find_line_number_of_relevant_line_in_file

find_line_number_of_relevant_line_in_file

pr_agent/algo/pr_processing.py

@@ -1,8 +1,9 @@
 from __future__ import annotations
 
+import re
+import difflib
 import logging
-from typing import Callable, Tuple
+from typing import Callable, Tuple, List, Any, Sequence
-
 from github import RateLimitExceededException
 
 from pr_agent.algo import MAX_TOKENS
@@ -10,7 +11,7 @@ from pr_agent.algo.git_patch_processing import convert_to_hunks_with_lines_numbe
 from pr_agent.algo.language_handler import sort_files_by_main_languages
 from pr_agent.algo.token_handler import TokenHandler
 from pr_agent.config_loader import get_settings
-from pr_agent.git_providers.git_provider import GitProvider
+from pr_agent.git_providers.git_provider import GitProvider, FilePatchInfo
 
 DELETED_FILES_ = "Deleted files:\n"
 
@@ -217,3 +218,53 @@ async def retry_with_fallback_models(f: Callable):
             logging.warning(f"Failed to generate prediction with {model}: {e}")
             if i == len(all_models) - 1:  # If it's the last iteration
                 raise  # Re-raise the last exception
+
+
+def find_line_number_of_relevant_line_in_file(diff_files: list[FilePatchInfo], relevant_file: str,
+                                              relevant_line_in_file: str) -> Tuple[int, int]:
+    position = -1
+    absolute_position = -1
+    RE_HUNK_HEADER = re.compile(
+        r"^@@ -(\d+)(?:,(\d+))? \+(\d+)(?:,(\d+))? @@[ ]?(.*)")
+    for file in diff_files:
+        if file.filename.strip() == relevant_file:
+            patch = file.patch
+            patch_lines = patch.splitlines()
+
+            # try to find the line in the patch using difflib, with some margin of error
+            matches_difflib: list[str | Any] = difflib.get_close_matches(relevant_line_in_file,
+                                                                         file.patch.splitlines(), n=3, cutoff=0.95)
+            if len(matches_difflib) == 1 and matches_difflib[0].startswith('+'):
+                relevant_line_in_file = matches_difflib[0]
+
+            delta = 0
+            for i, line in enumerate(patch_lines):
+
+                if line.startswith('@@'):
+                    delta = 0
+                    match = RE_HUNK_HEADER.match(line)
+                    start1, size1, start2, size2 = map(int, match.groups()[:4])
+                elif not line.startswith('-'):
+                    delta += 1
+
+                if relevant_line_in_file in line and line[0] != '-':
+                    position = i
+                    absolute_position = start2 + delta - 1
+                    break
+            if position == -1:
+                for i, line in enumerate(patch_lines):
+                    if line.startswith('@@'):
+                        delta = 0
+                        match = RE_HUNK_HEADER.match(line)
+                        start1, size1, start2, size2 = map(int, match.groups()[:4])
+                    elif not line.startswith('-'):
+                        delta += 1
+
+                    if relevant_line_in_file[0] == '+' and relevant_line_in_file[1:].lstrip() in line and line[
+                        0] != '-':
+                        # The model often adds a '+' to the beginning of the relevant_line_in_file even if originally
+                        # it's a context line
+                        position = i
+                        absolute_position = start2 + delta - 1
+                        break
+    return position, absolute_position

pr_agent/algo/utils.py

@@ -40,7 +40,7 @@ def convert_to_markdown(output_data: dict) -> str:
         "Security concerns": "🔒",
         "General PR suggestions": "💡",
         "Insights from user's answers": "📝",
-        "Code suggestions": "🤖",
+        "Code feedback": "🤖",
     }
 
     for key, value in output_data.items():
@@ -50,12 +50,12 @@ def convert_to_markdown(output_data: dict) -> str:
             markdown_text += f"## {key}\n\n"
             markdown_text += convert_to_markdown(value)
         elif isinstance(value, list):
-            if key.lower() == 'code suggestions':
+            if key.lower() == 'code feedback':
                 markdown_text += "\n"  # just looks nicer with additional line breaks
             emoji = emojis.get(key, "")
             markdown_text += f"- {emoji} **{key}:**\n\n"
             for item in value:
-                if isinstance(item, dict) and key.lower() == 'code suggestions':
+                if isinstance(item, dict) and key.lower() == 'code feedback':
                     markdown_text += parse_code_suggestion(item)
                 elif item:
                     markdown_text += f"  - {item}\n"
@@ -100,7 +100,7 @@ def try_fix_json(review, max_iter=10, code_suggestions=False):
     Args:
     - review: A string containing the JSON message to be fixed.
     - max_iter: An integer representing the maximum number of iterations to try and fix the JSON message.
-    - code_suggestions: A boolean indicating whether to try and fix JSON messages with code suggestions.
+    - code_suggestions: A boolean indicating whether to try and fix JSON messages with code feedback.
 
