ci/github-script/lint-commits: error when conventional commit format is used

E.g. https://redirect.github.com/NixOS/nixpkgs/pull/495442

ci/github-script/get-pr-commit-details.js

@@ -23,6 +23,11 @@ async function runGit({ args, repoPath, core, quiet }) {
 }
 
 /**
+ * Gets the SHA, subject and changed files for each commit in the given PR.
+ *
+ * Don't use GitHub API at all: the "list commits on PR" endpoint has a limit
+ * of 250 commits and doesn't return the changed files.
+ *
  * @param {{
  *  core: import('@actions/core'),
  *  pr: Awaited<ReturnType<InstanceType<import('@actions/github/lib/utils').GitHub>["rest"]["pulls"]["get"]>>["data"]
@@ -32,6 +37,8 @@ async function runGit({ args, repoPath, core, quiet }) {
  * @returns {Promise<{
  *  subject: string,
  *  sha: string,
+ *  changedPaths: string[],
+ *  changedPathSegments: Set<string>,
  * }[]>}
  */
 async function getCommitDetailsForPR({ core, pr, repoPath }) {
@@ -63,9 +70,10 @@ async function getCommitDetailsForPR({ core, pr, repoPath }) {
 
   return Promise.all(
     shas.map(async (sha) => {
+      // Subject first, then a blank line, then filenames.
       const result = (
         await runGit({
-          args: ['log', '--format=%s', '--numstat', '-1', sha],
+          args: ['log', '--format=%s', '--name-only', '-1', sha],
           repoPath,
           core,
           quiet: true,
@@ -74,9 +82,17 @@ async function getCommitDetailsForPR({ core, pr, repoPath }) {
 
       const subject = result[0]
 
+      const changedPaths = result.slice(2, -1)
+
+      const changedPathSegments = new Set(
+        changedPaths.flatMap((path) => path.split('/')),
+      )
+
       return {
         sha,
         subject,
+        changedPaths,
+        changedPathSegments,
       }
     }),
   )

ci/github-script/lint-commits.js

@@ -46,17 +46,60 @@ async function checkCommitMessages({ github, context, core, repoPath }) {
     return
   }
 
-  const commits = await getCommitDetailsForPR({
-    core,
-    pr,
-    repoPath,
-  })
+  const commits = await getCommitDetailsForPR({ core, pr, repoPath })
 
   const failures = new Set()
 
+  const conventionalCommitTypes = [
+    'build',
+    'chore',
+    'ci',
+    'doc',
+    'docs',
+    'feat',
+    'feature',
+    'fix',
+    'perf',
+    'refactor',
+    'style',
+    'test',
+  ]
+
+  /**
+   * @param {string[]} types e.g. ["fix", "feat"]
+   * @param {string?} sha commit hash
+   */
+  function makeConventionalCommitRegex(types, sha = null) {
+    core.info(
+      `${
+        sha
+          ? `Conventional commit types for ${sha?.slice(0, 16)}`
+          : 'Default conventional commit types'
+      }: ${JSON.stringify(types)}`,
+    )
+
+    return new RegExp(`^(${types.join('|')})!?(\\(.*\\))?!?:`)
+  }
+
+  // Optimize for the common case that we don't have path segments with the
+  // same name as a conventional commit type.
+  const fullConventionalCommitRegex = makeConventionalCommitRegex(
+    conventionalCommitTypes,
+  )
+
   for (const commit of commits) {
     const logMsgStart = `Commit ${commit.sha}'s message's subject ("${commit.subject}")`
 
+    // If we have a commit `perf: ...`, and we touch a file containing the path
+    // segment "perf", we don't want to flag this.
+    const filteredTypes = conventionalCommitTypes.filter(
+      (type) => !commit.changedPathSegments.has(type),
+    )
+    const conventionalCommitRegex =
+      filteredTypes.length === conventionalCommitTypes.length
+        ? fullConventionalCommitRegex
+        : makeConventionalCommitRegex(filteredTypes, commit.sha)
+
     if (!commit.subject.includes(': ')) {
       core.error(
         `${logMsgStart} was detected as not meeting our guidelines because ` +
@@ -84,6 +127,16 @@ async function checkCommitMessages({ github, context, core, repoPath }) {
       failures.add(commit.sha)
     }
 
+    if (conventionalCommitRegex.test(commit.subject)) {
+      core.error(
+        `${logMsgStart} was detected as not meeting our guidelines because ` +
+          'it seems to use conventional commit (conventionalcommits.org) ' +
+          'formatting. Nixpkgs has its own, different, commit message ' +
+          'formatting standards.',
+      )
+      failures.add(commit.sha)
+    }
+
     if (!failures.has(commit.sha)) {
       core.info(`${logMsgStart} passed our automated checks!`)
     }