youdozi's memory
← Writing

Compare Dotted Numeric Versions in Shell with Python

youdozi

Do not compare 1.10 and 1.2 as strings or decimal numbers. Compare integer components from left to right and fill missing trailing components with zero; that makes 1.2 equal to 1.2.0. This example implements that rule for numeric dotted versions only.

It rejects v1.2, 1.2-rc1 and 1.2+build. For SemVer pre-release precedence, use a parser for that specification. Blindly removing suffixes can change the meaning.

A runnable comparison script

Python 3 is required. Save this as version-compare.sh and invoke it with sh.

version-compare.sh

#!/bin/sh
set -eu
python3 - "$@" <<'PYTHON'
import re
import sys
from itertools import zip_longest

if len(sys.argv) != 3 or any(not re.fullmatch(r"[0-9]+(?:\.[0-9]+)*", value) for value in sys.argv[1:]):
    print("usage: version-compare.sh CURRENT LATEST (numeric dotted versions)", file=sys.stderr)
    raise SystemExit(2)
current, latest = sys.argv[1:]
try:
    left = [int(part) for part in current.split('.')]
    right = [int(part) for part in latest.split('.')]
except ValueError:
    print("numeric component exceeds this Python runtime's conversion limit", file=sys.stderr)
    raise SystemExit(2)
comparison = 0
for a, b in zip_longest(left, right, fillvalue=0):
    if a != b:
        comparison = -1 if a < b else 1
        break
operator = '<' if comparison < 0 else '>' if comparison > 0 else '='
print(f"{current} {operator} {latest}")
PYTHON
sh version-compare.sh 1.10.5 1.3.1
sh version-compare.sh 1.2 1.2.0
sh version-compare.sh 1.1000 2.0

The outputs are 1.10.5 > 1.3.1, 1.2 = 1.2.0 and 1.1000 < 2.0. A valid comparison returns status 0 regardless of the relationship. Invalid argument counts, unsupported syntax or numeric conversion limits produce a stderr explanation and status 2.

Inputs Result
1.2.0.0.1 and 1.2 1.2.0.0.1 > 1.2
01.002 and 1.2 01.002 = 1.2
1.2-rc1 and 1.2 Unsupported format, status 2
1..2 and 1.2 Unsupported format, status 2

Why not concatenate fixed-width numbers?

A width such as awk’s %03d is a minimum display width, not a maximum component length. Packing those fields into a shell integer can lose the original ordering for large values or additional components.

This example converts each component to a Python integer and uses zip_longest(..., fillvalue=0) to align lengths. A runtime refusing the numeric conversion is not treated as a successful comparison.

In deployment automation, comparison and command success are separate decisions. This script does not fail just because the current version is lower; read the relationship from stdout before applying your policy. GitHub release tags need not use this numeric syntax, so do not pass them through blindly. For interpreter selection, see the POSIX sh function example.

More to read

Discussion