     Returns:
     - data: A dictionary containing the parsed JSON data.
@@ -108,7 +108,7 @@ def try_fix_json(review, max_iter=10, code_suggestions=False):
     The function attempts to fix broken or incomplete JSON messages by parsing until the last valid code suggestion.
     If the JSON message ends with a closing bracket, the function calls the fix_json_escape_char function to fix the
     message.
-    If code_suggestions is True and the JSON message contains code suggestions, the function tries to fix the JSON
+    If code_suggestions is True and the JSON message contains code feedback, the function tries to fix the JSON
     message by parsing until the last valid code suggestion.
     The function uses regular expressions to find the last occurrence of "}," with any number of whitespaces or
     newlines.
@@ -128,7 +128,8 @@ def try_fix_json(review, max_iter=10, code_suggestions=False):
     else:
         closing_bracket = "]}}"
 
-    if review.rfind("'Code suggestions': [") > 0 or review.rfind('"Code suggestions": [') > 0:
+    if (review.rfind("'Code feedback': [") > 0 or review.rfind('"Code feedback": [') > 0) or \
+            (review.rfind("'Code suggestions': [") > 0 or review.rfind('"Code suggestions": [') > 0) :
         last_code_suggestion_ind = [m.end() for m in re.finditer(r"\}\s*,", review)][-1] - 1
         valid_json = False
         iter_count = 0

pr_agent/git_providers/github_provider.py

@@ -1,4 +1,6 @@
 import logging
+import hashlib
+
 from datetime import datetime
 from typing import Optional, Tuple
 from urllib.parse import urlparse
@@ -10,6 +12,7 @@ from starlette_context import context
 from .git_provider import FilePatchInfo, GitProvider, IncrementalPR
 from ..algo.language_handler import is_valid_file
 from ..algo.utils import load_large_diff
+from ..algo.pr_processing import find_line_number_of_relevant_line_in_file
 from ..config_loader import get_settings
 from ..servers.utils import RateLimitExceeded
 
@@ -148,22 +151,9 @@ class GithubProvider(GitProvider):
     def publish_inline_comment(self, body: str, relevant_file: str, relevant_line_in_file: str):
         self.publish_inline_comments([self.create_inline_comment(body, relevant_file, relevant_line_in_file)])
 
+
     def create_inline_comment(self, body: str, relevant_file: str, relevant_line_in_file: str):
-        diff_files = self.get_diff_files()
+        position = find_line_number_of_relevant_line_in_file(self.diff_files, relevant_file.strip('`'), relevant_line_in_file)
-        position = -1
-        for file in diff_files:
-            if file.filename.strip() == relevant_file:
-                patch = file.patch
-                patch_lines = patch.splitlines()
-                for i, line in enumerate(patch_lines):
-                    if relevant_line_in_file in line:
-                        position = i
-                        break
-                    elif relevant_line_in_file[0] == '+' and relevant_line_in_file[1:].lstrip() in line:
-                        # The model often adds a '+' to the beginning of the relevant_line_in_file even if originally
-                        # it's a context line
-                        position = i
-                        break
         if position == -1:
             if get_settings().config.verbosity_level >= 2:
                 logging.info(f"Could not find position for {relevant_file} {relevant_line_in_file}")
@@ -171,8 +161,6 @@ class GithubProvider(GitProvider):
         else:
             subject_type = "LINE"
         path = relevant_file.strip()
-        # placeholder for future API support (already supported in single inline comment)
-        # return dict(body=body, path=path, position=position, subject_type=subject_type)
         return dict(body=body, path=path, position=position) if subject_type == "LINE" else {}
 
     def publish_inline_comments(self, comments: list[dict]):
@@ -384,3 +372,25 @@ class GithubProvider(GitProvider):
         except:
             commit_messages_str = ""
         return commit_messages_str
+
+    def generate_link_to_relevant_line_number(self, suggestion) -> str:
+        try:
+            relevant_file = suggestion['relevant file']
+            relevant_line_str = suggestion['relevant line']
+            position, absolute_position = find_line_number_of_relevant_line_in_file \
+                (self.diff_files, relevant_file.strip('`'), relevant_line_str)
+
+            if absolute_position != -1:
+                # # link to right file only
+                # link = f"https://github.com/{self.repo}/blob/{self.pr.head.sha}/{relevant_file}" \
+                #        + "#" + f"L{absolute_position}"
+
+                # link to diff
+                sha_file = hashlib.sha256(relevant_file.encode('utf-8')).hexdigest()
+                link = f"https://github.com/{self.repo}/pull/{self.pr_num}/files#diff-{sha_file}R{absolute_position}"
+                return link
+        except Exception as e:
+            if get_settings().config.verbosity_level >= 2:
+                logging.info(f"Failed adding line link, error: {e}")
+
+        return ""

