Update gemini_review.py

richabanker · richabanker · commit 99ea87f5908a · 2025-02-19T11:45:37.000-08:00
diff --git a/.github/workflows/gemini-pr-review.yml b/.github/workflows/gemini-pr-review.yml
@@ -11,8 +11,8 @@ jobs:
     steps:
       - name: Checkout code
         uses: actions/checkout@v3
-        with:
-          fetch-depth: 0
+        # with:
+          # fetch-depth: 0
 
       - name: Set up Python
         uses: actions/setup-python@v4
diff --git a/hack/gemini_review.py b/hack/gemini_review.py
@@ -2,9 +2,15 @@
 import os
 from github import Github
 from google.cloud import storage
+import re
+
+# Set the maximum number of comments to post on the PR
+MAX_COMMENTS = 20
+
+total_comments_posted = 0
 
 def get_pr_latest_commit_diff_files(repo_name, pr_number, github_token):
-    """Retrieves diff information for each file in the latest commit of a PR, excluding test files."""
+    """Retrieves diff information for each file in the latest commit of a PR, excluding test files and generated files."""
     g = Github(github_token)
     repo = g.get_repo(repo_name)
     pr = repo.get_pull(pr_number)
@@ -16,7 +22,7 @@ def get_pr_latest_commit_diff_files(repo_name, pr_number, github_token):
             files = latest_commit.files
             diff_files = []
             for file in files:
-                if not file.filename.endswith("_test.go") and not file.filename.endswith("_test.py") and not "/test/" in file.filename:
+                if not file.filename.endswith("_test.go") and not file.filename.endswith("_test.py") and not "/test/" in file.filename and "_generated" not in file.filename:
                     if file.patch:
                         diff_files.append(file)
             return diff_files
@@ -41,117 +47,175 @@ def download_and_combine_guidelines(bucket_name, prefix):
 
     except Exception as e:
         print(f"Error downloading or combining guidelines: {e}")
-        return ""
 
 def download_and_combine_pr_comments(bucket_name, prefix):
     """Downloads text files from GCS using the google-cloud-storage library."""
     try:
         storage_client = storage.Client()
         bucket = storage_client.bucket(bucket_name)
         blobs = bucket.list_blobs(prefix=prefix)  # Use prefix for efficiency
-
         pr_comments_content = ""
         # TODO: Skip for now, since it is too large
         # for blob in blobs:
         #     if blob.name.endswith(".txt"):
         #         pr_comments_content += blob.download_as_text() + "\n\n"
         return pr_comments_content
-
     except Exception as e:
         print(f"Error downloading or combining PR comments: {e}")
         return ""
 
 def generate_gemini_review_with_annotations(diff_file, api_key, guidelines, pr_comments):
-    """Generates a code review with annotations, incorporating guidelines."""
+    """Generates a code review with annotations using Gemini."""
     genai.configure(api_key=api_key)
     model = genai.GenerativeModel('gemini-2.0-flash')
 
     diff = diff_file.patch
-    max_diff_length = 20000  # Adjust based on token count
+    max_diff_length = 100000
     if len(diff) > max_diff_length:
-        diff = diff[:max_diff_length]
-        diff += "\n... (truncated due to length limit) ..."
+        diff = diff[:max_diff_length] + "\n... (truncated due to length limit)..."
 
     prompt = f"""
-    The following are the API review guidelines:
+    You are an expert Kubernetes API reviewer. Follow these guidelines:
 
     {guidelines}
 
-    The following are the previous PR comments history:
+    Review the following code diff from `{diff_file.filename}`. 
+
+    Your task is to identify potential issues and suggest concrete improvements. 
+
+    Prioritize comments that highlight potential bugs, suggest improvements.
+
+    Avoid general comments that simply acknowledge correct code or good practices.
 
-    {pr_comments}
+    Provide your review comments in the following format:
 
-    Review the following code diff from file `{diff_file.filename}` and provide feedback.
-    Point out potential issues, based on the guidelines and the previous PR comments history.
-    Keep the review concise.
+    ```
+    line <line_number>: <comment>
+    line <line_number>: <comment>
+    ...and so on
+    ```
+
+* **Adhere to Conventions:**
+    * Duration fields use `fooSeconds`.
+    * Condition types are `PascalCase`.
+    * Constants are `CamelCase`.
+    * No unsigned integers.
+    * Floating-point values are avoided in `spec`.
+    * Use `int32` unless `int64` is necessary.
+    * `Reason` is a one-word, `CamelCase` category of cause.
+    * `Message` is a human-readable phrase with specifics.
+    * Label keys are lowercase with dashes.
+    * Annotations are for tooling and extensions.
+* **Compatibility:**
+    * Added fields must have non-nil default values in all API versions.
+    * New enum values must be handled safely by older clients.
+    * Validation rules on spec fields cannot be relaxed nor strengthened.
+    * Changes must be round-trippable with no loss of information.
+* **Changes:**
+    * New fields should be optional and added in a new API version if possible.
+    * Singular fields should not be made plural without careful consideration of compatibility.
+    * Avoid renaming fields within the same API version.
+    * When adding new fields or enum values, use feature gates to control enablement and ensure compatibility with older API servers.
 
