Use GHA eval to assign rebuild labels (#359704)

.github/workflows/eval.yml

@@ -1,6 +1,16 @@
 name: Eval
 
-on: pull_request_target
+on:
+  pull_request_target:
+  push:
+    # Keep this synced with ci/request-reviews/dev-branches.txt
+    branches:
+      - master
+      - staging
+      - release-*
+      - staging-*
+      - haskell-updates
+      - python-updates
 
 permissions:
   contents: read
@@ -11,6 +21,7 @@ jobs:
     runs-on: ubuntu-latest
     outputs:
       mergedSha: ${{ steps.merged.outputs.mergedSha }}
+      baseSha: ${{ steps.baseSha.outputs.baseSha }}
       systems: ${{ steps.systems.outputs.systems }}
     steps:
       # Important: Because of `pull_request_target`, this doesn't check out the PR,
@@ -24,14 +35,22 @@ jobs:
         id: merged
         env:
           GH_TOKEN: ${{ github.token }}
+          GH_EVENT: ${{ github.event_name }}
         run: |
-          if mergedSha=$(base/ci/get-merge-commit.sh ${{ github.repository }} ${{ github.event.number }}); then
+          case "$GH_EVENT" in
-            echo "Checking the merge commit $mergedSha"
+            push)
-            echo "mergedSha=$mergedSha" >> "$GITHUB_OUTPUT"
+              echo "mergedSha=${{ github.sha }}" >> "$GITHUB_OUTPUT"
-          else
+              ;;
-            # Skipping so that no notifications are sent
+            pull_request_target)
-            echo "Skipping the rest..."
+              if mergedSha=$(base/ci/get-merge-commit.sh ${{ github.repository }} ${{ github.event.number }}); then
-          fi
+                echo "Checking the merge commit $mergedSha"
+                echo "mergedSha=$mergedSha" >> "$GITHUB_OUTPUT"
+              else
+                # Skipping so that no notifications are sent
+                echo "Skipping the rest..."
+              fi
+              ;;
+          esac
           rm -rf base
       - name: Check out the PR at the test merge commit
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -39,8 +58,16 @@ jobs:
         if: steps.merged.outputs.mergedSha
         with:
           ref: ${{ steps.merged.outputs.mergedSha }}
+          fetch-depth: 2
           path: nixpkgs
 
+      - name: Determine base commit
+        if: github.event_name == 'pull_request_target' && steps.merged.outputs.mergedSha
+        id: baseSha
+        run: |
+          baseSha=$(git -C nixpkgs rev-parse HEAD^1)
+          echo "baseSha=$baseSha" >> "$GITHUB_OUTPUT"
+
       - name: Install Nix
         uses: cachix/install-nix-action@08dcb3a5e62fa31e2da3d490afc4176ef55ecd72 # v30
         if: steps.merged.outputs.mergedSha
@@ -105,6 +132,8 @@ jobs:
     name: Process
     runs-on: ubuntu-latest
     needs: [ outpaths, attrs ]
+    outputs:
+      baseRunId: ${{ steps.baseRunId.outputs.baseRunId }}
     steps:
       - name: Download output paths and eval stats for all systems
         uses: actions/download-artifact@fa0a91b85d4f404e444e00e005971372dc801d16 # v4.1.8
@@ -124,18 +153,93 @@ jobs:
       - name: Combine all output paths and eval stats
         run: |
           nix-build nixpkgs/ci -A eval.combine \
-            --arg resultsDir ./intermediate
+            --arg resultsDir ./intermediate \
+            -o prResult
 
       - name: Upload the combined results
         uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882 # v4.4.3
         with:
           name: result
-          path: result/*
+          path: prResult/*
 
+      - name: Get base run id
+        if: needs.attrs.outputs.baseSha
+        id: baseRunId
+        run: |
+          # Get the latest eval.yml workflow run for the PR's base commit
+          if ! run=$(gh api --method GET /repos/"$REPOSITORY"/actions/workflows/eval.yml/runs \
+            -f head_sha="$BASE_SHA" \
+            --jq '.workflow_runs | sort_by(.run_started_at) | .[-1]') \
+            || [[ -z "$run" ]]; then
+            echo "Could not find an eval.yml workflow run for $BASE_SHA, cannot make comparison"
+            exit 0
+          fi
+          echo "Comparing against $(jq .html_url <<< "$run")"
+          runId=$(jq .id <<< "$run")
+          conclusion=$(jq -r .conclusion <<< "$run")
 