pr_agent/git_providers/gitlab_provider.py

@@ -344,4 +344,4 @@ class GitLabProvider(GitProvider):
             commit_messages_str = "\n".join([f"{i + 1}. {message}" for i, message in enumerate(commit_messages_list)])
         except:
             commit_messages_str = ""
-        return commit_messages_str
+        return commit_messages_str

pr_agent/settings/configuration.toml

@@ -13,8 +13,8 @@ require_focused_review=true
 require_score_review=false
 require_tests_review=true
 require_security_review=true
-num_code_suggestions=0
+num_code_suggestions=3
-inline_code_comments = true
+inline_code_comments = false
 ask_and_reflect=false
 extra_instructions = ""
 

pr_agent/settings/pr_reviewer_prompts.toml

@@ -1,9 +1,9 @@
 [pr_review_prompt]
 system="""You are CodiumAI-PR-Reviewer, a language model designed to review git pull requests.
 Your task is to provide constructive and concise feedback for the PR, and also provide meaningfull code suggestions to improve the new PR code (the '+' lines).
-- Provide up to {{ num_code_suggestions }} code suggestions.
 {%- if num_code_suggestions > 0 %}
-- Try to focus on important suggestions like fixing code problems, issues and bugs. As a second priority, provide suggestions for meaningfull code improvements, like performance, vulnerability, modularity, and best practices.
+- Provide up to {{ num_code_suggestions }} code suggestions.
+- Try to focus on the most important suggestions, like fixing code problems, issues and bugs. As a second priority, provide suggestions for meaningfull code improvements, like performance, vulnerability, modularity, and best practices.
 - Suggestions should focus on improving the new added code lines.
 - Make sure not to provide suggestions repeating modifications already implemented in the new PR code (the '+' lines).
 {%- endif %}
@@ -24,7 +24,7 @@ You must use the following JSON schema to format your answer:
     },
     "Type of PR": {
       "type": "string",
-      "enum": ["Bug fix", "Tests", "Bug fix with tests", "Refactoring", "Enhancement", "Documentation", "Other"]
+      "enum": ["Bug fix", "Tests", "Refactoring", "Enhancement", "Documentation", "Other"]
     },
 {%- if require_score %}
     "Score": {
@@ -47,17 +47,17 @@ You must use the following JSON schema to format your answer:
 {%- if require_focused %}
     "Focused PR": {
       "type": "string",
-      "description": "Is this a focused PR, in the sense that it has a clear and coherent title and description, and all PR code diff changes are properly derived from the title and description? Explain your response."
+      "description": "Is this a focused PR, in the sense that all the PR code diff changes are united under a single focused theme ? If the theme is too broad, or the PR code diff changes are too scattered, then the PR is not focused. Explain your answer shortly."
     }
   },
 {%- endif %}
   "PR Feedback": {
-    "General PR suggestions": {
+    "General suggestions": {
       "type": "string",
-      "description": "General suggestions and feedback for the contributors and maintainers of this PR. May include important suggestions for the overall structure, primary purpose, best practices, critical bugs, and other aspects of the PR. Explain your suggestions."
+      "description": "General suggestions and feedback for the contributors and maintainers of this PR. May include important suggestions for the overall structure, primary purpose, best practices, critical bugs, and other aspects of the PR. Don't address PR title and description, or lack of tests. Explain your suggestions."
     },
 {%- if num_code_suggestions > 0 %}
-    "Code suggestions": {
+    "Code feedback": {
       "type": "array",
       "maxItems": {{ num_code_suggestions }},
       "uniqueItems": true,
@@ -66,13 +66,13 @@ You must use the following JSON schema to format your answer:
           "type": "string",
           "description": "the relevant file full path"
         },
-        "suggestion content": {
+        "suggestion": {
           "type": "string",
           "description": "a concrete suggestion for meaningfully improving the new PR code. Also describe how, specifically, the suggestion can be applied to new PR code. Add tags with importance measure that matches each suggestion ('important' or 'medium'). Do not make suggestions for updating or adding docstrings, renaming PR title and description, or linter like.
         },
-        "relevant line in file": {
+        "relevant line": {
           "type": "string",
-          "description": "an authentic single code line from the PR git diff section, to which the suggestion applies."
+          "description": "a single code line taken from the relevant file, to which the suggestion applies. The line should be a '+' line. Make sure to output the line exactly as it appears in the relevant file"
         }
       }
     },
@@ -80,8 +80,8 @@ You must use the following JSON schema to format your answer:
 {%- if require_security %}
     "Security concerns": {
       "type": "string",
-      "description": "yes\\no question: does this PR code introduce possible security concerns or issues, like SQL injection, XSS, CSRF, and others ? explain your answer"
+      "description": "yes\\no question: does this PR code introduce possible security concerns or issues, like SQL injection, XSS, CSRF, and others ? If answered 'yes', explain your answer shortly"
-       ? explain your answer"
+       ? explain your answer shortly"
     }
 {%- endif %}
   }
@@ -109,11 +109,11 @@ Example output:
     {
         "General PR suggestions": "..., `xxx`...",
 {%- if num_code_suggestions > 0 %}
-        "Code suggestions": [
+        "Code feedback": [
             {
                 "relevant file": "directory/xxx.py",
-                "suggestion content": "xxx [important]",
+                "suggestion": "xxx [important]",
-                "relevant line in file": "xxx",
+                "relevant line": "xxx",
             },
             ...
         ]

