How to parse a string to an int in C++?

Question

What's the C++ way of parsing a string (given as char *) into an int? Robust and clear error handling is a plus (instead of returning zero).

Answer 1

I think these three links sum it up:

• http://tinodidriksen.com/2010/02/07/cpp-convert-int-to-string-speed/
• http://tinodidriksen.com/2010/02/16/cpp-convert-string-to-int-speed/
• http://www.fastformat.org/performance.html

stringstream and lexical_cast solutions are about the same as lexical cast is using stringstream.

Some specializations of lexical cast use different approach see http://www.boost.org/doc/libs/release/boost/lexical_cast.hpp for details. Integers and floats are now specialized for integer to string conversion.

One can specialize lexical_cast for his/her own needs and make it fast. This would be the ultimate solution satisfying all parties, clean and simple.

Articles already mentioned show comparison between different methods of converting integers <-> strings. Following approaches make sense: old c-way, spirit.karma, fastformat, simple naive loop.

Lexical_cast is ok in some cases e.g. for int to string conversion.

Converting string to int using lexical cast is not a good idea as it is 10-40 times slower than atoi depending on the platform/compiler used.

Boost.Spirit.Karma seems to be the fastest library for converting integer to string.

ex.: generate(ptr_char, int_, integer_number);

and basic simple loop from the article mentioned above is a fastest way to convert string to int, obviously not the safest one, strtol() seems like a safer solution

int naive_char_2_int(const char *p) {
    int x = 0;
    bool neg = false;
    if (*p == '-') {
        neg = true;
        ++p;
    }
    while (*p >= '0' && *p <= '9') {
        x = (x*10) + (*p - '0');
        ++p;
    }
    if (neg) {
        x = -x;
    }
    return x;
}

Answer 2

I like Dan's answer, esp because of the avoidance of exceptions. For embedded systems development and other low level system development, there may not be a proper Exception framework available.

Added a check for white-space after a valid string...these three lines

    while (isspace(*end)) {
        end++;
    }

Added a check for parsing errors too.

    if ((errno != 0) || (s == end)) {
        return INCONVERTIBLE;
    }

Here is the complete function..

#include <cstdlib>
#include <cerrno>
#include <climits>
#include <stdexcept>

enum STR2INT_ERROR { SUCCESS, OVERFLOW, UNDERFLOW, INCONVERTIBLE };

STR2INT_ERROR str2long (long &l, char const *s, int base = 0)
{
    char *end = (char *)s;
    errno = 0;

    l = strtol(s, &end, base);

    if ((errno == ERANGE) && (l == LONG_MAX)) {
        return OVERFLOW;
    }
    if ((errno == ERANGE) && (l == LONG_MIN)) {
        return UNDERFLOW;
    }
    if ((errno != 0) || (s == end)) {
        return INCONVERTIBLE;
    }    
    while (isspace((unsigned char)*end)) {
        end++;
    }

    if (*s == '\0' || *end != '\0') {
        return INCONVERTIBLE;
    }

    return SUCCESS;
}

Answer 3

I know this is an older question, but I've come across it so many times and, to date, have still not found a nicely templated solution having the following characteristics:

• Can convert any base (and detect base type)
• Will detect erroneous data (i.e. ensure the entire string, less leading/trailing whitespace, is consumed by the conversion)
• Will ensure that, regardless of the type converted to, the range of the string's value is acceptable.

So, here is mine, with a test strap. Because it uses the C functions strtoull/strtoll under the hood, it always converts first to the largest type available. Then, if you are not using the largest type, it will perform additional range checks to verify your type was not over(under)flowed. For this, it is a little less performant than if one properly chose strtol/strtoul. However, it also works for shorts/chars and, to the best of my knowledge, there exists no standard library function that does that, too.

Enjoy; hopefully someone finds it useful.

#include <cstdlib>
#include <cerrno>
#include <limits>
#include <stdexcept>
#include <sstream>

static const int DefaultBase = 10;

template<typename T>
static inline T CstrtoxllWrapper(const char *str, int base = DefaultBase)
{
    while (isspace(*str)) str++; // remove leading spaces; verify there's data
    if (*str == '\0') { throw std::invalid_argument("str; no data"); } // nothing to convert

    // NOTE:  for some reason strtoull allows a negative sign, we don't; if
    //          converting to an unsigned then it must always be positive!
    if (!std::numeric_limits<T>::is_signed && *str == '-')
    { throw std::invalid_argument("str; negative"); }

    // reset errno and call fn (either strtoll or strtoull)
    errno = 0;
    char *ePtr;
    T tmp = std::numeric_limits<T>::is_signed ? strtoll(str, &ePtr, base)
                                              : strtoull(str, &ePtr, base);

