Shell Programming and Scripting

$ bash -c '(( $(expr 1.20 \> 1.3) == 1 )) && echo ok || echo ko' 
ko

bash -c 'case $(expr 1.20 \> 1.3) in (1) echo ok;; (*) echo ko;; esac'
ko

Like bash, expr only does integer arithmetic; you cannot use it with decimal fractions.

In any POSIX shell, you can strip the fraction and compare the integer portions. If they are equal, then compare the decimals:

Code:

if [ ${1%.*} -gt  ${2%.*} ]
then
  printf "%s is larger than %s\n" "$1" "$2"
elif [ ${1%.*} -eq  ${2%.*} ]
then
  : decimal comparison left as exercise for the reader 
else
  printf "%s is smaller than %s\n" "$1" "$2"
fi

There's no need to use an external program when it can be done many times faster with bash (or any POSIX shell).

When dealing with a string, an external program is many, many times slower than using shell internals.

Code:

$ x=123.456
$ time echo "${x%.*}"
123

real    0m0.000s
user    0m0.000s
sys     0m0.000s
$ time echo "$x" | awk -F. '{print $1}'
123

real    0m0.045s
user    0m0.002s
sys     0m0.001s
$ time echo "$x" | sed 's/\(.*\)\..*/\1/'
123

real    0m0.048s
user    0m0.000s
sys     0m0.004s

In the time it take to create the new process, a hundred (give or take) lines of shell code can be executed.

I have an 80-line shell function, fpmul, which multiplies floating point numbers.

Code:

$ time fpmul 12.34 56.78 90.12
63143.947824

real    0m0.000s
user    0m0.000s
sys     0m0.001s

Awk is much slower:

Code:

$ time awk 'BEGIN { print 12.34 * 56.78 * 90.12; exit }'
63143.9

real    0m0.033s
user    0m0.000s
sys     0m0.001s

That is exactly when a more efficient use of the shell is important.

On the contrary, complicated programs can be written in the shell that perform faster than those using another language.

There are times when an external program is better, especially dealing with large files.

Calling an external command to deal with a single string is the most inefficient code possible.

Write for the POSIX shell and there are no compatibility issues.

Write for bash or ksh when extra efficiency is needed for specific tasks.

Since most people will have some familiarity with the shell, it is faster to learn to use it well than to learn a new programming language.

# time awk -v v="$x" -F"." 'BEGIN{print $1}'

real    0m0.003s
user    0m0.001s
sys     0m0.001s

Code:

$ time awk -v v="$x" -F"." 'BEGIN{print $1}'


real    0m0.023s
user    0m0.000s
sys     0m0.002s

Not much different from the pipe, as a subshell adds very little time compared to that of a new process.

# time echo "$x" | awk -F. '{print $1}'
123

real    0m0.004s
user    0m0.001s
sys     0m0.003s

What maintenance? It's a black box function that was written once, many years ago and hasn't been touched since.

Its use is a one-liner.

$ time awk 'BEGIN { print 12.34 * 56.78 * 90.12; exit }'
63143.9

real    0m0.003s
user    0m0.000s
sys     0m0.002s

$ time echo "12.34 * 56.78 * 90.12" |bc
63143.47

real    0m0.003s
user    0m0.000s
sys     0m0.005s

No, I now write one line of code: fpmul ...

Absolute nonsense.

The shell is a very good programming language. It is the only one I need.

There are also things that python and perl don't do well.

File globbing and external commands are seamless in the shell; not so in other languages.

But ridiculously inefficient to use them on a single string, which I how I see them often used.

How often have you seen something like:

Code:

int=$(echo $x | awk -F. '{print $1}')
dec=$(echo $x | awk -F. '{print $2}')

Such coding can slow a script to a crawl.

Speed of execution is still very important.

The difference between a command that executes immediately and one that takes a second or two is the difference between a good user experience and a bad one.

Development of shell programs can be just as fast as writing perl or python, and just as legible.