Merge pull request qodo-ai#78 from Codium-ai/tr/agent_logic

Enhancement of PR Agent with User Interaction

README.md

@@ -78,6 +78,7 @@ To set up your own PR-Agent, see the [Quickstart](#Quickstart) section
 |       | Ask                                         |   :white_check_mark:    |   :white_check_mark:    |           |
 |       | Auto-Description                            |   :white_check_mark:    |        |           |
 |       | Improve Code                                |   :white_check_mark:    |   :white_check_mark:    |           |
+|       | Reflect and Review                          |   :white_check_mark:    |                         |           |
 |       |                                             |        |        |           |
 | USAGE | CLI                                         |   :white_check_mark:    |   :white_check_mark:    |   :white_check_mark:       |
 |       | Tagging bot                                 |   :white_check_mark:    |   :white_check_mark:    |           |
@@ -92,6 +93,7 @@ Examples for invoking the different tools via the [CLI](#quickstart):
 - **Describe**:     python cli.py --pr-url=<pr_url>  describe
 - **Improve**:      python cli.py --pr-url=<pr_url>  improve
 - **Ask**:          python cli.py --pr-url=<pr_url>  ask "Write me a poem about this PR"
+- **Reflect**:      python cli.py --pr-url=<pr_url>  reflect
 
 "<pr_url>" is the url of the relevant PR (for example: https://github.com/Codium-ai/pr-agent/pull/50).
 
@@ -120,12 +122,13 @@ Here are several ways to install and run PR-Agent:
 
 ## Usage and Tools
 
-**PR-Agent** provides four types of interactions ("tools"): `"PR Reviewer"`, `"PR Q&A"`, `"PR Description"` and `"PR Code Sueggestions"`.
+**PR-Agent** provides five types of interactions ("tools"): `"PR Reviewer"`, `"PR Q&A"`, `"PR Description"`, `"PR Code Sueggestions"` and `"PR Reflect and Review"`.
 
 - The "PR Reviewer" tool automatically analyzes PRs, and provides various types of feedback.
 - The "PR Q&A" tool answers free-text questions about the PR.
 - The "PR Description" tool automatically sets the PR Title and body.
 - The "PR Code Suggestion" tool provide inline code suggestions for the PR that can be applied and committed.
+- The "PR Reflect and Review" tool first initiates a dialog with the user and asks them to reflect on the PR, and then provides a review.
 
 ## How it works
 
@@ -138,11 +141,11 @@ Check out the [PR Compression strategy](./PR_COMPRESSION.md) page for more detai
 - [ ] Support open-source models, as a replacement for openai models. (Note - a minimal requirement for each open-source model is to have 8k+ context, and good support for generating json as an output)
 - [x] Support other Git providers, such as Gitlab and Bitbucket.
 - [ ] Develop additional logics for handling large PRs, and compressing git patches
-- [ ] Dedicated tools and sub-tools for specific programming languages (Python, Javascript, Java, C++, etc)
 - [ ] Add additional context to the prompt. For example, repo (or relevant files) summarization, with tools such a [ctags](https://github.com/universal-ctags/ctags)
 - [ ] Adding more tools. Possible directions:
   - [x] PR description
   - [x] Inline code suggestions
+  - [x] Reflect and review
   - [ ] Enforcing CONTRIBUTING.md guidelines
   - [ ] Performance (are there any performance issues)
   - [ ] Documentation (is the PR properly documented)

pr_agent/agent/pr_agent.py

@@ -2,17 +2,24 @@
 
 from pr_agent.tools.pr_code_suggestions import PRCodeSuggestions
 from pr_agent.tools.pr_description import PRDescription
+from pr_agent.tools.pr_information_from_user import PRInformationFromUser
 from pr_agent.tools.pr_questions import PRQuestions
 from pr_agent.tools.pr_reviewer import PRReviewer
+from pr_agent.config_loader import settings
 
 
 class PRAgent:
     def __init__(self):
         pass
 
     async def handle_request(self, pr_url, request) -> bool:
-        if any(cmd in request for cmd in ["/review", "/review_pr"]):
-            await PRReviewer(pr_url).review()
+        if any(cmd in request for cmd in ["/answer"]):
+            await PRReviewer(pr_url, is_answer=True).review()
+        elif any(cmd in request for cmd in ["/review", "/review_pr", "/reflect_and_review"]):
+            if settings.pr_reviewer.ask_and_reflect or "/reflect_and_review" in request:
+                await PRInformationFromUser(pr_url).generate_questions()
+            else:
+                await PRReviewer(pr_url).review()
         elif any(cmd in request for cmd in ["/describe", "/describe_pr"]):
             await PRDescription(pr_url).describe()
         elif any(cmd in request for cmd in ["/improve", "/improve_code"]):

pr_agent/cli.py

@@ -18,19 +18,22 @@ def run():
 - cli.py --pr-url=... describe
 - cli.py --pr-url=... improve
 - cli.py --pr-url=... ask "write me a poem about this PR"
+- cli.py --pr-url=... reflect
 
