std::isdigit(static_cast(*it)))
return false;
break;
case SUFFIX_F:
return false;
case SUFFIX_L:
return false;
}
}
return (state==BASE_DIGITS2 || state==MANTISSA_DIGITS || state==TRAILING_DECIMAL || state==SUFFIX_F || state==SUFFIX_L);
}
bool MathLib::isNegative(const std::string &s)
{
if (s.empty())
return false;
return (s[0] == '-');
}
bool MathLib::isPositive(const std::string &s)
{
if (s.empty())
return false;
return !MathLib::isNegative(s);
}
/*! \brief Does the string represent an octal number?
* In case leading or trailing white space is provided, the function
* returns false.
* Additional information can be found here:
* http://gcc.gnu.org/onlinedocs/gcc/Binary-constants.html
*
* \param[in] s The string to check. In case the string is empty, the function returns false.
* \return Return true in case a octal number is provided and false otherwise.
**/
bool MathLib::isOct(const std::string& s)
{
enum Status {
START, PLUSMINUS, OCTAL_PREFIX, DIGITS
} state = START;
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it) {
switch (state) {
case START:
if (*it == '+' || *it == '-')
state = PLUSMINUS;
else if (*it == '0')
state = OCTAL_PREFIX;
else
return false;
break;
case PLUSMINUS:
if (*it == '0')
state = OCTAL_PREFIX;
else
return false;
break;
case OCTAL_PREFIX:
if (isOctalDigit(static_cast(*it)))
state = DIGITS;
else
return false;
break;
case DIGITS:
if (isOctalDigit(static_cast(*it)))
state = DIGITS;
else
return isValidIntegerSuffix(it,s.end());
break;
}
}
return state == DIGITS;
}
bool MathLib::isIntHex(const std::string& s)
{
enum Status {
START, PLUSMINUS, HEX_PREFIX, DIGIT, DIGITS
} state = START;
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it) {
switch (state) {
case START:
if (*it == '+' || *it == '-')
state = PLUSMINUS;
else if (*it == '0')
state = HEX_PREFIX;
else
return false;
break;
case PLUSMINUS:
if (*it == '0')
state = HEX_PREFIX;
else
return false;
break;
case HEX_PREFIX:
if (*it == 'x' || *it == 'X')
state = DIGIT;
else
return false;
break;
case DIGIT:
if (isxdigit(static_cast(*it)))
state = DIGITS;
else
return false;
break;
case DIGITS:
if (isxdigit(static_cast(*it)))
state = DIGITS;
else
return isValidIntegerSuffix(it,s.end());
break;
}
}
return state == DIGITS;
}
bool MathLib::isFloatHex(const std::string& s)
{
enum Status {
START, PLUSMINUS, HEX_PREFIX, WHOLE_NUMBER_DIGIT, WHOLE_NUMBER_DIGITS, FRACTION, EXPONENT_DIGIT, EXPONENT_DIGITS
} state = START;
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it) {
switch (state) {
case START:
if (*it == '+' || *it == '-')
state = PLUSMINUS;
else if (*it == '0')
state = HEX_PREFIX;
else
return false;
break;
case PLUSMINUS:
if (*it == '0')
state = HEX_PREFIX;
else
return false;
break;
case HEX_PREFIX:
if (*it == 'x' || *it == 'X')
state = WHOLE_NUMBER_DIGIT;
else
return false;
break;
case WHOLE_NUMBER_DIGIT:
if (isxdigit(static_cast(*it)))
state = WHOLE_NUMBER_DIGITS;
else
return false;
break;
case WHOLE_NUMBER_DIGITS:
if (isxdigit(static_cast(*it)))
state = WHOLE_NUMBER_DIGITS;
else if (*it=='.')
