entrolint

README

entrolint

Code rots toward disorder. entrolint measures the entropy — and keeps it from growing on every PR.

entrolint gauges code quality and maintainability through the metaphor of entropy from statistical physics. Clean, ordered, easy-to-change code is a system with low entropy. Tangled, tightly-coupled, hard-to-maintain code is a system with high entropy.

And just as in the second law of thermodynamics, the entropy of a codebase only grows when left unattended. entrolint's job is to measure it and keep it from growing unnoticed.

🔒 Runs entirely locally — no network calls, no telemetry, ever.

Why

Ordinary linters catch individual style violations. entrolint looks at the whole picture: it gives you a single number — entropy — that reflects how hard the code will be to maintain, and shows how that number changes with every change.

Two modes

scan — a heat map of the whole repo

Walks the codebase, computes an entropy score (S) for every file and package, and highlights hotspots — the places with the highest entropy. These are the first candidates for refactoring.

check — a pull-request gate

Computes ΔS (the entropy delta) between base and head: did the change raise disorder or lower it. It plugs into CI and blocks PRs that worsen maintainability beyond a configured threshold.

Positive ΔS = the code became harder to maintain.

Installation

go install github.com/pavlov061356/entrolint/cmd/entrolint@latest

Or download a prebuilt binary from the latest release.

Quick start

# The 10 hottest files in the repo:
entrolint scan --top 10

# PR gate: compare a feature branch against dev, fail if the threshold is exceeded.
entrolint check --base dev --head HEAD

# Machine-readable report for CI / a PR bot:
entrolint check --base origin/dev --head HEAD --format json > delta.json

scan prints a table with the columns PATH | S | T | DOMINANT. check prints a verdict line (PASS/FAIL) plus a per-file breakdown and returns exit code 1 when delta_s_max is exceeded.

Output formats

Both commands take --format:

Command	Formats	Notable use
scan	table (default), json, sarif	sarif → GitHub Code Scanning
check	table (default), json, markdown	markdown → a PR-comment body

(--json is a deprecated alias for --format json.)

Configuration

.entrolint.yaml at the repository root (optional — defaults are used when the file is absent):

weights:
  cyclomatic: 1.0
  nesting:    0.8
  coupling:   0.6
  length:     0.5
  duplication: 0.7
delta_s_max:       0.05   # ΔS_density threshold for check
churn_since_days:  90     # window for the churn factor (lives in T, not S)

Using entrolint in CI

entrolint ships a composite GitHub Action that posts a sticky PR comment with ΔS, the scaling class and the hottest changed files, and (optionally) uploads a SARIF log to GitHub Code Scanning. No Go toolchain is needed on the runner — the Action downloads the released binary.

# .github/workflows/entrolint.yml
name: entrolint
on:
  pull_request:
permissions:
  contents: read
  pull-requests: write     # post the PR comment
  security-events: write   # upload SARIF (optional)
jobs:
  entrolint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
        with:
          fetch-depth: 0
      - run: git fetch --no-tags origin "+refs/heads/${{ github.base_ref }}:refs/remotes/origin/${{ github.base_ref }}"
      - uses: pavlov061356/entrolint@v0
        with:
          comment: true
          upload-sarif: true
          fail-on-gate: false   # set true to block PRs that raise entropy

Input	Default	Description
version	latest	latest, a tag (v0.4.0), or source (build from the checkout).
base	PR base branch	Base ref for ΔS (origin/<base_ref> on PRs).
head	HEAD	Head ref.
config	.entrolint.yaml	Path to the config; defaults to .entrolint.yaml at the root.
comment	true	Post / update the sticky PR comment.
upload-sarif	false	Upload the scan SARIF to Code Scanning.
fail-on-gate	false	Fail the job when the ΔS / scaling gate trips.
github-token	github.token	Token used to post the PR comment and upload SARIF.

Fork PRs get a read-only GITHUB_TOKEN, so the comment step is skipped on PRs from forks (the Action stays on pull_request, not pull_request_target).

The entropy model

Entropy is a weighted sum of "microstates" — individual measurable factors of disorder. In the spirit of Boltzmann's formula S = k · ln(W): the more ways the code can be tangled, the higher the entropy.

Microstates that contribute to S:

Microstate	What it measures	Since
Cyclomatic complexity	how many branches there are in the code	v0.1
Nesting depth	how deeply blocks are nested	v0.1
Function / file length	the size of code units	v0.1
Coupling	how many import specs a file has	v0.3
Duplication	repeated AST subtrees within a file	v0.3

Churn (how often a file changes) feeds the temperature T = S · ξ(churn) — the hottest spots are not merely complex but also frequently rewritten.

Upcoming microstates and milestones are tracked in the ROADMAP.

Terminology

• Entropy (S) — a measure of disorder and maintenance difficulty. Higher = worse.
• ΔS — the change in entropy introduced by a pull request.
• Hotspot — a file or package with high entropy.
• Microstate — an individual factor contributing to S.
• Temperature — a file's normalized entropy, used for the heat map.

Status

📦 v0.4 (latest release) — CI integration: a drop-in entrolint-check GitHub Action that comments ΔS / scaling class / hotspots on a PR and uploads SARIF to Code Scanning, plus --format markdown|sarif output. Built on the v0.3 structural microstates (coupling, duplication) and the v0.2 predictive scaling class (O-class detectors shotgun, implementor_scan, switch_case_symmetry, identifier_fanout, state_multiplier emit a scaling_class line next to ΔS). The formula, weights, and threshold are considered unstable until v1.0 — S values may shift between releases.

License

MIT (see LICENSE).