     ```diff
     {diff}
     ```
     """
-    # print("total_tokens: ", model.count_tokens(prompt))
     response = model.generate_content(prompt)
-    return response.text if response.text else None
+    if response and response.text:
+        return response.text
+    else:
+        print("=== Gemini Response (Empty) ===")
+        return None
 
 def post_github_review_comments(repo_name, pr_number, diff_file, review_comment, github_token):
-    """Posts review comments to a GitHub pull request, annotating specific lines."""
+    """Posts review comments to GitHub PR, annotating specific lines."""
+    global total_comments_posted  # Declare total_comments_posted as global
     g = Github(github_token)
     repo = g.get_repo(repo_name)
     pr = repo.get_pull(pr_number)
 
     if review_comment:
         commits = list(pr.get_commits())
         if not commits:
-            print(f"WARNING: No commits found for PR {pr_number}. Posting general issue comment for {diff_file.filename}.")
+            print(f"WARNING: No commits for PR {pr_number}. Posting general comment for {diff_file.filename}.")
             pr.create_issue_comment(f"Review for {diff_file.filename}:\n{review_comment}")
             return
 
         latest_commit = commits[-1]
+        diff_lines = diff_file.patch.splitlines()
+
+        # Use regex to find line numbers and comments
+        line_comments = [(int(match.group(1)), match.group(2).strip())
+                         for match in re.finditer(r"line (\d+): (.*)", review_comment)]
+
+        for line_num, comment in line_comments:
+            if total_comments_posted >= MAX_COMMENTS:
+                print("Comment limit reached.")
+                break
+            try:
+                corrected_line_num = None
+                right_side_line = 0
+                current_line = 0
+
+                for diff_line in diff_lines:
+                    if diff_line.startswith("@@"):
+                        # Extract right-side line number from hunk info
+                        hunk_info = diff_line.split("@@")[1].strip()
+                        right_side_info = hunk_info.split("+")[1].split(" ")[0]
+                        right_side_line = int(right_side_info.split(",")[0])
+                        current_line = right_side_line - 1
+
+                    elif diff_line.startswith("+"):
+                        current_line += 1
+                        if current_line == line_num:
+                            corrected_line_num = current_line
+                            break
+
+                    elif not diff_line.startswith("-") and not diff_line.startswith("@@"): #count unchanged lines.
+                        current_line += 1
+                        if current_line == line_num:
+                            corrected_line_num = current_line
+                            break
+
+                if corrected_line_num:
+                    pr.create_review_comment(
+                        body=comment,
+                        commit=latest_commit,
+                        path=diff_file.filename,
+                        line=corrected_line_num,
+                        side="RIGHT",
+                    )
+                    total_comments_posted += 1
+                    print(f"Review comments for {diff_file.filename} posted.")
+                else:
+                    print(f"WARNING: Could not find line {line_num} in {diff_file.filename}.")
+                    print(f"Diff file: {diff_file.filename}")
+                    print(f"Gemini comment: {comment}")
+
+            except Exception as e:
+                print(f"ERROR: Failed to create comment for line {line_num} in {diff_file.filename}: {e}")
 
-        # Parse the review comment for line number annotations
-        lines_to_comment = []
-        for line in review_comment.split('\n'):
-            if "line" in line.lower() and ":" in line:
-                try:
-                    line_num = int(line.lower().split("line")[1].split(":")[0].strip())
-                    lines_to_comment.append(line_num)
-                except ValueError:
-                    continue
-
-        if lines_to_comment:
-            for line_num in lines_to_comment:
-                try:
-                    pr.create_review_comment(body=review_comment, commit=latest_commit, path=diff_file.filename, line=line_num, side="RIGHT")
-                except Exception as e:
-                    print(f"ERROR: Failed to create review comment for line {line_num} in {diff_file.filename}: {e}")
-            print(f"Review comments for {diff_file.filename} posted successfully.")
-        else:
-            pr.create_issue_comment(f"Review for {diff_file.filename}:\n{review_comment}")
-            print(f"Review for {diff_file.filename} posted as general comment since no line number was found.")
     else:
-        print(f"Gemini API returned no response for {diff_file.filename}.")
+        print(f"Gemini returned no response for {diff_file.filename}.")
 
 def main():
-    """Main function to orchestrate the Gemini PR review with annotations."""
+    """Main function to orchestrate Gemini PR review."""
     api_key = os.environ.get('GEMINI_API_KEY')
     pr_number = int(os.environ.get('PR_NUMBER'))
     repo_name = os.environ.get('GITHUB_REPOSITORY')
     github_token = os.environ.get('GITHUB_TOKEN')
 
-    # Use the GCS client library
     guidelines = download_and_combine_guidelines("hackathon-2025-sme-code-review-train", "guidelines/")
     if not guidelines:
-        print("Warning: No guidelines loaded.  Review will proceed without guidelines.")
+        print("Warning: No guidelines loaded.")
 
     diff_files = get_pr_latest_commit_diff_files(repo_name, pr_number, github_token)
-
     if diff_files is None:
-        print("Failed to retrieve PR diff files from latest commit. Exiting.")
+        print("Failed to retrieve PR diff files. Exiting.")
         return
 
     pr_comments = download_and_combine_pr_comments("hackathon-2025-sme-code-review-train", "pr_comments/")
-    if not pr_comments:
-        print("Warning: No PR comments loaded. Review will proceed without PR comments history.")
 
     for diff_file in diff_files:
         review_comment = generate_gemini_review_with_annotations(diff_file, api_key, guidelines, pr_comments)
