How to use C++ std::ostream with printf-like forma

Question

I am learning C++. cout is an instance of std::ostream class. How can I print a formatted string with it?

I can still use printf, but I want to learn a proper C++ method which can take advantage of all C++ benefits. I think this should be possible with std::ostream, but I can't find the proper way.

Answer 1

Field Width

Setting field width is very simple. For each variable, simply precede it with "setw(n)". Like this:

#include <iostream>
#include <iomanip>

using namespace std;

int main()
{
  const int max = 12;
  const int width = 6;
  for(int row = 1; row <= max; row++) {
      for(int col = 1; col <= max; col++) {
          cout << setw(width) << row * col;
      }
      cout << endl;
  }
  return 0;
}

Notice how "setw(n)" controls the field width, so each number is printed inside a field that stays the same width regardless of the width of the number itself.

-- From "Programming/C++ tutorial" by P. Lutus.

Answer 2

I wrote independently but came up with answer similar to user3283405

My solution uses vasprintf() to acheive formatting, and uses operator overloading of << of std::ostream to free the memory in right place.

Usage:

std::cout << putf(const char *format, ...); //Same format as C printf(3)

Code:

#define _GNU_SOURCE
#include <cstdarg>
#include <iostream>
#include <cstdio>

struct putf_r{
        char *s;
};

putf_r putf(const char *fmt, ...){
        va_list ap;
        va_start(ap, fmt);
        putf_r a;
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wformat-nonliteral"
        vasprintf(&a.s, fmt, ap);
#pragma GCC diagnostic pop
        va_end(ap);
        return a;
}

std::ostream& operator<<(std::ostream& os, putf_r a){
        os<<a.s;
        free(a.s);
        return os;
}

int main(){
        std::cout << putf("%3d\n", 23) << putf("%a\n", 256.);
}

Note that compiler doesn't check format inside putf(), so compiler flag -Wformat-nonliteral will not warn for suspicious code in putf() and you need to care uncontrolled format string problem by yourself.
Detailed info can be found on GitHub

Answer 3

This is an idiom I have gotten used to. Hopefully it helps:

// Hacky but idiomatic printf style syntax with c++ <<

#include <cstdlib> // for sprintf

char buf[1024]; sprintf(buf, "%d score and %d years ago", 4, 7);
cout << string(buf) <<endl;

&

Answer 4

To implement printf one could use c++11 template parameters:

#include <iostream>
#include <string>

inline std::ostream & mprintf(std::ostream & ostr, const char * fstr) throw()
{
    return ostr << fstr;
}

template<typename T, typename... Args> 
std::ostream & mprintf(std::ostream & ostr, 
        const char * fstr, const T & x) throw()
{
    size_t i=0;
    char c = fstr[0];

    while (c != '%')
    {
        if(c == 0) return ostr; // string is finished
        ostr << c;
        c = fstr[++i];
    };
    c = fstr[++i];
    ostr << x;

    if(c==0) return ostr; // 

    // print the rest of the stirng
    ostr << &fstr[++i];
    return ostr;
}


template<typename T, typename... Args> 
std::ostream & mprintf(std::ostream & ostr,
        const char * fstr, const T & x, Args... args) throw()
{
    size_t i=0;
    char c = fstr[0];

    while (c != '%')
    {
        if(c == 0) return ostr; // string is finished
        ostr << c;
        c = fstr[++i];
    };
    c = fstr[++i];
    ostr << x;

    if(c==0) return ostr; // string is finished

    return mprintf(ostr, &fstr[++i], args...);
}

int main()
{
    int c = 50*6;
    double a = 34./67.;
    std::string q = "Hello!";

    // put only two arguments
    // the symbol after % does not matter at all
    mprintf(std::cout, "%f + %f = %a \n", c, a);

    // print string object: for real printf one should write q.c_str() 
    mprintf(std::cout, "message: \"%s\". \n", q);

    // the last argument will be ignored
    mprintf(std::cout, "%z + %f\n", (long)a, 12, 544 );

}

Output

300 + 2 = %a 
message: "Hello!". 
2 + 12

This a very simple code and it can be improved.

