Random yes experiments

A series of implementations of the simplest form of yes, outputting an endless loop of y\n.

Inspired by, and C solutions based on, this Reddit post.

hsbyte based on this Reddit comment.

Usage

To compile, simply run

make

To run a specific test, run

./speedtest.sh [test]

To see the list of compiled tests, just run

./speedtest.sh

To test GNU yes, or the yes installed on your system, run

yes | pv > /dev/null

Results

These results are completely not rigorous. Each test was only run once for a total of 5 seconds each, which I believe was sufficient enough to see the improvements in each.

The percentages for each speed is of the speed of the intial test of GNU yes.

GNU yes

yes

5.53 GiB/s

Bash scripts

sh.sh (echo "y")

593 KiB/s (0.01%)—There is a lot of overhead from each call to echo.

shunroll.sh (echo "y\ny\ny\n...")

40.1 MiB/s (0.71%)—Less overhead due to less calls to echo, since each call prints multiple lines of y\n.

Haskell (ghc)

hs

12.3 MiB/s (0.21%)

hslist (infinite list)

9.07 MiB/s (0.16%)—Overhead of list usage.

hsbuf (manual flush every 8192)

196 MiB/s (3.46%)—Vast improvements from manually flushing and disabling flushing upon every newline.

hsbyte (use ByteString.Char 8)

4.08 GiB/s (73.78%)—Extreme improvements from using ByteString instead (based on this Reddit comment).

C (gcc -O3 -g)

c (printf)

138 MiB/s (2.44%)

cputs (puts)

203 MiB/s (3.58%)—Savings from not doing unnecessary formatting.

cbuf (write with buffer size 8192)

5.51 GiB/s (99.64%)—Extreme improvements from taking a leaf out of GNU yes's book—we write directly using buffer size of 8192.

Prerequisites

• bash
• ghc
• gcc
• make
• pv
• yes (maybe)