Using std::chrono::duration::rep with printf in 32bit and 64bit programs

Question

There is this code:

#include <cstdio>
#include <chrono>

int main()
{
  auto d = std::chrono::microseconds(1).count();
  printf("%lld", d);
  return 0;
}

When this is compiled in 64bit mode, then there is a warning:

main.cpp: In function ‘int main()’:
main.cpp:7:19: warning: format ‘%lld’ expects argument of type ‘long long int’, but argument 2 has type ‘long int’ [-Wformat=]
   printf("%lld", d);
                   ^

This warning is not present when compiling in 32bit mode (with -m32 flag). It looks like that std::chrono::duration::rep is of type long int in 64bit programs and long long int in 32bit programs.

Is there a portable way to print it like %zu specifier for size_t?

Answer 1

Instead of using the auto qualifier, use a fixed size integer int64_t.

#include <cstdio>
#include <chrono>
#include <cinttypes>

int main()
{
    int64_t d = std::chrono::microseconds(1).count();
    printf("%" PRId64 "\n", d);
    return 0;
}

Answer 2

A portable (i.e. C++) approach to consider, that does not use std::cout

 {
   // create a string:
   std::ostringstream ss; 
   ss << d;" 
   // then
   printf("%s", ss.str().c_str());
 } 

or perhaps

 {
   printf("%s", std::to_string(d).c_str() );
 }

Answer 3

I suggest you use std::cout, since you are in C++. This will be portable.

---

However, if you must use printf, change this:

printf("%lld", d);

to this:

#include <cinttypes>
printf("%" PRId64 "", d); 

Another approach would be to cast d to the highest data type (which can hold both types), like this:

printf("%lld", static_cast<long long int>(d));

Answer 4

You can cast it to long long int before printing:

#include <cstdio>
#include <chrono>

int main()
{
  auto d = std::chrono::microseconds(1).count();
  printf("%lld", static_cast<long long int>(d));
  return 0;
}

But it seems me that it is better to use std::cout

Answer 5

As you said that the usage of std::cout is not an option you can cast the value to the smallest needed data type¹ here it's long long int² and use the corresponding conversion specifier:

printf("%lld", static_cast<long long int>(d));

To avoid the explicit cast you can also use the data type directly instead of the auto specifier:

long long int d = std::chrono::microseconds(1).count();
printf("%lld", d);

---

¹ With smallest needed data type I mean the smallest type that can represent the value in both implementations.
² The long long int type has to be at least 64 bit wide, see here on SO.

Answer 6

To avoid the warning you can cast d to long long int.

printf("%lld", static_cast<long long int> (d));

Answer 7

Perhaps not directly related to the 32/64 bit problem at hand, but some of us are on embedded systems with odd output consoles and C++ libraries. (Plus we know that if we have to do any serious output formatting, printf is saner than iomanip!)

Anyway, this prints the guts of a duration and may be useful for debugging. Modify to taste.

template<typename Rep, typename Ratio>
printf_dur( std::chrono::duration< Rep, Ratio > dur )
{
    printf( "%lld ticks of %lld/%lld == %.3fs",
            (long long int) dur.count(),
            (long long int) Ratio::num,
            (long long int) Ratio::den,
            ( (Ratio::num == 1LL)
              ? (float) dur.count() / (float) Ratio::den
              : (float) dur.count() * (float) Ratio::num
            )
         );
}