EDIT Since c++17, some parts of the standard library were removed. Fortunately, starting with c++11, we have lambdas which are a superior solution.

#include <algorithm> 
#include <cctype>
#include <locale>

// trim from start (in place)
static inline void ltrim(std::string &s) {
    s.erase(s.begin(), std::find_if(s.begin(), s.end(), [](unsigned char ch) {
        return !std::isspace(ch);
    }));
}

// trim from end (in place)
static inline void rtrim(std::string &s) {
    s.erase(std::find_if(s.rbegin(), s.rend(), [](unsigned char ch) {
        return !std::isspace(ch);
    }).base(), s.end());
}

// trim from both ends (in place)
static inline void trim(std::string &s) {
    ltrim(s);
    rtrim(s);
}

// trim from start (copying)
static inline std::string ltrim_copy(std::string s) {
    ltrim(s);
    return s;
}

// trim from end (copying)
static inline std::string rtrim_copy(std::string s) {
    rtrim(s);
    return s;
}

// trim from both ends (copying)
static inline std::string trim_copy(std::string s) {
    trim(s);
    return s;
}

Thanks to https://stackoverflow.com/a/44973498/524503 for bringing up the modern solution.

Original answer:

I tend to use one of these 3 for my trimming needs:

#include <algorithm> 
#include <functional> 
#include <cctype>
#include <locale>

// trim from start
static inline std::string &ltrim(std::string &s) {
    s.erase(s.begin(), std::find_if(s.begin(), s.end(),
            std::not1(std::ptr_fun<int, int>(std::isspace))));
    return s;
}

// trim from end
static inline std::string &rtrim(std::string &s) {
    s.erase(std::find_if(s.rbegin(), s.rend(),
            std::not1(std::ptr_fun<int, int>(std::isspace))).base(), s.end());
    return s;
}

// trim from both ends
static inline std::string &trim(std::string &s) {
    return ltrim(rtrim(s));
}

They are fairly self-explanatory and work very well.

EDIT: BTW, I have std::ptr_fun in there to help disambiguate std::isspace because there is actually a second definition which supports locales. This could have been a cast just the same, but I tend to like this better.

EDIT: To address some comments about accepting a parameter by reference, modifying and returning it. I Agree. An implementation that I would likely prefer would be two sets of functions, one for in place and one which makes a copy. A better set of examples would be:

#include <algorithm> 
#include <functional> 
#include <cctype>
#include <locale>

// trim from start (in place)
static inline void ltrim(std::string &s) {
    s.erase(s.begin(), std::find_if(s.begin(), s.end(),
            std::not1(std::ptr_fun<int, int>(std::isspace))));
}

// trim from end (in place)
static inline void rtrim(std::string &s) {
    s.erase(std::find_if(s.rbegin(), s.rend(),
            std::not1(std::ptr_fun<int, int>(std::isspace))).base(), s.end());
}

// trim from both ends (in place)
static inline void trim(std::string &s) {
    ltrim(s);
    rtrim(s);
}

// trim from start (copying)
static inline std::string ltrim_copy(std::string s) {
    ltrim(s);
    return s;
}

// trim from end (copying)
static inline std::string rtrim_copy(std::string s) {
    rtrim(s);
    return s;
}

// trim from both ends (copying)
static inline std::string trim_copy(std::string s) {
    trim(s);
    return s;
}

I am keeping the original answer above though for context and in the interest of keeping the high voted answer still available.

For iterating backwards see this answer.

Iterating forwards is almost identical. Just change the iterators / swap decrement by increment. You should prefer iterators. Some people tell you to use std::size_t as the index variable type. However, that is not portable. Always use the size_type typedef of the container (While you could get away with only a conversion in the forward iterating case, it could actually go wrong all the way in the backward iterating case when using std::size_t, in case std::size_t is wider than what is the typedef of size_type):

Using std::vector

Using iterators

for(std::vector<T>::iterator it = v.begin(); it != v.end(); ++it) {
    /* std::cout << *it; ... */
}

Important is, always use the prefix increment form for iterators whose definitions you don't know. That will ensure your code runs as generic as possible.

Using Range C++11

for(auto const& value: a) {
     /* std::cout << value; ... */

Using indices

for(std::vector<int>::size_type i = 0; i != v.size(); i++) {
    /* std::cout << v[i]; ... */
}

Using arrays

Using iterators

for(element_type* it = a; it != (a + (sizeof a / sizeof *a)); it++) {
    /* std::cout << *it; ... */
}

Using Range C++11

for(auto const& value: a) {
     /* std::cout << value; ... */

Using indices

for(std::size_t i = 0; i != (sizeof a / sizeof *a); i++) {
    /* std::cout << a[i]; ... */
}

Read in the backward iterating answer what problem the sizeof approach can yield to, though.