state = FRACTION;
else if (*it=='p' || *it=='P')
state = EXPONENT_DIGIT;
else
return false;
break;
case FRACTION:
if (isxdigit(static_cast(*it)))
state = FRACTION;
else if (*it=='p' || *it=='P')
state = EXPONENT_DIGIT;
break;
case EXPONENT_DIGIT:
if (isxdigit(static_cast(*it)))
state = EXPONENT_DIGITS;
else if (*it=='+' || *it=='-')
state = EXPONENT_DIGITS;
else
return false;
break;
case EXPONENT_DIGITS:
if (isxdigit(static_cast(*it)))
state = EXPONENT_DIGITS;
else
return *it=='f'||*it=='F'||*it=='l'||*it=='L';
break;
}
}
return state==EXPONENT_DIGITS;
}
bool MathLib::isValidIntegerSuffix(std::string::const_iterator it, std::string::const_iterator end)
{
enum {START, SUFFIX_U, SUFFIX_UL, SUFFIX_ULL, SUFFIX_L, SUFFIX_LU, SUFFIX_LL, SUFFIX_LLU, SUFFIX_I, SUFFIX_I6, SUFFIX_I64} state = START;
for (; it != end; ++it) {
switch (state) {
case START:
if (*it == 'u' || *it == 'U')
state = SUFFIX_U;
else if (*it == 'l' || *it == 'L')
state = SUFFIX_L;
else if (*it == 'i')
state = SUFFIX_I;
else
return false;
break;
case SUFFIX_U:
if (*it == 'l' || *it == 'L')
state = SUFFIX_UL; // UL
else
return false;
break;
case SUFFIX_UL:
if (*it == 'l' || *it == 'L')
state = SUFFIX_ULL; // ULL
else
return false;
break;
case SUFFIX_L:
if (*it == 'u' || *it == 'U')
state = SUFFIX_LU; // LU
else if (*it == 'l' || *it == 'L')
state = SUFFIX_LL; // LL
else
return false;
break;
case SUFFIX_LU:
return false;
case SUFFIX_LL:
if (*it == 'u' || *it == 'U')
state = SUFFIX_LLU; // LLU
else
return false;
break;
case SUFFIX_I:
if (*it == '6')
state = SUFFIX_I6;
else
return false;
break;
case SUFFIX_I6:
if (*it == '4')
state = SUFFIX_I64;
else
return false;
break;
default:
return false;
}
}
return ((state == SUFFIX_U) ||
(state == SUFFIX_L) ||
(state == SUFFIX_UL) ||
(state == SUFFIX_LU) ||
(state == SUFFIX_LL) ||
(state == SUFFIX_ULL) ||
(state == SUFFIX_LLU) ||
(state == SUFFIX_I64));
}
/*! \brief Does the string represent a binary number?
* In case leading or trailing white space is provided, the function
* returns false.
* Additional information can be found here:
* http://gcc.gnu.org/onlinedocs/gcc/Binary-constants.html
*
* \param[in] s The string to check. In case the string is empty, the function returns false.
* \return Return true in case a binary number is provided and false otherwise.
**/
bool MathLib::isBin(const std::string& s)
{
enum Status {
START, PLUSMINUS, GNU_BIN_PREFIX, DIGIT, DIGITS
} state = START;
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it) {
switch (state) {
case START:
if (*it == '+' || *it == '-')
state = PLUSMINUS;
else if (*it == '0')
state = GNU_BIN_PREFIX;
else
return false;
break;
case PLUSMINUS:
if (*it == '0')
state = GNU_BIN_PREFIX;
else
return false;
break;
case GNU_BIN_PREFIX:
if (*it == 'b' || *it == 'B')
state = DIGIT;
else
return false;
break;
case DIGIT:
if (*it == '0' || *it == '1')
state = DIGITS;
else
return false;
break;
case DIGITS:
if (*it == '0' || *it == '1')
state = DIGITS;
else
return isValidIntegerSuffix(it,s.end());
break;
}
}
return state == DIGITS;
}
bool MathLib::isDec(const std::string & s)
{
enum Status {
START, PLUSMINUS, DIGIT
} state = START;
for (std::string::const_iterator it = s.begin(); it != s.end(); ++it) {
switch (state) {
case START:
if (*it == '+' || *it == '-')
state = PLUSMINUS;
else if (isdigit(static_cast(*it)))
state = DIGIT;
else
return false;
break;
case PLUSMINUS:
if (isdigit(static_cast(*it)))
state = DIGIT;
else
return false;
break;
case DIGIT:
if (isdigit(static_cast(*it)))
state = DIGIT;
else
return isValidIntegerSuffix(it,s.end());
break;
}
}
return state == DIGIT;
}
bool MathLib::isInt(const std::string & s)
{
return isDec(s) || isIntHex(s) || isOct(s) || isBin(s);
}
static std::string intsuffix(const std::string & first, const std::string & second)
{
std::string suffix1, suffix2;
for (std::size_t i = 1U; i < first.size(); ++i) {
char c = first[first.size() - i];
if (c == 'l' || c == 'u')
c = c - 'a' + 'A';
if (c != 'L' && c != 'U')
break;
suffix1 = c + suffix1;
}
for (std::size_t i = 1U; i < second.size(); ++i) {
char c = second[second.size() - i];
if (c == 'l' || c == 'u')
c = c - 'a' + 'A';
if (c != 'L' && c != 'U')
break;
suffix2 = c + suffix2;
}
if (suffix1 == "ULL" || suffix2 == "ULL"
