Add Docker support for Goose in CI/CD pipelines (#4434)

Adds a production-ready Docker image for Goose that enables using Goose itself in CI pipelines to improve the codebase.
2025-09-03 10:14:41 -04:00
parent a40a9aa8f8
commit 7034d1593c
5 changed files with 703 additions and 0 deletions
@@ -0,0 +1,143 @@
+name: Daily Test Coverage Finder
+
+on:
+  schedule:
+    # Run daily at 2 AM UTC
+    - cron: '0 2 * * *'
+  workflow_dispatch:
+    inputs:
+      dry_run:
+        description: 'Dry run (no PR creation)'
+        required: false
+        default: false
+        type: boolean
+
+permissions:
+  contents: write
+  pull-requests: write
+
+jobs:
+  find-untested-code:
+    runs-on: ubuntu-latest
+    container:
+      image: ghcr.io/block/goose:latest
+      options: --user root
+      env:
+        GOOSE_PROVIDER: ${{ vars.GOOSE_PROVIDER || 'openai' }}
+        GOOSE_MODEL: ${{ vars.GOOSE_MODEL || 'gpt-4o' }}
+        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # pin@v4
+        with:
+          fetch-depth: 0
+
+      - name: Install analysis tools
+        run: |
+          apt-get update
+          apt-get install -y jq ripgrep
+
+      - name: Find untested code and create working test
+        id: find_untested
+        run: |
+          # Create analysis and test creation script
+          cat << 'EOF' > /tmp/create_working_test.txt
+          Your task is to find ONE untested function in the Rust codebase and create a working test for it.
+
+          Requirements:
+          1. The function MUST be in the crates/ directory
+          2. It MUST have actual logic (not just a simple getter/setter)
+          3. It MUST not already have a test
+          4. Prefer functions with complexity but that are still testable in isolation
+          5. Focus on the goose crate first, then goose-cli, then others
+
+          Process:
+          1. Find a suitable untested function
+          2. Write a comprehensive unit test for it
+          3. Apply your changes to the codebase
+          4. Run the test with: cargo test --test <test_name>
+          5. If the test fails:
+             - Read and understand the error message
+             - Fix the test code
+             - Apply the fix and run the test again
+             - Repeat up to 3 times until the test passes
+          6. Once the test passes (or after 3 attempts), save the final changes as a git diff to /tmp/test_addition.patch
+          7. Report the outcome:
+             - If successful: write "SUCCESS" to /tmp/test_result.txt
+             - If no suitable function found: write "NO_FUNCTION_FOUND" to /tmp/test_result.txt
+             - If test couldn't be fixed: write "TEST_FAILED" to /tmp/test_result.txt
+
+          Important:
+          - Only add ONE test for ONE function
+          - The test MUST compile and pass before creating the patch
+          - Keep changes minimal and focused
+          - Include a descriptive test name that explains what is being tested
+          EOF
+
+          goose run -i /tmp/create_working_test.txt
+
+          # Check the result
+          if [ -f /tmp/test_result.txt ]; then
+            RESULT=$(cat /tmp/test_result.txt)
+            echo "Test creation result: $RESULT"
+
+            if [ "$RESULT" = "SUCCESS" ] && [ -f /tmp/test_addition.patch ]; then
+              echo "patch_created=true" >> $GITHUB_OUTPUT
+              # Extract function name from patch for PR title
+              FUNC_NAME=$(grep -E "fn test_|#\[test\]" /tmp/test_addition.patch | head -1 | sed 's/.*test_//' | sed 's/(.*//' || echo "function")
+              echo "function_name=${FUNC_NAME}" >> $GITHUB_OUTPUT
+            else
+              echo "patch_created=false" >> $GITHUB_OUTPUT
+              echo "Reason: $RESULT"
+            fi
+          else
+            echo "patch_created=false" >> $GITHUB_OUTPUT
+            echo "No result file created"
+          fi
+
+      - name: Create Pull Request
+        if: steps.find_untested.outputs.patch_created == 'true' && github.event.inputs.dry_run != 'true'
+        uses: peter-evans/create-pull-request@271a8d0340265f705b14b6d32b9829c1cb33d45e # pin@v7.0.8
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          commit-message: "test: add test for ${{ steps.find_untested.outputs.function_name }}"
+          title: "test: add test coverage for ${{ steps.find_untested.outputs.function_name }}"
+          body: |
+            ## 🤖 Automated Test Addition
+
+            This PR was automatically generated by Goose to improve test coverage.
+
+            ### What changed?
+            Added a unit test for a previously untested function.
+
+            ### Why?
+            Part of our daily automated test coverage improvement initiative. Goose analyzes the codebase to find untested but important functions and creates focused unit tests for them.
+
+            ### Review checklist
+            - [ ] Test is meaningful and actually tests the function
+            - [ ] Test name is descriptive
+            - [ ] Test passes locally
+            - [ ] No unnecessary changes included
+
+            ---
+            *Generated by the Daily Test Coverage Finder workflow*
+          branch: goose/test-coverage-${{ github.run_number }}
+          delete-branch: true
+          labels: |
+            goose-generated
+            test
+            automated
+
+      - name: Summary
+        if: always()
+        env:
+          PATCH_CREATED: ${{ steps.find_untested.outputs.patch_created }}
+          FUNCTION_NAME: ${{ steps.find_untested.outputs.function_name }}
+        run: |
+          if [ "$PATCH_CREATED" == "true" ]; then
+            echo "✅ Successfully found untested code and created a test"
+            echo "📝 Function tested: $FUNCTION_NAME"
+          else
+            echo "ℹ️ No suitable untested code found today"
+          fi