 Supported commands:
 review / review_pr - Add a review that includes a summary of the PR and specific suggestions for improvement.
 ask / ask_question [question] - Ask a question about the PR.
 describe / describe_pr - Modify the PR title and description based on the PR's contents.
 improve / improve_code - Suggest improvements to the code in the PR as pull request comments ready to commit.
+reflect - Ask the PR author questions about the PR.
 """)
     parser.add_argument('--pr_url', type=str, help='The URL of the PR to review', required=True)
     parser.add_argument('command', type=str, help='The', choices=['review', 'review_pr',
                                                                   'ask', 'ask_question',
                                                                   'describe', 'describe_pr',
                                                                   'improve', 'improve_code',
-                                                                  'user_questions'], default='review')
+                                                                  'reflect', 'review_after_reflect'],
+                        default='review')
     parser.add_argument('rest', nargs=argparse.REMAINDER, default=[])
     args = parser.parse_args()
     logging.basicConfig(level=os.environ.get("LOGLEVEL", "INFO"))
@@ -56,10 +59,14 @@ def run():
         print(f"Reviewing PR: {args.pr_url}")
         reviewer = PRReviewer(args.pr_url, cli_mode=True)
         asyncio.run(reviewer.review())
-    elif command in ['user_questions']:
+    elif command in ['reflect']:
         print(f"Asking the PR author questions: {args.pr_url}")
         reviewer = PRInformationFromUser(args.pr_url)
         asyncio.run(reviewer.generate_questions())
+    elif command in ['review_after_reflect']:
+        print(f"Processing author's answers and sending review: {args.pr_url}")
+        reviewer = PRReviewer(args.pr_url, cli_mode=True, is_answer=True)
+        asyncio.run(reviewer.review())
     else:
         print(f"Unknown command: {command}")
         parser.print_help()

pr_agent/git_providers/bitbucket_provider.py

@@ -25,6 +25,11 @@ def __init__(self, pr_url: Optional[str] = None):
         if pr_url:
             self.set_pr(pr_url)
 
+    def is_supported(self, capability: str) -> bool:
+        if capability == 'get_issue_comments':
+            return False
+        return True
+
     def set_pr(self, pr_url: str):
         self.workspace_slug, self.repo_slug, self.pr_num = self._parse_pr_url(pr_url)
         self.pr = self._get_pr()
@@ -74,6 +79,9 @@ def get_pr_description(self):
     def get_user_id(self):
         return 0
 
+    def get_issue_comments(self):
+        raise NotImplementedError("Bitbucket provider does not support issue comments yet")
+
     @staticmethod
     def _parse_pr_url(pr_url: str) -> Tuple[str, int]:
         parsed_url = urlparse(pr_url)

pr_agent/git_providers/git_provider.py

@@ -21,6 +21,10 @@ class FilePatchInfo:
 
 
 class GitProvider(ABC):
+    @abstractmethod
+    def is_supported(self, capability: str) -> bool:
+        pass
+
     @abstractmethod
     def get_diff_files(self) -> list[FilePatchInfo]:
         pass
@@ -62,6 +66,10 @@ def get_user_id(self):
     def get_pr_description(self):
         pass
 
+    @abstractmethod
+    def get_issue_comments(self):
+        pass
+
 
 def get_main_pr_language(languages, files) -> str:
     """

pr_agent/git_providers/github_provider.py

@@ -23,6 +23,9 @@ def __init__(self, pr_url: Optional[str] = None):
             self.set_pr(pr_url)
             self.last_commit_id = list(self.pr.get_commits())[-1]
 
+    def is_supported(self, capability: str) -> bool:
+        return True
+
     def set_pr(self, pr_url: str):
         self.repo, self.pr_num = self._parse_pr_url(pr_url)
         self.pr = self._get_pr()
@@ -161,6 +164,9 @@ def get_notifications(self, since: datetime):
         notifications = self.github_client.get_user().get_notifications(since=since)
         return notifications
 
