From 195482c8e2811a84cb003845a64992c69b64237b Mon Sep 17 00:00:00 2001
From: Francesco Novy <francesco.novy@sentry.io>
Date: Tue, 27 Aug 2024 12:37:03 +0200
Subject: [PATCH] ci: Stabilize CI dependency cache key (#13401)

Ensure it only changes if actual dependencies change.

Previously, we would invalidate the dependency cache every time a
package.json of the workspace changed in any way. This is defensive, but
it also means that we also invalidate if one of these things happen:

1. A script or similar is added/edited for workspace package
2. A release is made, bumping internal dependency versions

This change updates this to calculate the hash with a slightly more
sophisticated approach, which should hopefully ensure we only actually
bust the cache when a dependency _actually_ changes. This should lead to
the dependency cache being re-used much more, because only rarely is an
actual dependency changed.

---------

Co-authored-by: Charly Gomez <charly.gomez@sentry.io>
---
 .../actions/install-dependencies/action.yml   |  4 +-
 scripts/dependency-hash-key.js                | 73 +++++++++++++++++++
 2 files changed, 74 insertions(+), 3 deletions(-)
 create mode 100644 scripts/dependency-hash-key.js

diff --git a/.github/actions/install-dependencies/action.yml b/.github/actions/install-dependencies/action.yml
index 8cb80ac7440e..5fbf87c67d79 100644
--- a/.github/actions/install-dependencies/action.yml
+++ b/.github/actions/install-dependencies/action.yml
@@ -9,11 +9,9 @@ outputs:
 runs:
   using: "composite"
   steps:
-      # we use a hash of yarn.lock as our cache key, because if it hasn't changed, our dependencies haven't changed,
-      # so no need to reinstall them
     - name: Compute dependency cache key
       id: compute_lockfile_hash
-      run: echo "hash=dependencies-${{ hashFiles('yarn.lock', 'packages/*/package.json', 'dev-packages/*/package.json') }}" >> "$GITHUB_OUTPUT"
+      run: node ./scripts/dependency-hash-key.js >> "$GITHUB_OUTPUT"
       shell: bash
 
     - name: Check dependency cache
diff --git a/scripts/dependency-hash-key.js b/scripts/dependency-hash-key.js
new file mode 100644
index 000000000000..55e38e4a385e
--- /dev/null
+++ b/scripts/dependency-hash-key.js
@@ -0,0 +1,73 @@
+const crypto = require('crypto');
+const fs = require('fs');
+const path = require('path');
+
+/**
+ * Build a cache key for the dependencies of the monorepo.
+ * In addition to the content of the yarn.lock file, we also include
+ * dependencies of all workspace packages in the cache key.
+ * This ensures that we get a consistent cache key even if a dependency change does not affect
+ * the yarn.lock file.
+ */
+function outputDependencyCacheKey() {
+  const lockfileContent = fs.readFileSync(path.join(process.cwd(), 'yarn.lock'), 'utf8');
+
+  const hashParts = [lockfileContent];
+
+  const packageJson = require(path.join(process.cwd(), 'package.json'));
+
+  const workspacePackages = packageJson.workspaces || [];
+
+  // Get the package name (e.g. @sentry/browser) of all workspace packages
+  // we want to ignore their version numbers later
+  const workspacePackageNames = getWorkspacePackageNames(workspacePackages);
+
+  // Add the dependencies of the workspace itself
+  hashParts.push(getNormalizedDependencies(packageJson, workspacePackageNames));
+
+  // Now for each workspace package, add the dependencies
+  workspacePackages.forEach(workspace => {
+    const packageJsonPath = path.join(process.cwd(), workspace, 'package.json');
+    const packageJson = require(packageJsonPath);
+    hashParts.push(getNormalizedDependencies(packageJson, workspacePackageNames));
+  });
+
+  const hash = crypto.createHash('md5').update(hashParts.join('\n')).digest('hex');
+  // We log the output in a way that the GitHub Actions can append it to the output
+  // We prefix it with `dependencies-` so it is easier to identify in the logs
+  // eslint-disable-next-line no-console
+  console.log(`hash=dependencies-${hash}`);
+}
+
+function getNormalizedDependencies(packageJson, workspacePackageNames) {
+  const { dependencies, devDependencies } = packageJson;
+
+  const mergedDependencies = {
+    ...devDependencies,
+    ...dependencies,
+  };
+
+  const normalizedDependencies = {};
+
+  // Sort the keys to ensure a consistent order
+  Object.keys(mergedDependencies)
+    .sort()
+    .forEach(key => {
+      // If the dependency is a workspace package, ignore the version
+      // No need to invalidate a cache after every release
+      const version = workspacePackageNames.includes(key) ? '**workspace**' : mergedDependencies[key];
+      normalizedDependencies[key] = version;
+    });
+
+  return JSON.stringify(normalizedDependencies);
+}
+
+function getWorkspacePackageNames(workspacePackages) {
+  return workspacePackages.map(workspace => {
+    const packageJsonPath = path.join(process.cwd(), workspace, 'package.json');
+    const packageJson = require(packageJsonPath);
+    return packageJson.name;
+  });
+}
+
+outputDependencyCacheKey();