|| suffix1 == "LLU" || suffix2 == "LLU")
return "ULL";
if (suffix1 == "LL" || suffix2 == "LL")
return "LL";
if (suffix1 == "UL" || suffix2 == "UL"
|| suffix1 == "LU" || suffix2 == "LU")
return "UL";
if (suffix1 == "L" || suffix2 == "L")
return "L";
if (suffix1 == "U" || suffix2 == "U")
return "U";
return suffix1.empty() ? suffix2 : suffix1;
}
std::string MathLib::add(const std::string & first, const std::string & second)
{
#ifdef TEST_MATHLIB_VALUE
return (value(first) + value(second)).str();
#else
if (MathLib::isInt(first) && MathLib::isInt(second)) {
return toString(toLongNumber(first) + toLongNumber(second)) + intsuffix(first, second);
}
double d1 = toDoubleNumber(first);
double d2 = toDoubleNumber(second);
int count = 0;
while (d1 > 100000.0 * d2 && toString(d1+d2)==first && ++count<5)
d2 *= 10.0;
while (d2 > 100000.0 * d1 && toString(d1+d2)==second && ++count<5)
d1 *= 10.0;
return toString(d1 + d2);
#endif
}
std::string MathLib::subtract(const std::string &first, const std::string &second)
{
#ifdef TEST_MATHLIB_VALUE
return (value(first) - value(second)).str();
#else
if (MathLib::isInt(first) && MathLib::isInt(second)) {
return toString(toLongNumber(first) - toLongNumber(second)) + intsuffix(first, second);
}
if (first == second)
return "0.0" ;
double d1 = toDoubleNumber(first);
double d2 = toDoubleNumber(second);
int count = 0;
while (d1 > 100000.0 * d2 && toString(d1-d2)==first && ++count<5)
d2 *= 10.0;
while (d2 > 100000.0 * d1 && toString(d1-d2)==second && ++count<5)
d1 *= 10.0;
return toString(d1 - d2);
#endif
}
std::string MathLib::incdec(const std::string & var, const std::string & op)
{
#ifdef TEST_MATHLIB_VALUE
if (op == "++")
return value(var).add(1).str();
else if (op == "--")
return value(var).add(-1).str();
#else
if (op == "++")
return MathLib::add(var, "1");
else if (op == "--")
return MathLib::subtract(var, "1");
#endif
throw InternalError(0, std::string("Unexpected operation '") + op + "' in MathLib::incdec(). Please report this to Cppcheck developers.");
}
std::string MathLib::divide(const std::string &first, const std::string &second)
{
#ifdef TEST_MATHLIB_VALUE
return (value(first) / value(second)).str();
#else
if (MathLib::isInt(first) && MathLib::isInt(second)) {
const bigint a = toLongNumber(first);
const bigint b = toLongNumber(second);
if (b == 0)
throw InternalError(0, "Internal Error: Division by zero");
if (a == std::numeric_limits::min() && std::abs(b)<=1)
throw InternalError(0, "Internal Error: Division overflow");
return toString(toLongNumber(first) / b) + intsuffix(first, second);
} else if (isNullValue(second)) {
if (isNullValue(first))
return "nan.0";
return isPositive(first) ? "inf.0" : "-inf.0";
}
return toString(toDoubleNumber(first) / toDoubleNumber(second));
#endif
}
std::string MathLib::multiply(const std::string &first, const std::string &second)
{
#ifdef TEST_MATHLIB_VALUE
return (value(first) * value(second)).str();
#else
if (MathLib::isInt(first) && MathLib::isInt(second)) {
return toString(toLongNumber(first) * toLongNumber(second)) + intsuffix(first, second);
}
return toString(toDoubleNumber(first) * toDoubleNumber(second));
#endif
}
std::string MathLib::mod(const std::string &first, const std::string &second)
{
#ifdef TEST_MATHLIB_VALUE
return (value(first) % value(second)).str();
#else
if (MathLib::isInt(first) && MathLib::isInt(second)) {
const bigint b = toLongNumber(second);
if (b == 0)
throw InternalError(0, "Internal Error: Division by zero");
return toString(toLongNumber(first) % b) + intsuffix(first, second);
}
return toString(std::fmod(toDoubleNumber(first),toDoubleNumber(second)));
#endif
}
std::string MathLib::calculate(const std::string &first, const std::string &second, char action)
{
switch (action) {
case '+':
return MathLib::add(first, second);
case '-':
return MathLib::subtract(first, second);
case '*':
return MathLib::multiply(first, second);
case '/':
return MathLib::divide(first, second);
case '%':
return MathLib::mod(first, second);
case '&':
return MathLib::toString(MathLib::toLongNumber(first) & MathLib::toLongNumber(second)) + intsuffix(first,second);
case '|':
return MathLib::toString(MathLib::toLongNumber(first) | MathLib::toLongNumber(second)) + intsuffix(first,second);