+    def get_issue_comments(self):
+        return self.pr.get_issue_comments()
+
     @staticmethod
     def _parse_pr_url(pr_url: str) -> Tuple[str, int]:
         parsed_url = urlparse(pr_url)

pr_agent/git_providers/gitlab_provider.py

@@ -32,6 +32,11 @@ def __init__(self, merge_request_url: Optional[str] = None):
         self.RE_HUNK_HEADER = re.compile(
             r"^@@ -(\d+)(?:,(\d+))? \+(\d+)(?:,(\d+))? @@[ ]?(.*)")
 
+    def is_supported(self, capability: str) -> bool:
+        if capability == 'get_issue_comments':
+            return False
+        return True
+
     @property
     def pr(self):
         '''The GitLab terminology is merge request (MR) instead of pull request (PR)'''
@@ -210,6 +215,9 @@ def get_pr_branch(self):
     def get_pr_description(self):
         return self.mr.description
 
+    def get_issue_comments(self):
+        raise NotImplementedError("GitLab provider does not support issue comments yet")
+
     def _parse_merge_request_url(self, merge_request_url: str) -> Tuple[int, int]:
         parsed_url = urlparse(merge_request_url)
 

pr_agent/servers/github_action_runner.py

@@ -3,9 +3,11 @@
 import os
 import re
 
+from pr_agent.agent.pr_agent import PRAgent
 from pr_agent.config_loader import settings
 from pr_agent.tools.pr_code_suggestions import PRCodeSuggestions
 from pr_agent.tools.pr_description import PRDescription
+from pr_agent.tools.pr_information_from_user import PRInformationFromUser
 from pr_agent.tools.pr_questions import PRQuestions
 from pr_agent.tools.pr_reviewer import PRReviewer
 
@@ -53,20 +55,7 @@ async def run_action():
                 pr_url = event_payload.get("issue", {}).get("pull_request", {}).get("url", None)
                 if pr_url:
                     body = comment_body.strip().lower()
-                    if any(cmd in body for cmd in ["/review", "/review_pr"]):
-                        await PRReviewer(pr_url).review()
-                    elif any(cmd in body for cmd in ["/describe", "/describe_pr"]):
-                        await PRDescription(pr_url).describe()
-                    elif any(cmd in body for cmd in ["/improve", "/improve_code"]):
-                        await PRCodeSuggestions(pr_url).suggest()
-                    elif any(cmd in body for cmd in ["/ask", "/ask_question"]):
-                        pattern = r'(/ask|/ask_question)\s*(.*)'
-                        matches = re.findall(pattern, comment_body, re.IGNORECASE)
-                        if matches:
-                            question = matches[0][1]
-                            await PRQuestions(pr_url, question).answer()
-                    else:
-                        print(f"Unknown command: {body}")
+                    await PRAgent().handle_request(pr_url, body)
 
 
 if __name__ == '__main__':

pr_agent/settings/configuration.toml

@@ -10,6 +10,7 @@ require_tests_review=true
 require_security_review=true
 num_code_suggestions=3
 inline_code_comments = true
+ask_and_reflect=false
 
 [pr_description]
 publish_description_as_comment=false

pr_agent/settings/pr_information_from_user_prompts.toml

@@ -1,16 +1,17 @@
 [pr_information_from_user_prompt]
 system="""You are CodiumAI-PR-Reviewer, a language model designed to review git pull requests.
-Given the PR Info and the PR Git Diff, generate 4 questions about the PR for the PR author.
+Given the PR Info and the PR Git Diff, generate 3 short questions about the PR code for the PR author.
 The goal of the questions is to help the language model understand the PR better, so the questions should be insightful, informative, non-trivial, and relevant to the PR.
-Prefer yes\\no or multiple choice questions. If you have to ask open-ended questions, make sure they are not too difficult, and can be answered in a sentence or two.
+You should prefer asking yes\\no questions, or multiple choice questions. Also add at least one open-ended question, but make sure they are not too difficult, and can be answered in a sentence or two.
 
