Documentation › CODE_COUNT

Code Statistics

tools/count_code.py builds a filtered list of project files and passes it to the selected counter. cloc is the default and preserves the historical report format. scc adds complexity, COCOMO, and LOCOMO estimates. The default configuration is tools/codecount.json.

Run it from the repository root:

python3 tools/count_code.py
python3 tools/count_code.py --languages
python3 tools/count_code.py --format json --output build/code-count.json
python3 tools/count_code.py --format markdown --output build/code-count.md
python3 tools/count_code.py --jobs 8
python3 tools/count_code.py --no-progress --format json
python3 tools/count_code.py --engine scc
python3 tools/count_code.py --engine scc --cocomo-project-type semi-detached \
  --avg-wage 75000 --overhead 2.4 --eaf 1.0 --locomo-preset local
python3 tools/count_code.py --history --history-chart build/code-growth.svg
python3 tools/count_code.py --history --format json --output build/code-history.json
python3 tools/count_code.py --history --format markdown --output build/code-history.md

--format markdown produces the complete human-readable report: totals, directory groups, language groups, and when --engine scc is used, COCOMO and LOCOMO estimates. It is appropriate for CI summaries and review artifacts.

GitHub Actions

The Code Statistics workflow is intentionally manual-only. In GitHub, open Actions, select Code Statistics, then choose Run workflow. It recursively checks out every Git submodule, runs this tool with the repository exclusions, and writes the complete Markdown report to the workflow's standard job summary by writing directly to $GITHUB_STEP_SUMMARY. The summary is rendered directly in the workflow run page and is not dependent on a generated or uploaded report file.

The standard exclusions cover build products and temporary directories such as build/, dist/, .git/, pycache/, buildsystem/deps/, and buildsystem/tmp/. Common binary and compiler-generated file suffixes are also excluded. The repository configuration additionally excludes the optional third_party/llama2.c component and third_party/cmd, because those trees are not part of LeonOS's code-size statistics. These path exclusions are recursive, so all files below each directory are omitted.

Adjust persistent exclusions in tools/codecount.json:

{
  "exclude": ["third_party/example", "system/generated/**"],
  "exclude_dirs": ["fixtures"],
  "exclude_files": ["*_generated.c"],
  "exclude_languages": ["Markdown"],
  "include_languages": []
}

Command-line exclusions apply only to the current run:

python3 tools/count_code.py --exclude userland/apps/doom --exclude-file '*.snap'
python3 tools/count_code.py --exclude-dir test --group-depth 2

By default, the tool starts up to four worker threads. The report always keeps top-level parts such as userland and third_party separate. Large parts are internally split into bounded batches, so a large dependency tree or the userland sources do not monopolize one worker. Use --shard-threshold and --shard-size to tune this behavior. Results are merged back into the same part before the final report. Use --jobs 1 to force a single worker, or choose a larger value for a fast machine. Progress is sent to stderr so JSON/text output on stdout stays machine-readable; use --no-progress to suppress it.

Use --no-config to ignore the repository configuration, or --no-default-excludes when a complete diagnostic scan of generated files is needed. cloc must be installed and available in PATH; use --cloc to select a specific executable. For scc, install the upstream tool and use --engine scc or --scc PATH. scc receives the same explicit filtered file list and ignores repository-local ignore files, so codecount.json remains the single source of truth for exclusions.

The scc COCOMO values use Basic COCOMO coefficients. --avg-wage is an annual wage, --overhead is the non-salary multiplier, and --eaf is the effort adjustment factor. LOCOMO is a rough LLM code-regeneration estimate; its large, medium, small, and local presets describe only pricing and output throughput, not a promise about any specific model.

Git history growth is available with --history. It performs one git log --numstat scan, so it is substantially faster than checking out every commit and running cloc repeatedly. The default follows the first-parent history and reports cumulative physical text lines after the same path and file exclusions used by the normal report. --history-all-branches includes non-merge commits from all refs, but its cumulative line series should be read as aggregate change activity rather than one branch's exact tree size. Use --history-chart FILE.svg to create a dependency-free SVG line chart; JSON and text output remain available through --format and --output.

The host-side regression test uses a temporary fixture and does not modify the repository:

python3 tools/test_code_count.py