case '^':
return MathLib::toString(MathLib::toLongNumber(first) ^ MathLib::toLongNumber(second)) + intsuffix(first,second);
default:
throw InternalError(0, std::string("Unexpected action '") + action + "' in MathLib::calculate(). Please report this to Cppcheck developers.");
}
}
std::string MathLib::sin(const std::string &tok)
{
return toString(std::sin(toDoubleNumber(tok)));
}
std::string MathLib::cos(const std::string &tok)
{
return toString(std::cos(toDoubleNumber(tok)));
}
std::string MathLib::tan(const std::string &tok)
{
return toString(std::tan(toDoubleNumber(tok)));
}
std::string MathLib::abs(const std::string &tok)
{
return toString(std::abs(toDoubleNumber(tok)));
}
bool MathLib::isEqual(const std::string &first, const std::string &second)
{
// this conversion is needed for formatting
// e.g. if first=0.1 and second=1.0E-1, the direct comparison of the strings would fail
return toString(toDoubleNumber(first)) == toString(toDoubleNumber(second));
}
bool MathLib::isNotEqual(const std::string &first, const std::string &second)
{
return !isEqual(first, second);
}
bool MathLib::isGreater(const std::string &first, const std::string &second)
{
return toDoubleNumber(first) > toDoubleNumber(second);
}
bool MathLib::isGreaterEqual(const std::string &first, const std::string &second)
{
return toDoubleNumber(first) >= toDoubleNumber(second);
}
bool MathLib::isLess(const std::string &first, const std::string &second)
{
return toDoubleNumber(first) < toDoubleNumber(second);
}
bool MathLib::isLessEqual(const std::string &first, const std::string &second)
{
return toDoubleNumber(first) <= toDoubleNumber(second);
}
/*! \brief Does the string represent the numerical value of 0?
* In case leading or trailing white space is provided, the function
* returns false.
* Requirement for this function:
* - This code is allowed to be slow because of simplicity of the code.
*
* \param[in] str The string to check. In case the string is empty, the function returns false.
* \return Return true in case the string represents a numerical null value.
**/
bool MathLib::isNullValue(const std::string &str)
{
if (str.empty() || (!std::isdigit(static_cast(str[0])) && (str.size() < 1 || (str[0] != '.' && str[0] != '-' && str[0] != '+'))))
return false; // Has to be a number
for (size_t i = 0; i < str.size(); i++) {
if (std::isdigit(static_cast(str[i])) && str[i] != '0') // May not contain digits other than 0
return false;
if (str[i] == 'E' || str[i] == 'e')
return true;
}
return true;
}
bool MathLib::isOctalDigit(char c)
{
return (c >= '0' && c <= '7');
}
bool MathLib::isDigitSeparator(const std::string& iCode, std::string::size_type iPos)
{
if (iPos == 0 || iPos >= iCode.size() || iCode[iPos] != '\'')
return false;
std::string::size_type i = iPos - 1;
while (std::isxdigit(iCode[i])) {
if (i == 0)
return true; // Only xdigits before '
--i;
}
if (i == iPos - 1) { // No xdigit before '
return false;
} else {
switch (iCode[i]) {
case ' ':
case '.':
case ',':
case 'x':
case '(':
case '{':
case '+':
case '-':
case '*':
case '%':
case '/':
case '&':
case '|':
case '^':
case '~':
case '=':
return true;
case '\'':
return isDigitSeparator(iCode, i);
default:
return false;
}
}
}
MathLib::value operator+(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('+',v1,v2);
}
MathLib::value operator-(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('-',v1,v2);
}
MathLib::value operator*(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('*',v1,v2);
}
MathLib::value operator/(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('/',v1,v2);
}
MathLib::value operator%(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('%',v1,v2);
}
MathLib::value operator&(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('&',v1,v2);
}
MathLib::value operator|(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('|',v1,v2);
}
MathLib::value operator^(const MathLib::value &v1, const MathLib::value &v2)
{
return MathLib::value::calc('^',v1,v2);
}
MathLib::value operator<<(const MathLib::value &v1, const MathLib::value &v2)
{
return v1.shiftLeft(v2);
}
MathLib::value operator>>(const MathLib::value &v1, const MathLib::value &v2)
{
return v1.shiftRight(v2);
}