    // check for any C errors -- note these are range errors on T, which may
    //   still be out of the range of the actual type we're using; the caller
    //   may need to perform additional range checks.
    if (errno != 0) 
    {
            if (errno == ERANGE) { throw std::range_error("str; out of range"); }
            else if (errno == EINVAL) { throw std::invalid_argument("str; EINVAL"); }
            else { throw std::invalid_argument("str; unknown errno"); }
    }

    // verify everything converted -- extraneous spaces are allowed
    if (ePtr != NULL)
    {
            while (isspace(*ePtr)) ePtr++;
            if (*ePtr != '\0') { throw std::invalid_argument("str; bad data"); }
    }

    return tmp;
}

template<typename T>
T StringToSigned(const char *str, int base = DefaultBase)
{
    static const long long max = std::numeric_limits<T>::max();
    static const long long min = std::numeric_limits<T>::min();

    long long tmp = CstrtoxllWrapper<typeof(tmp)>(str, base); // use largest type

    // final range check -- only needed if not long long type; a smart compiler
    //   should optimize this whole thing out
    if (sizeof(T) == sizeof(tmp)) { return tmp; }

    if (tmp < min || tmp > max)
    {
            std::ostringstream err;
            err << "str; value " << tmp << " out of " << sizeof(T) * 8
                << "-bit signed range (";
            if (sizeof(T) != 1) err << min << ".." << max;
            else err << (int) min << ".." << (int) max;  // don't print garbage chars
            err << ")";
            throw std::range_error(err.str());
    }

    return tmp;
}

template<typename T>
T StringToUnsigned(const char *str, int base = DefaultBase)
{
    static const unsigned long long max = std::numeric_limits<T>::max();

    unsigned long long tmp = CstrtoxllWrapper<typeof(tmp)>(str, base); // use largest type

    // final range check -- only needed if not long long type; a smart compiler
    //   should optimize this whole thing out
    if (sizeof(T) == sizeof(tmp)) { return tmp; }

    if (tmp > max)
    {
            std::ostringstream err;
            err << "str; value " << tmp << " out of " << sizeof(T) * 8
                << "-bit unsigned range (0..";
            if (sizeof(T) != 1) err << max;
            else err << (int) max;  // don't print garbage chars
            err << ")";
            throw std::range_error(err.str());
    }

    return tmp;
}

template<typename T>
inline T
StringToDecimal(const char *str, int base = DefaultBase)
{
    return std::numeric_limits<T>::is_signed ? StringToSigned<T>(str, base)
                                             : StringToUnsigned<T>(str, base);
}

template<typename T>
inline T
StringToDecimal(T &out_convertedVal, const char *str, int base = DefaultBase)
{
    return out_convertedVal = StringToDecimal<T>(str, base);
}

/*============================== [ Test Strap ] ==============================*/ 

#include <inttypes.h>
#include <iostream>

static bool _g_anyFailed = false;

template<typename T>
void TestIt(const char *tName,
            const char *s, int base,
            bool successExpected = false, T expectedValue = 0)
{
    #define FAIL(s) { _g_anyFailed = true; std::cout << s; }

    T x;
    std::cout << "converting<" << tName << ">b:" << base << " [" << s << "]";
    try
    {
            StringToDecimal<T>(x, s, base);
            // get here on success only
            if (!successExpected)
            {
                    FAIL(" -- TEST FAILED; SUCCESS NOT EXPECTED!" << std::endl);
            }
            else
            {
                    std::cout << " -> ";
                    if (sizeof(T) != 1) std::cout << x;
                    else std::cout << (int) x;  // don't print garbage chars
                    if (x != expectedValue)
                    {
                            FAIL("; FAILED (expected value:" << expectedValue << ")!");
                    }
                    std::cout << std::endl;
            }
    }
    catch (std::exception &e)
    {
            if (successExpected)
            {
                    FAIL(   " -- TEST FAILED; EXPECTED SUCCESS!"
                         << " (got:" << e.what() << ")" << std::endl);
            }
            else
            {
                    std::cout << "; expected exception encounterd: [" << e.what() << "]" << std::endl;
            }
    }
}

#define TEST(t, s, ...) \
    TestIt<t>(#t, s, __VA_ARGS__);

int main()
{
    std::cout << "============ variable base tests ============" << std::endl;
    TEST(int, "-0xF", 0, true, -0xF);
    TEST(int, "+0xF", 0, true, 0xF);
    TEST(int, "0xF", 0, true, 0xF);
    TEST(int, "-010", 0, true, -010);
    TEST(int, "+010", 0, true, 010);
    TEST(int, "010", 0, true, 010);
    TEST(int, "-10", 0, true, -10);
    TEST(int, "+10", 0, true, 10);
    TEST(int, "10", 0, true, 10);