-      # TODO: Run this workflow also on `push` (on at least the main development branches)
+          while [[ "$conclusion" == null ]]; do
-      # Then add an extra step here that waits for the base branch (not the merge base, because that could be very different)
+            echo "Workflow not done, waiting 10 seconds before checking again"
-      # to have completed the eval, then use
+            sleep 10
-      # gh api --method GET /repos/NixOS/nixpkgs/actions/workflows/eval.yml/runs -f head_sha=<BASE>
+            conclusion=$(gh api /repos/"$REPOSITORY"/actions/runs/"$runId" --jq '.conclusion')
-      # and follow it to the artifact results, where you can then download the outpaths.json from the base branch
+          done
-      # That can then be used to compare the number of changed paths, get evaluation stats and ping appropriate reviewers
+
+          if [[ "$conclusion" != "success" ]]; then
+            echo "Workflow was not successful, cannot make comparison"
+            exit 0
+          fi
+
+          echo "baseRunId=$runId" >> "$GITHUB_OUTPUT"
+        env:
+          REPOSITORY: ${{ github.repository }}
+          BASE_SHA: ${{ needs.attrs.outputs.baseSha }}
+          GH_TOKEN: ${{ github.token }}
+
+      - uses: actions/download-artifact@v4
+        if: steps.baseRunId.outputs.baseRunId
+        with:
+          name: result
+          path: baseResult
+          github-token: ${{ github.token }}
+          run-id: ${{ steps.baseRunId.outputs.baseRunId }}
+
+      - name: Compare against the base branch
+        if: steps.baseRunId.outputs.baseRunId
+        run: |
+          nix-build nixpkgs/ci -A eval.compare \
+            --arg beforeResultDir ./baseResult \
+            --arg afterResultDir ./prResult \
+            -o comparison
+
+          # TODO: Request reviews from maintainers for packages whose files are modified in the PR
+
+      - name: Upload the combined results
+        if: steps.baseRunId.outputs.baseRunId
+        uses: actions/upload-artifact@b4b15b8c7c6ac21ea08fcf65892d2ee8f75cf882 # v4.4.3
+        with:
+          name: comparison
+          path: comparison/*
+
+  # Separate job to have a very tightly scoped PR write token
+  tag:
+    name: Tag
+    runs-on: ubuntu-latest
+    needs: process
+    if: needs.process.outputs.baseRunId
+    permissions:
+      pull-requests: write
+    steps:
+      - name: Download process result
+        uses: actions/download-artifact@fa0a91b85d4f404e444e00e005971372dc801d16 # v4.1.8
+        with:
+          name: comparison
+          path: comparison
+
+      - name: Tagging pull request
+        run: |
+          gh api \
+            --method POST \
+            /repos/${{ github.repository }}/issues/${{ github.event.number }}/labels \
+            --input <(jq -c '{ labels: .labels }' comparison/changed-paths.json)
+        env:
+          GH_TOKEN: ${{ github.token }}

