More robust test finder patch handling (#4667)

.github/workflows/test-finder.yml

@@ -58,6 +58,7 @@ jobs:
           5. Focus on the goose crate first, then goose-cli, then others
 
           Process:
+          0. Immediately write these requirements in your TODO list tool and periodically check against them
           1. Find a suitable untested function (use your `analyze` tool and ripgrep)
           2. Write a comprehensive unit test for it
           3. Apply your changes to the codebase
@@ -68,10 +69,8 @@ jobs:
              - 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:
+          7. If successful, write the name of the function you tested to /tmp/function_tested.txt (just the function name, e.g., "check_tool_call" or "MyStruct::my_method")
-             - If successful: write "SUCCESS" to /tmp/test_result.txt
+          8. Only create the patch file if the test actually passes
-             - 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
@@ -82,23 +81,45 @@ jobs:
 
           goose run -i /tmp/create_working_test.txt --with-builtin developer
 
-          # Check the result
+          # Debug: Check what files were created
-          if [ -f /tmp/test_result.txt ]; then
+          echo "Checking for patch file..."
-            RESULT=$(cat /tmp/test_result.txt)
+          if [ -f /tmp/test_addition.patch ]; then
-            echo "Test creation result: $RESULT"
+            echo "Patch file exists"
+            echo "Patch size: $(wc -c < /tmp/test_addition.patch) bytes"
+            echo "First few lines of patch:"
+            head -5 /tmp/test_addition.patch || true
+          else
+            echo "No patch file found at /tmp/test_addition.patch"
+          fi
 
-            if [ "$RESULT" = "SUCCESS" ] && [ -f /tmp/test_addition.patch ]; then
+          # Check for new commits that Goose might have made
-              echo "patch_created=true" >> $GITHUB_OUTPUT
+          COMMITS_AHEAD=$(git rev-list HEAD --not --remotes=origin --count 2>/dev/null || echo "0")
-              # Extract function name from patch for PR title
+          echo "Commits ahead of origin: $COMMITS_AHEAD"
-              FUNC_NAME=$(grep -E "fn test_|#\[test\]" /tmp/test_addition.patch | head -1 | sed 's/.*test_//' | sed 's/(.*//' || echo "function")
+
-              echo "function_name=${FUNC_NAME}" >> $GITHUB_OUTPUT
+          # Check if we have changes to create a PR from
+          # Either: 1) A patch file exists, OR 2) There are new commits
+          if [ -f /tmp/test_addition.patch ] && [ -s /tmp/test_addition.patch ] || [ "$COMMITS_AHEAD" -gt 0 ]; then
+            echo "Changes detected (patch file or new commits)"
+            echo "Attempting to apply patch..."
+            # Apply the patch for the PR
+            git apply /tmp/test_addition.patch 2>/dev/null || echo "Patch already applied or changes already present"
+            echo "patch_created=true" >> $GITHUB_OUTPUT
+            echo "Test creation successful - patch file created"
+
+            # Try to get the function name from Goose's output file
+            if [ -f /tmp/function_tested.txt ]; then
+              FUNC_NAME=$(cat /tmp/function_tested.txt | head -1)
             else
-              echo "patch_created=false" >> $GITHUB_OUTPUT
+              # Fallback: Extract from test name in the actual changes (staged or committed)
-              echo "Reason: $RESULT"
+              # Try staged changes first, then last commit, then patch file
+              FUNC_NAME=$(git diff --cached | grep "^+.*fn test_" | head -1 | sed 's/.*fn test_//' | sed 's/(.*//' || git diff HEAD~1 | grep "^+.*fn test_" | head -1 | sed 's/.*fn test_//' | sed 's/(.*//' || grep "fn test_" /tmp/test_addition.patch 2>/dev/null | head -1 | sed 's/.*fn test_//' | sed 's/(.*//' || echo "function")
             fi
+            # Clean up the function name (remove any trailing whitespace or special chars)
+            FUNC_NAME=$(echo "$FUNC_NAME" | tr -d '\n\r' | sed 's/[[:space:]]*$//')
+            echo "function_name=${FUNC_NAME}" >> $GITHUB_OUTPUT
           else
             echo "patch_created=false" >> $GITHUB_OUTPUT
-            echo "No result file created"
+            echo "No patch file created - either no suitable function found or test failed"
           fi
 
       - name: Extract token metrics
@@ -136,6 +157,7 @@ jobs:
           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 }}"
+          draft: true
           body: |
             ## 🤖 Automated Test Addition