    std::cout << "============ base-10 tests ============" << std::endl;
    TEST(int, "-010", 10, true, -10);
    TEST(int, "+010", 10, true, 10);
    TEST(int, "010", 10, true, 10);
    TEST(int, "-10", 10, true, -10);
    TEST(int, "+10", 10, true, 10);
    TEST(int, "10", 10, true, 10);
    TEST(int, "00010", 10, true, 10);

    std::cout << "============ base-8 tests ============" << std::endl;
    TEST(int, "777", 8, true, 0777);
    TEST(int, "-0111 ", 8, true, -0111);
    TEST(int, "+0010 ", 8, true, 010);

    std::cout << "============ base-16 tests ============" << std::endl;
    TEST(int, "DEAD", 16, true, 0xDEAD);
    TEST(int, "-BEEF", 16, true, -0xBEEF);
    TEST(int, "+C30", 16, true, 0xC30);

    std::cout << "============ base-2 tests ============" << std::endl;
    TEST(int, "-10011001", 2, true, -153);
    TEST(int, "10011001", 2, true, 153);

    std::cout << "============ irregular base tests ============" << std::endl;
    TEST(int, "Z", 36, true, 35);
    TEST(int, "ZZTOP", 36, true, 60457993);
    TEST(int, "G", 17, true, 16);
    TEST(int, "H", 17);

    std::cout << "============ space deliminated tests ============" << std::endl;
    TEST(int, "1337    ", 10, true, 1337);
    TEST(int, "   FEAD", 16, true, 0xFEAD);
    TEST(int, "   0711   ", 0, true, 0711);

    std::cout << "============ bad data tests ============" << std::endl;
    TEST(int, "FEAD", 10);
    TEST(int, "1234 asdfklj", 10);
    TEST(int, "-0xF", 10);
    TEST(int, "+0xF", 10);
    TEST(int, "0xF", 10);
    TEST(int, "-F", 10);
    TEST(int, "+F", 10);
    TEST(int, "12.4", 10);
    TEST(int, "ABG", 16);
    TEST(int, "10011002", 2);

    std::cout << "============ int8_t range tests ============" << std::endl;
    TEST(int8_t, "7F", 16, true, std::numeric_limits<int8_t>::max());
    TEST(int8_t, "80", 16);
    TEST(int8_t, "-80", 16, true, std::numeric_limits<int8_t>::min());
    TEST(int8_t, "-81", 16);
    TEST(int8_t, "FF", 16);
    TEST(int8_t, "100", 16);

    std::cout << "============ uint8_t range tests ============" << std::endl;
    TEST(uint8_t, "7F", 16, true, std::numeric_limits<int8_t>::max());
    TEST(uint8_t, "80", 16, true, std::numeric_limits<int8_t>::max()+1);
    TEST(uint8_t, "-80", 16);
    TEST(uint8_t, "-81", 16);
    TEST(uint8_t, "FF", 16, true, std::numeric_limits<uint8_t>::max());
    TEST(uint8_t, "100", 16);

    std::cout << "============ int16_t range tests ============" << std::endl;
    TEST(int16_t, "7FFF", 16, true, std::numeric_limits<int16_t>::max());
    TEST(int16_t, "8000", 16);
    TEST(int16_t, "-8000", 16, true, std::numeric_limits<int16_t>::min());
    TEST(int16_t, "-8001", 16);
    TEST(int16_t, "FFFF", 16);
    TEST(int16_t, "10000", 16);

    std::cout << "============ uint16_t range tests ============" << std::endl;
    TEST(uint16_t, "7FFF", 16, true, std::numeric_limits<int16_t>::max());
    TEST(uint16_t, "8000", 16, true, std::numeric_limits<int16_t>::max()+1);
    TEST(uint16_t, "-8000", 16);
    TEST(uint16_t, "-8001", 16);
    TEST(uint16_t, "FFFF", 16, true, std::numeric_limits<uint16_t>::max());
    TEST(uint16_t, "10000", 16);

    std::cout << "============ int32_t range tests ============" << std::endl;
    TEST(int32_t, "7FFFFFFF", 16, true, std::numeric_limits<int32_t>::max());
    TEST(int32_t, "80000000", 16);
    TEST(int32_t, "-80000000", 16, true, std::numeric_limits<int32_t>::min());
    TEST(int32_t, "-80000001", 16);
    TEST(int32_t, "FFFFFFFF", 16);
    TEST(int32_t, "100000000", 16);

    std::cout << "============ uint32_t range tests ============" << std::endl;
    TEST(uint32_t, "7FFFFFFF", 16, true, std::numeric_limits<int32_t>::max());
    TEST(uint32_t, "80000000", 16, true, std::numeric_limits<int32_t>::max()+1);
    TEST(uint32_t, "-80000000", 16);
    TEST(uint32_t, "-80000001", 16);
    TEST(uint32_t, "FFFFFFFF", 16, true, std::numeric_limits<uint32_t>::max());
    TEST(uint32_t, "100000000", 16);