ci/eval/compare.jq

@@ -0,0 +1,152 @@
+# Turns
+#
+#   {
+#     "hello.aarch64-linux": "a",
+#     "hello.x86_64-linux": "b",
+#     "hello.aarch64-darwin": "c",
+#     "hello.x86_64-darwin": "d"
+#   }
+#
+# into
+#
+#   {
+#     "hello": {
+#       "linux": {
+#         "aarch64": "a",
+#         "x86_64": "b"
+#       },
+#       "darwin": {
+#         "aarch64": "c",
+#         "x86_64": "d"
+#       }
+#     }
+#   }
+#
+# while filtering out any attribute paths that don't match this pattern
+def expand_system:
+  to_entries
+  | map(
+    .key |= split(".")
+    | select(.key | length > 1)
+    | .double = (.key[-1] | split("-"))
+    | select(.double | length == 2)
+  )
+  | group_by(.key[0:-1])
+  | map(
+    {
+      key: .[0].key[0:-1] | join("."),
+      value:
+        group_by(.double[1])
+        | map(
+          {
+            key: .[0].double[1],
+            value: map(.key = .double[0]) | from_entries
+          }
+        )
+        | from_entries
+    })
+  | from_entries
+  ;
+
+# Transposes
+#
+#   {
+#     "a": [ "x", "y" ],
+#     "b": [ "x" ],
+#   }
+#
+# into
+#
+#   {
+#     "x": [ "a", "b" ],
+#     "y": [ "a" ]
+#   }
+def transpose:
+  [
+    to_entries[]
+    | {
+      key: .key,
+      value: .value[]
+    }
+  ]
+  | group_by(.value)
+  | map({
+    key: .[0].value,
+    value: map(.key)
+  })
+  | from_entries
+  ;
+
+# Computes the key difference for two objects:
+# {
+#   added: [ <keys only in the second object> ],
+#   removed: [ <keys only in the first object> ],
+#   changed: [ <keys with different values between the two objects> ],
+# }
+#
+def diff($before; $after):
+  {
+    added: $after | delpaths($before | keys | map([.])) | keys,
+    removed: $before | delpaths($after | keys | map([.])) | keys,
+    changed:
+      $before
+      | to_entries
+      | map(
+        $after."\(.key)" as $after2
+        | select(
+          # Filter out attributes that don't exist anymore
+          ($after2 != null)
+          and
+          # Filter out attributes that are the same as the new value
+          (.value != $after2)
+        )
+        | .key
+      )
+  }
+  ;
+
+($before[0] | expand_system) as $before
+| ($after[0] | expand_system) as $after
+| .attrdiff = diff($before; $after)
+| .rebuildsByKernel = (
+  .attrdiff.changed
+  | map({
+    key: .,
+    value: diff($before."\(.)"; $after."\(.)").changed
+  })
+  | from_entries
+  | transpose
+)
+| .rebuildCountByKernel = (
+  .rebuildsByKernel
+  | with_entries(.value |= length)
+  | pick(.linux, .darwin)
+  | {
+    linux: (.linux // 0),
+    darwin: (.darwin // 0),
+  }
+)
+| .labels = (
+  .rebuildCountByKernel
+  | to_entries
+  | map(
+    "10.rebuild-\(.key): " +
+      if .value == 0 then
+        "0"
+      elif .value <= 10 then
+        "1-10"
+      elif .value <= 100 then
+        "11-100"
+      elif .value <= 500 then
+        "101-500"
+      elif .value <= 1000 then
+        "501-1000"
+      elif .value <= 2500 then
+        "1001-2500"
+      elif .value <= 5000 then
+        "2501-5000"
+      else
+        "5000+"
+      end
+  )
+)

ci/eval/default.nix

@@ -246,6 +246,24 @@ let
           jq -s from_entries > $out/stats.json
       '';
 
+  compare =
+    { beforeResultDir, afterResultDir }:
+    runCommand "compare"
+      {
+        nativeBuildInputs = [
+          jq
+        ];
+      }
+      ''
+        mkdir $out
+        jq -n -f ${./compare.jq} \
+          --slurpfile before ${beforeResultDir}/outpaths.json \
+          --slurpfile after ${afterResultDir}/outpaths.json \
+          > $out/changed-paths.json
+
+        # TODO: Compare eval stats
+      '';
+
   full =
     {
       # Whether to evaluate just a single system, by default all are evaluated
@@ -276,6 +294,7 @@ in
     attrpathsSuperset
     singleSystem
     combine
+    compare
     # The above three are used by separate VMs in a GitHub workflow,
     # while the below is intended for testing on a single local machine
     full

ci/request-reviews/dev-branches.txt

@@ -1,5 +1,6 @@
 # Trusted development branches:
 # These generally require PRs to update and are built by Hydra.
+# Keep this synced with the branches in .github/workflows/eval.yml
 master
 staging
 release-*