# review_bot.py
import os
from github import Github
from anthropic import Anthropic
from dotenv import load_dotenv

load_dotenv()

anthropic = Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
gh = Github(os.environ["GITHUB_TOKEN"])

def get_pr_diff(repo_name: str, pr_number: int) -> dict:
    """Fetch PR diff with file context."""
    repo = gh.get_repo(repo_name)
    pr = repo.get_pull(pr_number)

    files = []
    for f in pr.get_files():
        files.append({
            "filename": f.filename,
            "status": f.status,    # added, modified, removed
            "additions": f.additions,
            "deletions": f.deletions,
            "patch": f.patch or "",  # The unified diff
        })

    return {
        "title": pr.title,
        "body": pr.body or "",
        "base_branch": pr.base.ref,
        "head_branch": pr.head.ref,
        "files": files,
    }

SYSTEM_PROMPT = """You are a senior software engineer performing a code review.
Your goal is to catch bugs, security issues, and violations of best practices.

Review guidelines:
- Flag bugs and logic errors (SEVERITY: critical)
- Flag security vulnerabilities (SEVERITY: critical)
- Flag performance issues that matter at scale (SEVERITY: warning)
- Flag code clarity issues that will cause future bugs (SEVERITY: warning)
- Skip style preferences unless they affect readability significantly
- Skip observations that are already obvious from the diff description

Output format — respond with a JSON object:
{
  "summary": "1-2 sentence overall assessment",
  "verdict": "approve" | "request_changes" | "comment",
  "issues": [
    {
      "filename": "path/to/file.py",
      "line": 42,
      "severity": "critical" | "warning" | "info",
      "category": "bug" | "security" | "performance" | "clarity",
      "comment": "Specific, actionable description of the issue",
      "suggestion": "What to do instead (optional)"
    }
  ]
}

If the code is clean, return an empty issues array and verdict "approve"."""

def build_review_prompt(pr_data: dict) -> str:
    """Build the user message with PR context."""
    files_text = ""
    for f in pr_data["files"]:
        if not f["patch"]:
            continue
        files_text += f"\n### {f['filename']} ({f['status']})\n"
        files_text += f"```diff\n{f['patch']}\n```\n"

    return f"""PR Title: {pr_data['title']}
PR Description: {pr_data['body'][:500] if pr_data['body'] else 'None'}
Base branch: {pr_data['base_branch']}

Changed files:
{files_text}

Review this pull request for bugs, security issues, and significant problems."""

import json

def review_pull_request(repo_name: str, pr_number: int) -> dict:
    """Run AI review and return structured results."""
    pr_data = get_pr_diff(repo_name, pr_number)

    # Skip trivial PRs (docs-only, small typos)
    total_changes = sum(f["additions"] + f["deletions"] for f in pr_data["files"])
    if total_changes < 5:
        return {"summary": "Trivial change, skipping AI review.", "verdict": "approve", "issues": []}

    response = anthropic.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=4096,
        system=SYSTEM_PROMPT,
        messages=[
            {"role": "user", "content": build_review_prompt(pr_data)}
        ]
    )

    # Parse structured JSON response
    text = response.content[0].text.strip()
    # Strip markdown code fences if present
    if text.startswith("```"):
 text = "\n".join(text.split("\n")[1:-1])

 try:
 return json.loads(text)
 except json.JSONDecodeError:
 # Fallback: return as plain comment if JSON parsing fails
 return {
 "summary": text,
 "verdict": "comment",
 "issues": []
 }

def post_github_review(repo_name: str, pr_number: int, review: dict) -> None:
 """Post the AI review as a GitHub PR review with inline comments."""
 repo = gh.get_repo(repo_name)
 pr = repo.get_pull(pr_number)

 # Build inline comments (requires commit SHA for position mapping)
 # Simpler approach: post review-level comment + file-level comments
 comments_text = ""
 critical_count = sum(1 for i in review["issues"] if i["severity"] == "critical")
 warning_count = sum(1 for i in review["issues"] if i["severity"] == "warning")

 if review["issues"]:
 comments_text = "\n\n### Issues Found\n\n"
 for issue in review["issues"]:
 severity_icon = "🔴" if issue["severity"] == "critical" else "🟡"
 comments_text += f"**{severity_icon} {issue['severity'].upper()} — {issue['category']}**\n"
 comments_text += f"`{issue['filename']}` line {issue['line']}\n"
 comments_text += f"{issue['comment']}\n"
 if issue.get("suggestion"):
 comments_text += f"*Suggestion: {issue['suggestion']}*\n"
 comments_text += "\n"

 body = f"""## AI Code Review

{review['summary']}

**Summary:** {critical_count} critical, {warning_count} warnings
{comments_text}---
*Reviewed by Claude claude-sonnet-4-6. This is automated analysis — use judgment.*"""

 # Map verdict to GitHub event type
 event_map = {
 "approve": "APPROVE",
 "request_changes": "REQUEST_CHANGES",
 "comment": "COMMENT"
 }

 pr.create_review(
 body=body,
 event=event_map.get(review["verdict"], "COMMENT")
 )
 print(f"Posted review: {review['verdict']} ({critical_count} critical issues)")

def get_file_content(repo_name: str, filename: str, ref: str) -> str:
 """Get full file content at a given ref."""
 repo = gh.get_repo(repo_name)
 try:
 content = repo.get_contents(filename, ref=ref)
 return content.decoded_content.decode("utf-8")
 except Exception:
 return ""

def build_enhanced_review_prompt(pr_data: dict, repo_name: str, head_sha: str) -> str:
 """Build prompt with full file context for critical files."""
 files_text = ""
 for f in pr_data["files"]:
 if not f["patch"]:
 continue
 files_text += f"\n### {f['filename']} (diff)\n```diff\n{f['patch']}\n```\n"

 # For Python/TypeScript files under 200 lines, include full content
 if f["filename"].endswith((".py", ".ts", ".js")) and f["additions"] < 200:
 full_content = get_file_content(repo_name, f["filename"], head_sha)
 if full_content and len(full_content.split("\n")) < 200:
 files_text += f"\n### {f['filename']} (full file after changes)\n"
 files_text += f"```\n{full_content}\n```\n"

 return f"""PR: {pr_data['title']}

{files_text}

Review for bugs, security vulnerabilities, and critical issues only."""

import time

MAX_DIFF_TOKENS = 50_000 # Roughly 200KB of diff text

def truncate_diff(pr_data: dict, max_chars: int = MAX_DIFF_TOKENS * 4) -> dict:
 """Truncate very large diffs to stay within token budget."""
 total_chars = sum(len(f["patch"]) for f in pr_data["files"])

 if total_chars <= max_chars:
 return pr_data

 # Keep the largest changed files, truncate the rest
 sorted_files = sorted(pr_data["files"], key=lambda f: len(f["patch"]), reverse=True)
 truncated = []
 chars_used = 0

 for f in sorted_files:
 if chars_used + len(f["patch"]) > max_chars:
 f = {**f, "patch": f["patch"][:max_chars - chars_used] + "\n... [truncated]"}
 truncated.append(f)
 chars_used += len(f["patch"])
 if chars_used >= max_chars:
 break

 return {**pr_data, "files": truncated}