pr_agent/tools/pr_reviewer.py

@@ -7,7 +7,8 @@ from typing import List, Tuple
 from jinja2 import Environment, StrictUndefined
 
 from pr_agent.algo.ai_handler import AiHandler
-from pr_agent.algo.pr_processing import get_pr_diff, retry_with_fallback_models
+from pr_agent.algo.pr_processing import get_pr_diff, retry_with_fallback_models, \
+    find_line_number_of_relevant_line_in_file
 from pr_agent.algo.token_handler import TokenHandler
 from pr_agent.algo.utils import convert_to_markdown, try_fix_json
 from pr_agent.config_loader import get_settings
@@ -160,27 +161,38 @@ class PRReviewer:
         the feedback.
         """
         review = self.prediction.strip()
-    
+
         try:
             data = json.loads(review)
         except json.decoder.JSONDecodeError:
             data = try_fix_json(review)
 
         # Move 'Security concerns' key to 'PR Analysis' section for better display
-        if 'PR Feedback' in data and 'Security concerns' in data['PR Feedback']:
+        pr_feedback = data.get('PR Feedback', {})
-            val = data['PR Feedback']['Security concerns']
+        security_concerns = pr_feedback.get('Security concerns')
-            del data['PR Feedback']['Security concerns']
+        if security_concerns:
-            data['PR Analysis']['Security concerns'] = val
+            del pr_feedback['Security concerns']
+            data.setdefault('PR Analysis', {})['Security concerns'] = security_concerns
 
-        # Filter out code suggestions that can be submitted as inline comments
+        # 
-        if get_settings().config.git_provider != 'bitbucket' and get_settings().pr_reviewer.inline_code_comments \
+        if 'Code feedback' in pr_feedback:
-                and 'Code suggestions' in data['PR Feedback']:
+            code_feedback = pr_feedback['Code feedback']
-            data['PR Feedback']['Code suggestions'] = [
+
-                d for d in data['PR Feedback']['Code suggestions']
+            # Filter out code suggestions that can be submitted as inline comments
-                if any(key not in d for key in ('relevant file', 'relevant line in file', 'suggestion content'))
+            if get_settings().pr_reviewer.inline_code_comments:
-            ]
+                del pr_feedback['Code feedback']
-            if not data['PR Feedback']['Code suggestions']:
+            else:
-                del data['PR Feedback']['Code suggestions']
+                for suggestion in code_feedback:
+                    relevant_line_str = suggestion['relevant line'].split('\n')[0]
+
+                    # removing '+'
+                    suggestion['relevant line'] = relevant_line_str.lstrip('+').strip()
+
+                    # try to add line numbers link to code suggestions
+                    if hasattr(self.git_provider, 'generate_link_to_relevant_line_number'):
+                        link = self.git_provider.generate_link_to_relevant_line_number(suggestion)
+                        if link:
+                            suggestion['relevant line'] = f"[{suggestion['relevant line']}]({link})"
 
         # Add incremental review section
         if self.incremental.is_incremental:
@@ -205,7 +217,7 @@ class PRReviewer:
         # Log markdown response if verbosity level is high
         if get_settings().config.verbosity_level >= 2:
             logging.info(f"Markdown response:\n{markdown_text}")
-    
+
         return markdown_text
 
     def _publish_inline_code_comments(self) -> None:
@@ -222,10 +234,10 @@ class PRReviewer:
             data = try_fix_json(review)
 
         comments: List[str] = []
-        for suggestion in data.get('PR Feedback', {}).get('Code suggestions', []):
+        for suggestion in data.get('PR Feedback', {}).get('Code feedback', []):
             relevant_file = suggestion.get('relevant file', '').strip()
-            relevant_line_in_file = suggestion.get('relevant line in file', '').strip()
+            relevant_line_in_file = suggestion.get('relevant line', '').strip()
-            content = suggestion.get('suggestion content', '')
+            content = suggestion.get('suggestion', '')
             if not relevant_file or not relevant_line_in_file or not content:
                 logging.info("Skipping inline comment with missing file/line/content")
                 continue