1) The advantage is that it uses << to print objects to the stream, so you can put arbitrary arguments that can be output via <<.

2) It ignores the type of the argument in the formatted string: after % can stand arbitrary symbol even a space. The output stream decides how to print the corresponding object. It also compatible with printf.

3) A disadvantage is that it can not print the percent symbol '%', one need to slightly improve the code.

4) It can not print formatted numbers, like %4.5f

5) If the number of arguments is less than predicted by formatted string, then the function just print the rest of the string.

6) If the number of arguments is greater than predicted by formatted string, then the remained arguments are ignored

One can improve the code to make 2)-6) to fully mimic the printf behaviour. However, if you follow the rules of printf, then only 3) and 4) need essentially to be fixed.

Answer 5

Sample output:

2017-12-20T16:24:47,604144+01:00 Hello, World!

Code (with put_printf usage demonstrated in put_timestamp):

#include <assert.h>
#include <chrono>
#include <iomanip>
#include <iostream>

class put_printf {
    static constexpr size_t failed = std::numeric_limits<size_t>::max(); // for any explicit error handling
    size_t stream_size; // excluding '\0'; on error set to 0 or to "failed"
    char buf_stack[2048+1]; // MAY be any size that fits on the stack (even 0), SHOULD be (just) large enough for most uses (including '\0')
    std::unique_ptr<char[]> buf_heap; // only used if the output doesn't fit in buf_stack
public:
    explicit put_printf(const char *format, ...)
            #if __GNUC__
            __attribute__ ((format (printf, 2, 3))) // most compelling reason for not using a variadic template; parameter 1 is implied "this"
            #endif
            {
        va_list args;
        va_start(args, format);
        const int res = vsnprintf(buf_stack, sizeof(buf_stack), format, args);
        va_end(args);
        if (res < 0) { // easily provoked, e.g., with "%02147483646i\n", i.e., more than INT_MAX-1 significant characters (only observed, no guarantee seen)
            stream_size = failed;
        } else if (res < sizeof(buf_stack)) { // preferred path
            stream_size = res;
        } else { // not artificially constrained
            try {
                const size_t buf_size = static_cast<size_t>(res) + 1; // avoids relying on "res < INT_MAX" (only observed, no guarantee seen)
                buf_heap.reset(new char[buf_size]); // observed to work even beyond INT_MAX=2^32-1 bytes
                va_start(args, format);
                if (vsnprintf(buf_heap.get(), buf_size, format, args) == res) stream_size = res;
                else stream_size = failed; // can't happen
                va_end(args);
            } catch (const std::bad_alloc&) { // insufficient free heap space (or an environment-specific constraint?)
                stream_size = failed;
            }
        }
    }
    friend std::ostream& operator<<(std::ostream& os, const put_printf& self) {
        if (self.stream_size == failed) {
            // (placeholder for any explicit error handling)
            return os;
        } else {
            // using write() rather than operator<<() to avoid a separate scan for '\0' or unintentional truncation at any internal '\0' character
            return os.write((self.buf_heap ? self.buf_heap.get() : self.buf_stack), self.stream_size);
        }
    }
};

