This repository contains code related to the ORCA project.

Navigate to the github-metrics subdirectory. If you have not installed Gatsby, do so, and then run

npm install
gatsby clean
gatsby develop

To manually run data retrieval scripts, you can:

• Run sql/repos_in_papers.sql to aggregate GitHub references that appear in papers. If you do not want to update the software extrated from scholarly literature, skip this step.

• Prepare your development environment:

virtualenv venv
. venv/bin/activate
pip install -r requirements.txt
export GITHUB_ACCESS_TOKEN=your access token
export GITHUB_USER=your username

• Run PYTHONPATH='.' python3 scripts/retrieve_repos.py to retrieve a clean list of software to pull metadata from. You can run with the --query_bq flag to retrieve software that appears in the scholarly literature (if you are a CSET employee with BigQuery access) or --query_topics to retrieve software that matches the GitHub topics that appear in input_data/topics.txt.

• At this point, we will have full metadata for repos we retrieved using the github API (i.e. repos retrieved by topic, at the moment), but not for repos that only appear in papers or other sources. The next script grabs the default metadata retrieved from the github API for repos that don't already have it: PYTHONPATH='.' python3 scripts/backfill_top_level_repo_data.py

• Now, we can scrape some additional metadata from GitHub itself, including text of README.md files which we can use to do further analysis. Run PYTHONPATH='.' python3 scripts/retrieve_repo_metadata.py curr_repos_filled.jsonl curr_repos_final.jsonl

• To prepare data for the web application, load curr_repos_final.jsonl in the previous step into BigQuery and run the sequence of queries in sequences/downstream_order.txt. Assuming your dataset in BigQuery is called orca, download the data from orca.website_stats in BigQuery as JSONL, then run PYTHONPATH='.' python3 scripts/preprocess_for_website.py.

These steps are automated and run on a monthly basis on the scholarly literature data using the orca_data_pipeline.py Airflow pipeline.

• Total stars - This comes from the GitHub API (see staging_github_metrics.repos_with_full_meta_raw.full_metadata.stargazers_count)
• Total watchers - This comes from the GitHub API (see staging_github_metrics.repos_with_full_meta_raw.full_metadata.subscribers_count)
• Total contributors - This is scraped by retrieve_repo_metadata.py (TODO: maybe retrieve from gh archive)
• Total references - This comes from our scholarly literature (see paper_meta in website_stats.sql)
• Total open issues - This comes from the GitHub API (see staging_github_metrics.repos_with_full_meta_raw.full_metadata.open_issues)
• Created date - This comes from the GitHub API (see staging_github_metrics.repos_with_full_meta_raw.full_metadata.created_at)
• Last push date - This comes from the GitHub API (see staging_github_metrics.repos_with_full_meta_raw.full_metadata.pushed_at)
• License - This comes from the GitHub API (see staging_github_metrics.repos_with_full_meta_raw.full_metadata.license.name)
• Top programming language - This comes from the GitHub API (see staging_github_metrics.repos_with_full_meta_raw.full_metadata.language)
• Stars over time - This counts the number of WatchEvents for the project in the githubarchive BQ public dataset. The sum of these counts may not equal the total stars because users may unstar (or even unstar and re-star!) a repo
• Commits over time - This counts the number of distinct commits based on the unnested commits for PushEvents in the githubarchive BQ public dataset. See also push_event_commits.sql
• Issues over time - This counts the number of opened and closed issues based on IssuesEvents in the githubarchive BQ public dataset. See also issue_events.sql. We determine whether the issue was opened or closed based on the action field
• New versus returning contributors over time - this metric is based on commits. If a contributor makes their first commit during a given time interval, we mark them as a new contributor. Otherwise, they are a returning contributor. See combination of repo_pushes in website_stats.sql and get_new_vs_returning_contributor_counts in preprocess_for_website.py.
• Contribution percentages - this metric is based on commits. For each contributor, we count their number of commits, then calculate the percentage of commits written by each contributor. See combination of repo_pushes in website_stats.sql and get_cumulative_contributor_counts in preprocess_for_website.py.
• Cumulative percentage of contributions by number of contributors - This metric is based on commits. See description of Contribution percentages above, and github-metrics/src/components/summary_panel.js:getContribTrace
• The deps.dev links are added if the repo is present in bigquery-public-data.deps_dev_v1
• The pypi downloads over time come from bigquery-public-data.pypi