lefed

Extracting source code metrics with shell scripting

08 Dec 2013 / posted by Nikku

Recently I was looking at a project and wanted to capture a number of code metrics to see how the project developed over time. It turns out some interesting metrics can easily be extracted with basic shell scripting.

I wanted to figure out, whether the project grew or shrinked over time and how that relates to test code and the number of test cases.

A few metrics that I was interested in were

  • source code size (lines of code and files)
  • test code size (lines of code and files)
  • number of tests

What I came up with is number of bash snippets that collected the information form me.

Bash for the win

Given I wanted to scan the current directory . for source files matching *.java we may implement the metrics as follows:

Number of Files

find . -type f -name "*.java" | wc -l

Lines of Code (LOC)

find . -type f -name "*.java" \
  -exec cat {} \; | sed '/^\s*$/d' | wc -l

This excludes empty lines (sed '/^\s*$/d').

Number of Tests

find . -type f -name "*.java" \
  -exec cat {} \; | grep @Test | wc -l

It scans for unit tests (in my case indentified by a @Test annotation).

Over Time Metrics

Based on the three metrics, over time data an be collected by browsing through the projects versioning history. I combined the above snippets in a small script that checks out a number of commits to collect the metrics of a project over a series of commits.

It logs commit_hash, src_files, test_files, src_loc, test_loc and num_tests as comma separated values into a specified file. To use it, collect all commits you would like to analyse via git rev-list first_commit..last_commit first.

This results in a CSV file with all information in place, ready to be visualized.

Summary

As it turned out, shell scripting is an appropriate environment do do quick and dirty extraction of source code metrics. The performance is well enough for a nightly job, too. On my developer machine for around 90k lines of code project, the scanning takes aproximately five seconds per commit.

Happy analyzing!