Slow output to nul:

[Matt Mahoney]

I found some strange behavior when testing output speed. The following program outputs the letter A 500,000 times to a file named as an argument, or to stdout by default.

// t.cpp
#include <stdio.h>
int main(int argc, char** argv) {
  FILE* out=stdout;
  if (argc==2) out=fopen(argv[1], "wb");
  for (int i=0; i<500000; ++i) putc('A', out);
  return 0;
}

The following commands all run in less than 0.05 seconds:

t x  (write to file x)
t nul: (discard output)
t > x (redirect stdout to file x)
copy x x1
copy x nul:
type x > x1
type x > nul:

But the following takes 14 seconds.

t > nul:

I get the same behavior whether I compile with g++ v4.6.1 (g++ -O3 -fomit-frame-pointer, also g++ with no optimization) or Microsoft 16.00.30319.01 (cl /O2 or cl). I'm testing under Vista SP2 32 bit on a 2.0 GHz T3200. I tried varying the output size, and time is proportional. Anyone else see this behavior?

Edit: same problem in Windows 7 Enterprise 32 bit, Intel Core i7 M620, 2.67 GHz, except it only takes 7 seconds because it's a faster machine.

Edit: the problem does NOT occur in Windows XP home edition (2002 SP3) 32 bit on a 2.2 GHz Athlon-64 3500+.

[schnaader]

Same problem here (Windows 7 Professional 64-Bit), 2.4 GHz, 15 seconds. But also note that with this change:

  for (int i=0; i<50000; ++i) {
    fprintf(out, "AAAAAAAAAA");
  }

it takes "only" 3 seconds. So I suppose this is some strange combination of flaws of shell pipes and using putc (instead of buffering), although it still is really strange (especially that it seems to work on XP). On my system, I can even see the speed difference between the original and the modified version when calling the program without any parameters, outputting to screen.

[Shelwien]

Its also slower on XP (0.2s instead of 0.02s).
But adding this:

setvbuf( out , NULL , _IOFBF , 1<<16 );

Seems to fix it.

[Matt Mahoney]

Yes, that works for me too. Thanks.