Why does it take so long to print "\n" in Perl?

Question

Why does it take so long to print a newline? Is this just my machine, or do others see the same effect?

With the newline:

#!/usr/bin/perl

use strict;
use Benchmark;

   timethis(100000,'main();');


   sub main {
      print "you are the bomb. \n";
   }


   # outputs: 
   # timethis 100000:  8 wallclock secs ( 0.15 usr +  0.45 sys =  0.60 CPU) @ 166666.67/s (n=100000)

W/o the newline:

#!/usr/bin/perl

use strict;
use Benchmark;

   timethis(100000,'main();');


   sub main {
      print "you are the bomb. ";
   }


   # outputs:
   # timethis 100000:  0 wallclock secs ( 0.09 usr +  0.04 sys =  0.13 CPU) @ 769230.77/s (n=100000)
   #     (warning: too few iterations for a reliable count)

Edit: I'd like to add that placing two "\n" causes the execution to take twice as long, at least for wallclock seconds.

timethis 100000: 16 wallclock secs ( 0.15 usr +  0.52 sys =  0.67 CPU) @ 149253.73/s (n=100000)

Answer 1

Newline flushes output.

In most stdio implementations, buffering varies with the type of output device ... Serial devices, including terminals, modems, mice, and joysticks, are normally line-buffered; stdio sends the entire line out only when it gets the newline

Answer 2

It's not the \n per se that causes this problem. Rather, successive calls to print are buffered by the OS until the \n character is encountered or the buffer is full. At that point, the output buffer is flushed to the screen. Flushing the output to the screen is a (relatively) expensive operation, so the loop in which you flush the output buffer many times has much slower performance than the loop in which you only flush the buffer once at the end (which happens implicitly when your program exits).

Answer 3

I don't think buffering has much to do with it. I'm guessing it's because the terminal needs to scroll when you print a newline to it (or print enough characters to fill a line). When I benchmark these functions writing to a file or to /dev/null, there is not much of a difference.

use Benchmark;
timethis(1000000, 'main');
timethis(1000000, 'main2');
select STDERR; $| = 0; select STDOUT;  # enable buffering on STDERR
sub main { print STDERR "you are the bomb. \n" }
sub main2 { print STDERR "you are the bomb. " }

$ perl benchmark.pl 2> a_file
timethis 1000000: 21 wallclock secs ( 4.67 usr + 13.38 sys = 18.05 CPU) @ 55410.87/s
timethis 1000000: 21 wallclock secs ( 4.91 usr + 13.34 sys = 18.25 CPU) @ 54797.52/s

$ perl benchmark.pl 2> /dev/null
timethis 1000000: 26 wallclock secs ( 2.86 usr + 10.36 sys = 13.22 CPU) @ 75648.69/s
timethis 1000000: 27 wallclock secs ( 2.86 usr + 10.30 sys = 13.16 CPU) @ 76010.95/s

$ perl benchmark.pl 2> a_file     (without buffering)
timethis 1000000: 29 wallclock secs ( 3.78 usr + 12.14 sys = 15.92 CPU) @ 62806.18/s
timethis 1000000: 29 wallclock secs ( 3.27 usr + 12.51 sys = 15.78 CPU) @ 63367.34/s

$ perl benchmark.pl 2> /dev/tty   (window has 35 lines and buffers 10000, YMMV)
[ 200000 declarations of how you are a bomb deleted ]
timethis 100000: 53 wallclock secs ( 0.98 usr +  3.73 sys =  4.72 CPU) @ 21190.93/s
timethis 100000:  9 wallclock secs ( 0.36 usr +  1.94 sys =  2.30 CPU) @ 43535.05/s

Summary: extra flushing reduces performance by about 10%. Extra scrolling on the terminal reduces performance by about 50%.