From 82a2207f4a5cc92d57400d3e60c5aecff7b2f173 Mon Sep 17 00:00:00 2001
From: Paul Gschwendtner <paulgschwendtner@gmail.com>
Date: Tue, 7 Jul 2020 18:29:13 +0200
Subject: [PATCH] build: filter out duplicate cherry-picked commits in
 changelog (#37956)

Often changelogs are generated from the patch branch and then
cherry-picked into the `CHANGELOG.md` file in `master` for
better access and readability. This is problematic though as
`conventional-changelog` (the tool we use for generating the
changelog), will duplicate commits when a future changelog
is generated from `master` then (i.e. for a new minor release).

This happens because conventional-changelog always generates the
changelog from the latest tag in a given branch to `HEAD`. The
tag in the patch branch does not correspond to any SHA in `master`
so the intersection of commits is not automatically omitted.

We work around this naively (until we have a better tool provided
by dev-infra), by deduping commits that are already part of the
changelog. This has proven to work as expected in the components
repo.

PR Close #37956
---
 tools/gulp-tasks/changelog.js | 59 +++++++++++++++++++++++++++++++----
 1 file changed, 53 insertions(+), 6 deletions(-)

diff --git a/tools/gulp-tasks/changelog.js b/tools/gulp-tasks/changelog.js
index a260052ea7..0caaf1c0d5 100644
--- a/tools/gulp-tasks/changelog.js
+++ b/tools/gulp-tasks/changelog.js
@@ -6,6 +6,9 @@
  * found in the LICENSE file at https://angular.io/license
  */
 
+const {readFileSync} = require('fs');
+const {bold, yellow} = require('chalk');
+
 module.exports = (gulp) => () => {
   const conventionalChangelog = require('gulp-conventional-changelog');
   const ignoredScopes = [
@@ -16,11 +19,55 @@ module.exports = (gulp) => () => {
   ];
 
   return gulp.src('CHANGELOG.md')
-      .pipe(conventionalChangelog({preset: 'angular'}, {}, {
-        // Ignore commits that start with `<type>(<scope>)` for any of the ignored scopes.
-        extendedRegexp: true,
-        grep: `^[^(]+\\((${ignoredScopes.join('|')})\\)`,
-        invertGrep: true,
-      }))
+      .pipe(conventionalChangelog(
+          /* core options */ {preset: 'angular'},
+          /* context options */ {},
+          /* raw-commit options */ {
+            // Ignore commits that start with `<type>(<scope>)` for any of the ignored scopes.
+            extendedRegexp: true,
+            grep: `^[^(]+\\((${ignoredScopes.join('|')})\\)`,
+            invertGrep: true,
+          },
+          /* commit parser options */ null,
+          /* writer options*/ createDedupeWriterOptions()))
       .pipe(gulp.dest('./'));
 };
+
+/**
+ * Creates changelog writer options which ensure that commits are not showing up multiple times.
+ * Commits can show up multiple times if a changelog has been generated on a publish branch
+ * and has been cherry-picked into "master". In that case, the changelog will already contain
+ * commits from master which might be added to the changelog again. This is because usually
+ * patch and minor releases are tagged from the publish branches and therefore
+ * conventional-changelog tries to build the changelog from last minor version to HEAD when a
+ * new minor version is being published from the "master" branch. We naively match commit
+ * headers as otherwise we would need to query Git and diff commits between a given patch branch.
+ * The commit header is reliable enough as it contains a direct reference to the source PR.
+ */
+function createDedupeWriterOptions() {
+  const existingChangelogContent = readFileSync('CHANGELOG.md', 'utf8');
+
+  return {
+    // Specify a writer option that can be used to modify the content of a new changelog section.
+    // See: conventional-changelog/tree/master/packages/conventional-changelog-writer
+    finalizeContext: (context) => {
+      context.commitGroups = context.commitGroups.filter((group) => {
+        group.commits = group.commits.filter((commit) => {
+          // NOTE: We cannot compare the SHAs because the commits will have a different SHA
+          // if they are being cherry-picked into a different branch.
+          if (existingChangelogContent.includes(commit.subject)) {
+            console.info(yellow(`  ↺   Skipping duplicate: "${bold(commit.header)}"`));
+            return false;
+          }
+          return true;
+        });
+
+        // Filter out commit groups which don't have any commits. Commit groups will become
+        // empty if we filter out all duplicated commits.
+        return group.commits.length !== 0;
+      });
+
+      return context;
+    }
+  };
+}