class put_timestamp {
    const bool basic = false;
    const bool local = true;
public:
    friend std::ostream& operator<<(std::ostream& os, const put_timestamp& self) {
        const auto now = std::chrono::system_clock::now();
        const std::time_t now_time_t = std::chrono::system_clock::to_time_t(now);
        struct tm tm; if ((self.local ? localtime_r(&now_time_t, &tm) : gmtime_r(&now_time_t, &tm)) == nullptr) return os; // TODO: explicit error handling?
        static_assert(4 <= sizeof(int), "");
        const int microseconds = std::chrono::duration_cast<std::chrono::microseconds>(now.time_since_epoch() % std::chrono::seconds(1)).count();
        assert(0 <= microseconds && microseconds < 1000000); // TODO: (how) do we know?
        // TODO: doesn't "point" in "decimal_point()" imply "dot"/"full stop"/"period", unlike an obviously neutral term like "mark"/"separator"/"sign"?
        const char decimal_sign = std::use_facet<std::numpunct<char>>(os.getloc()).decimal_point() == '.' ? '.' : ','; // full stop accepted, comma preferred
        // TODO: all well and good for a locale-specific decimal sign, but couldn't the locale also upset microseconds formatting by grouping digits?
        os << std::put_time(&tm, self.basic ? "%Y%m%dT%H%M%S" : "%FT%T") << put_printf("%c%06i", decimal_sign, microseconds);
        if (! self.local) return os << "Z";
        const int tz_minutes = std::abs(static_cast<int>(tm.tm_gmtoff)) / 60;
        return os << put_printf(self.basic ? "%c%02i%02i" : "%c%02i:%02i", 0 <= tm.tm_gmtoff ? '+' : '-', tz_minutes / 60, tz_minutes % 60);
    }
};

int main() {
    // testing decimal sign
    ///std::cout.imbue(std::locale("en_GB"));
    ///std::cout.imbue(std::locale("fr_FR"));

    std::cout << put_timestamp() << " Hello, World!\n";
    #if 0
    typedef put_printf pf; // just to demo local abbreviation
    std::cout << "1: " << pf("%02147483646i\n"  , 1     ) << std::endl; // res < 0
    std::cout << "2: " << pf("%02147483643i%i\n", 1, 100) << std::endl; // res < 0
    std::cout << "3: " << pf("%02147483643i%i\n", 1,  10) << std::endl; // works
    std::cout << "4: " << pf("%02147483646i"    , 1     ) << std::endl; // works
    #endif
    return 0;
}

Comments about put_printf:

// Reasons for the name "put_printf" (and not "putf" after all):
// - put_printf is self-documenting, while using the naming pattern also seen in std::put_time;
// - it is not clear whether the proposed std::putf would support exactly the same format syntax;
// - it has a niche purpose, so a longer name is not an objection, and for frequent local uses
//     it is easy enough to declare an even shorter "typedef put_printf pf;" or so.
// Evaluation of delegating to vsnprintf() with intermediate buffer:
// (+) identical result without implementation and/or maintenance issues,
// (?) succeeds or fails as a whole, no output of successful prefix before point of failure
// (-) (total output size limited to INT_MAX-1)
// (-) overhead (TODO: optimal buf_stack size considering cache and VM page locality?)
// Error handling (an STL design problem?):
// - std::cout.setstate(std::ios_base::failbit) discards further std::cout output (stdout still works),
//     so, to be aware of an error in business logic yet keep on trucking in diagnostics,
//     should there be separate classes, or a possibility to plug in an error handler, or what?
// - should the basic or default error handling print a diagnostic message? throw an exception?
// TODO: could a function "int ostream_printf(std::ostream& os, const char *format, ...)"
//           first try to write directly into os.rdbuf() before using buf_stack and buf_heap,
//           and would that significantly improve performance or not?

Answer 6

I suggest using ostringstream instead of ostream see following example :

#include <vector>
#include <string>
#include <iostream>
#include "CppUnitTest.h"

#define _CRT_NO_VA_START_VALIDATION

std::string format(const std::string& format, ...)
{
    va_list args;
    va_start(args, format);
    size_t len = std::vsnprintf(NULL, 0, format.c_str(), args);
    va_end(args);
    std::vector<char> vec(len + 1);
    va_start(args, format);
    std::vsnprintf(&vec[0], len + 1, format.c_str(), args);
    va_end(args);
    return &vec[0];
}

example usage:

std::ostringstream ss;
ss << format("%s => %d", "Version", Version) << std::endl;
Logger::WriteMessage(ss.str().c_str()); // write to unit test output
std::cout << ss.str() << std::endl; // write to standard output