Shellmath
Introducing floating-point arithmetic libraries for the Bash shell, because they said it couldn't be done... and because:
.
Quick-start guide
Download this project and source the file shellmath.sh into your shell script,
then fire away at the shellmath API!
The basic API looks like this:
_shellmath_add arg1 arg2 [...] argN
_shellmath_subtract arg1 arg2 # means arg1 - arg2
_shellmath_multiply arg1 arg2 [...] argN
_shellmath_divide arg1 arg2 # means arg1 / arg2
The extended API introduces one more function:
_shellmath_getReturnValue arg
This function optimizes away the need for $( subshelling ) in order to capture shellmath's output.
To use this feature, just be sure to set __shellmath_isOptimized=1 at the top
of your script. (You can find an example in faster_e_demo.sh.)
Operands to the shellmath functions can be integers or decimal numbers presented in either standard or scientific notation:
_shellmath_add 1.009 4.223e-2
_shellmath_getReturnValue sum
echo "The sum is $sum"
Addition and multiplication are of arbitrary arity; try this on for size:
_shellmath_multiply 1 2 3 4 5 6
_shellmath_getReturnValue sixFactorial
echo "6 factorial is $sixFactorial"
Subtraction and division, OTOH, are exclusively binary operations.
The demos
For a gentle introduction to shellmath run the demo slower_e_demo.sh
with a small whole-number argument, say 15:
$ slower_e_demo.sh 15
e = 2.7182818284589936
This script uses a few shellmath API calls to calculate e, the mathematical
constant also known as Euler's number. The argument
15 tells the script to evaluate the 15th-degree Maclaurin polynomial for e.
(That's the Taylor polynomial centered at 0.) Take a look inside the script to
see how it uses the shellmath APIs.
There is another demo script very much like this one but different, and the sensitive user can feel the difference. Try the following, but don't blink or you'll miss it ;)
$ faster_e_demo.sh 15
e = 2.7182818284589936
Did you feel the difference? Try the -t option with both scripts; this will produce
timing statistics. Here are my results
when running from my minGW64 command prompt on Windows 10 with an Intel i3 Core CPU:
$ for n in {1..5}; do faster_e_demo.sh -t 15 2>&1; done | awk '/^real/ {print $2}'
0m0.055s
0m0.051s
0m0.056s
0m0.054s
0m0.054s
$ for n in {1..5}; do slower_e_demo.sh -t 15 2>&1; done | awk '/^real/ {print $2}'
0m0.498s
0m0.594s
0m0.536s
0m0.511s
0m0.580s
(When sizing up these timings, do keep in mind that we are timing the calculation of e from its Maclaurin polynomial. Every invocation of either script is exercising the shellmath arithmetic subroutines 31 times.)
The comment header in faster_e_demo.sh explains the optimization and shows
how to put this faster version to work for you.
Runtime efficiency competitive with awk and bc
The file timingData.txt captures the results of some timing experiments that compare
shellmath against the GNU versions of the calculators awk and bc. The experiments
exercised each of the arithmetic operations and captured the results in a shell variable.
The result summary below shows that shellmath is competitive with awk and runs faster
than bc in these experiments. (One commenter noted that the differences in execution speed
can be partially explained by the fact that shellmath and awk use finite precision
whereas bc uses arbitrary precision. Another factor in these measurements is the need to
subshell 'awk' and 'bc' to capture their results, whereas 'shellmath' writes directly to
the shell's global memory.)
Here are the run times of shellmath as a percentage of the awk and bc equivalents:
versus awk versus bc
Addition: 82.2% 40.6%
Subtraction: 95.9% 50.5%
Multiplication: 135.9% 73.3%
Division: 80.3% 43.2%
Astute observers will note the experiments provide approximations to the sum, difference, product, and quotient of pi and e. Unfortunately I did not gain insight as to which of these values, if any, are transcendental.
You can find a deeper discussion of shellmath's runtime efficiency here.
Background
The Bash shell does not have built-in operators for decimal arithmetic, making it something of an oddity among well-known, widely-used programming languages. For the most part, practitioners in need of powerful computational building blocks have naturally opted for other languages and tools. Their widespread availability has diverted attention from the possibility of implementing decimal arithmetic in Bash and it's easy to assume that this cannot be done:
- From the indispensable Bash FAQ (on Greg's Wiki): How can I calculate with floating point numbers?
"For most operations... an external program must be used." - From Mendel Cooper's wonderful and encyclopedic Advanced Bash Scripting Guide:
Bash does not understand floating point arithmetic. Use bc instead. - From a community discussion on Stack Overflow, How do I use floating point division in bash?
The user's preferred answer is a good example of prevailing thought on this subject.
Meanwhile,
- Bash maintainer (BDFL?) Chet Ramey sounds a (brighter?) note in The Bash Reference Guide, Section 6.5 by emphasizing what the built-in arithmetic operators can do.
But finally, a glimmer of hope:
- A diamond-in-the-rough buried elsewhere
on Stack Overflow.
This down-and-dirty milestone computes the decimal quotient of two integer arguments. At a casual glance, it seems to have drawn inspiration from the Euclidean algorithm for computing GCDs, an entirely different approach thanshellmath's.
Please try shellmath on for size and draw your own conclusions!
How it works
shellmath splits decimal numbers into their integer and fractional parts,
performs the appropriate integer operations on the parts, and recombines the results.
(In the spirit of Bash, numerical overflow is silently ignored.)
Because if we can get carrying, borrowing, place value, and the distributive law right, then the sky's the limit! As they say--erm, as they said in Rome,
Ad astra per aspera.
And now...
You can run your floating-point calculations directly in Bash!