Cross Platform Support for sprintf's Format '-Flag

The Single UNIX Specification Version 2 specifies the sprintf's format '-flag behavior as:

The integer portion of the result of a decimal conversion (%i, %d, %u, %f, %g or %G) will be formatted with thousands' grouping characters[1]

I can't find the format '-flag in the or the specifications. g++ even warns:

ISO C++11 does not support the ' printf flag

The flag is not recognized to even warn about in Visual C; printf("%'d", foo) outputs:


I'd like to be able to write C-standard compliant code that uses the behavior of the format '-flag. Thus the answer I'm looking for one of the following:

  1. C-Standard specification of the format '-flag
  2. A cross platform compatible extrapolation of gcc's format '-flag
  3. Demonstration that a cross platform extrapolation is not possible


  • Standard C doesn't provide the formatting capability directly, but it does provide the ability to retrieve a...specification of what the formatting should be, on a locale-specific basis. So, it's up to you to retrieve the locale's specification of proper formatting, then put it to use to format your data (but even then, it's somewhat non-trivial). For example, here's a version for formatting long data:

    #include <stdlib.h>
    #include <locale.h>
    #include <string.h>
    #include <limits.h>
    static int next_group(char const **grouping) {
        if ((*grouping)[1] == CHAR_MAX)
            return 0;
        if ((*grouping)[1] != '\0')
        return **grouping;
    size_t commafmt(char   *buf,            /* Buffer for formatted string  */
                    int     bufsize,        /* Size of buffer               */
                    long    N)              /* Number to convert            */
        int i;
        int len = 1;
        int posn = 1;
        int sign = 1;
        char *ptr = buf + bufsize - 1;
        struct lconv *fmt_info = localeconv();
        char const *tsep = fmt_info->thousands_sep;
        char const *group = fmt_info->grouping;
        // char const *neg = fmt_info->negative_sign;
        size_t sep_len = strlen(tsep);
        size_t group_len = strlen(group);
        // size_t neg_len = strlen(neg);
        int places = (int)*group;
        if (bufsize < 2)
            *buf = '\0';
            return 0;
        *ptr-- = '\0';
        if (N < 0L)
            sign = -1;
            N = -N;
        for ( ; len <= bufsize; ++len, ++posn)
            *ptr-- = (char)((N % 10L) + '0');
            if (0L == (N /= 10L))
            if (places && (0 == (posn % places)))
                places = next_group(&group);
                for (int i=sep_len; i>0; i--) {
                    *ptr-- = tsep[i-1];
                    if (++len >= bufsize)
                        goto ABORT;
            if (len >= bufsize)
                goto ABORT;
        if (sign < 0)
            if (len >= bufsize)
                goto ABORT;
            *ptr-- = '-';
        memmove(buf, ++ptr, len + 1);
        return (size_t)len;
    #ifdef TEST
    #include <stdio.h>
    #define elements(x) (sizeof(x)/sizeof(x[0]))
    void show(long i) {
        char buffer[32];
        commafmt(buffer, sizeof(buffer), i);
        printf("%s\n", buffer);
        commafmt(buffer, sizeof(buffer), -i);
        printf("%s\n", buffer);
    int main() {
        long inputs[] = {1, 12, 123, 1234, 12345, 123456, 1234567, 12345678 };
        for (int i=0; i<elements(inputs); i++) {
            setlocale(LC_ALL, "");
        return 0;

    This does have a bug (but one I'd consider fairly minor). On two's complement hardware, it won't convert the most-negative number correctly, because it attempts to convert a negative number to its equivalent positive number with N = -N; In two's complement, the maximally negative number doesn't have a corresponding positive number, unless you promote it to a larger type. One way to get around this is by promoting the number the corresponding unsigned type (but it's is somewhat non-trivial).

    Implementing the same for other integer types is fairly trivial. For floating point types is a bit more work. Converting floating point types (even without formatting) correctly is enough more work that for them, I'd at least consider using something like sprintf to do the conversion, then inserting the formatting into the string that produced.