    std::cout << "============ int64_t range tests ============" << std::endl;
    TEST(int64_t, "7FFFFFFFFFFFFFFF", 16, true, std::numeric_limits<int64_t>::max());
    TEST(int64_t, "8000000000000000", 16);
    TEST(int64_t, "-8000000000000000", 16, true, std::numeric_limits<int64_t>::min());
    TEST(int64_t, "-8000000000000001", 16);
    TEST(int64_t, "FFFFFFFFFFFFFFFF", 16);
    TEST(int64_t, "10000000000000000", 16);

    std::cout << "============ uint64_t range tests ============" << std::endl;
    TEST(uint64_t, "7FFFFFFFFFFFFFFF", 16, true, std::numeric_limits<int64_t>::max());
    TEST(uint64_t, "8000000000000000", 16, true, std::numeric_limits<int64_t>::max()+1);
    TEST(uint64_t, "-8000000000000000", 16);
    TEST(uint64_t, "-8000000000000001", 16);
    TEST(uint64_t, "FFFFFFFFFFFFFFFF", 16, true, std::numeric_limits<uint64_t>::max());
    TEST(uint64_t, "10000000000000000", 16);

    std::cout << std::endl << std::endl
              << (_g_anyFailed ? "!! SOME TESTS FAILED !!" : "ALL TESTS PASSED")
              << std::endl;

    return _g_anyFailed;
}

StringToDecimal is the user-land method; it is overloaded so it can be called either like this:

int a; a = StringToDecimal<int>("100");

or this:

int a; StringToDecimal(a, "100");

I hate repeating the int type, so prefer the latter. This ensures that if the type of 'a' changes one does not get bad results. I wish the compiler could figure it out like:

int a; a = StringToDecimal("100");

...but, C++ does not deduce template return types, so that's the best I can get.

The implementation is pretty simple:

CstrtoxllWrapper wraps both strtoull and strtoll, calling whichever is necessary based on the template type's signed-ness and providing some additional guarantees (e.g. negative input is disallowed if unsigned and it ensures the entire string was converted).

CstrtoxllWrapper is used by StringToSigned and StringToUnsigned with the largest type (long long/unsigned long long) available to the compiler; this allows the maximal conversion to be performed. Then, if it is necessary, StringToSigned/StringToUnsigned performs the final range checks on the underlying type. Finally, the end-point method, StringToDecimal, decides which of the StringTo* template methods to call based on the underlying type's signed-ness.

I think most of the junk can be optimized out by the compiler; just about everything should be compile-time deterministic. Any commentary on this aspect would be interesting to me!

Answer 4

This is a safer C way than atoi()

const char* str = "123";
int i;

if(sscanf(str, "%d", &i)  == EOF )
{
   /* error */
}

C++ with standard library stringstream: (thanks CMS )

int str2int (const string &str) {
  stringstream ss(str);
  int num;
  if((ss >> num).fail())
  { 
      //ERROR 
  }
  return num;
}

With boost library: (thanks jk)

#include <boost/lexical_cast.hpp>
#include <string>

try
{
    std::string str = "123";
    int number = boost::lexical_cast< int >( str );
}
catch( const boost::bad_lexical_cast & )
{
    // Error
}

Edit: Fixed the stringstream version so that it handles errors. (thanks to CMS's and jk's comment on original post)

Answer 5

The good 'old C way still works. I recommend strtol or strtoul. Between the return status and the 'endPtr', you can give good diagnostic output. It also handles multiple bases nicely.

Answer 6

I like Dan Moulding's answer, I'll just add a bit of C++ style to it:

#include <cstdlib>
#include <cerrno>
#include <climits>
#include <stdexcept>

int to_int(const std::string &s, int base = 0)
{
    char *end;
    errno = 0;
    long result = std::strtol(s.c_str(), &end, base);
    if (errno == ERANGE || result > INT_MAX || result < INT_MIN)
        throw std::out_of_range("toint: string is out of range");
    if (s.length() == 0 || *end != '\0')
        throw std::invalid_argument("toint: invalid string");
    return result;
}

It works for both std::string and const char* through the implicit conversion. It's also useful for base conversion, e.g. all to_int("0x7b") and to_int("0173") and to_int("01111011", 2) and to_int("0000007B", 16) and to_int("11120", 3) and to_int("3L", 34); would return 123.

Unlike std::stoi it works in pre-C++11. Also unlike std::stoi, boost::lexical_cast and stringstream it throws exceptions for weird strings like "123hohoho".

NB: This function tolerates leading spaces but not trailing spaces, i.e. to_int(" 123") returns 123 while to_int("123 ") throws exception. Make sure this is acceptable for your use case or adjust the code.

Such function could be part of STL...