Codebase Analysis

​Challenge Day 2017

Sandy Suh,  Ezra Skolnik,  Suzi Curran,  Morgan Packard

What we did:

• Investigated the feasibility of doing static analysis of our codebase to inform our work
• Parsed git logs from the PHP repo to look at author counts, commit counts, and reverts on a file-by-file basis
• Explored correlating errors thrown in production to files and file metadata
• Mocked the sort of results we could expect to see from an investment in this type of analysis

What we learned:

• We have a treasure trove of data in Gitlab!
• Kibana's accessibility, lifespan, and formatting demand more investment
• Let the data do the driving: we are better served by looking at as many variables as possible and letting the numbers provide us with conclusions, rather than fishing for anything in particular
• While we need to be careful about our controls, there appear to be strong correlations in file metadata, ex:
  When regressing on both author count and commit count, a higher author count predicts fewer reverts

Example A: Authors to Reverts

# of Authors

Example B: Commits to Reverts

# of Commits

Kibana log batch exporter job:

https://git.csnzoo.com/mpackard/git-repo-analysis

Git repo analysis work:

https://git.csnzoo.com/wayfair/php/commits/challenge_day_17

Special Thanks To:

• Clover Bobker, for assistance with the Elasticsearch client
• The Challenge Day planning team, for their hard work