 
 Example output:
 '
 Questions to better understand the PR:
-1. ...
-2. ...
+1) ...
+2) ...
 ...
+'
 """
 
 user="""PR Info:

pr_agent/settings/pr_reviewer_prompts.toml

@@ -26,6 +26,12 @@ You must use the following JSON schema to format your answer:
       "description": "yes\\no question: does this PR have relevant tests ?"
     },
 {%- endif %}
+{%- if question_str %}
+    "Insights from user's answer": {
+      "type": "string",
+      "description": "shortly summarize the insights you gained from the user's answers to the questions"
+    },
+{%- endif %}
 {%- if require_focused %}
     "Focused PR": {
       "type": "string",
@@ -115,6 +121,16 @@ Description: '{{description}}'
 Main language: {{language}}
 {%- endif %}
 
+{%- if question_str %}
+######
+Here are questions to better understand the PR. Use the answers to provide better feedback.
+
+{{question_str|trim}}
+
+User answers:
+{{answer_str|trim}}
+######
+{%- endif %}
 
 The PR Git Diff:
 ```

pr_agent/tools/pr_information_from_user.py

@@ -35,7 +35,7 @@ def __init__(self, pr_url: str):
     async def generate_questions(self):
         logging.info('Generating question to the user...')
         if settings.config.publish_output:
-            self.git_provider.publish_comment("Preparing answer...", is_temporary=True)
+            self.git_provider.publish_comment("Preparing questions...", is_temporary=True)
         logging.info('Getting PR diff...')
         self.patches_diff = get_pr_diff(self.git_provider, self.token_handler)
         logging.info('Getting AI prediction...')
@@ -66,6 +66,6 @@ def _prepare_pr_answer(self) -> str:
         model_output = self.prediction.strip()
         if settings.config.verbosity_level >= 2:
             logging.info(f"answer_str:\n{model_output}")
-        answer_str = f"{model_output}\n\n Please respond to the question above in the following format:\n\n" + \
-                     f"/answer <question_id> <answer>\n\n" + f"Example:\n'\n/answer\n1. Yes, because ...\n2. No, because ...\n'"
+        answer_str = f"{model_output}\n\n Please respond to the questions above in the following format:\n\n" +\
+                     f"\n>/answer\n>1) ...\n>2) ...\n>...\n"
         return answer_str

pr_agent/tools/pr_reviewer.py

@@ -15,12 +15,16 @@
 
 
 class PRReviewer:
-    def __init__(self, pr_url: str, cli_mode=False):
+    def __init__(self, pr_url: str, cli_mode=False, is_answer: bool = False):
 
         self.git_provider = get_git_provider()(pr_url)
         self.main_language = get_main_pr_language(
             self.git_provider.get_languages(), self.git_provider.get_files()
         )
+        self.is_answer = is_answer
+        if self.is_answer and not self.git_provider.is_supported("get_issue_comments"):
+            raise Exception(f"Answer mode is not supported for {settings.config.git_provider} for now")
+        answer_str = question_str = self._get_user_answers()
         self.ai_handler = AiHandler()
         self.patches_diff = None
         self.prediction = None
@@ -35,6 +39,9 @@ def __init__(self, pr_url: str, cli_mode=False):
             "require_security": settings.pr_reviewer.require_security_review,
             "require_focused": settings.pr_reviewer.require_focused_review,
             'num_code_suggestions': settings.pr_reviewer.num_code_suggestions,
+            #
+            'question_str': question_str,
+            'answer_str': answer_str,
         }
         self.token_handler = TokenHandler(self.git_provider.pr,
                                           self.vars,
@@ -118,9 +125,26 @@ def _publish_inline_code_comments(self):
         except json.decoder.JSONDecodeError:
             data = try_fix_json(review)
 
-        for d in data['PR Feedback']['Code suggestions']:
-            relevant_file = d['relevant file'].strip()
-            relevant_line_in_file = d['relevant line in file'].strip()
-            content = d['suggestion content']
-
-            self.git_provider.publish_inline_comment(content, relevant_file, relevant_line_in_file)
+        if settings.pr_reviewer.num_code_suggestions > 0:
+            try:
+                for d in data['PR Feedback']['Code suggestions']:
+                    relevant_file = d['relevant file'].strip()
+                    relevant_line_in_file = d['relevant line in file'].strip()
+                    content = d['suggestion content']
+
+                    self.git_provider.publish_inline_comment(content, relevant_file, relevant_line_in_file)
+            except KeyError:
+                pass
+
+    def _get_user_answers(self):
+        answer_str = question_str = ""
+        if self.is_answer:
+            discussion_messages = self.git_provider.get_issue_comments()
+            for message in discussion_messages.reversed:
+                if "Questions to better understand the PR:" in message.body:
+                    question_str = message.body
+                elif '/answer' in message.body:
+                    answer_str = message.body
+                if answer_str and question_str:
+                    break
+        return question_str, answer_str