File size: 17,706 Bytes
60cec59 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 | """Feature extractor for PatchJudge.
Extracts structured features from code patches using:
- Unified diff parsing
- AST analysis (Python)
- Keyword/entity extraction for issue-patch alignment
- Code quality signal detection
"""
import ast
import re
import textwrap
from collections import Counter
from typing import Optional
from patchjudge.models import PatchExample, PatchFeatures
class FeatureExtractor:
"""Extracts structured features from a patch for LLM evaluation."""
# Core/infrastructure files that are risky to modify
CORE_FILE_PATTERNS = [
r'__init__\.py$',
r'settings\.py$',
r'conf\.py$',
r'config\.py$',
r'setup\.py$',
r'setup\.cfg$',
r'manage\.py$',
r'urls\.py$',
r'wsgi\.py$',
r'asgi\.py$',
r'migrations/',
r'base\.py$',
]
# Patterns that indicate hardcoded values
HARDCODE_PATTERNS = [
r'return\s+["\'].*["\']', # return "hardcoded string"
r'return\s+\d+\b', # return 42
r'==\s*["\'][^"\']{20,}', # comparing with long hardcoded string
r'if\s+.*==\s*\d{3,}', # comparing with specific large numbers
]
# Debug statement patterns
DEBUG_PATTERNS = [
r'\bprint\s*\(',
r'\bpdb\b',
r'\bbreakpoint\s*\(',
r'\bIPython\b',
r'\bipdb\b',
r'\bconsole\.log\b',
r'\bdebugger\b',
]
# TODO/FIXME patterns
TODO_PATTERNS = [
r'\bTODO\b',
r'\bFIXME\b',
r'\bHACK\b',
r'\bXXX\b',
r'\bTEMP\b',
r'\bWORKAROUND\b',
]
def extract(self, example: PatchExample) -> PatchFeatures:
"""Extract all features from a PatchExample."""
patch = example.agent_patch
features = PatchFeatures()
# --- Diff statistics ---
added_lines, removed_lines = self._parse_diff_lines(patch)
features.num_files_changed = self._count_files_changed(patch)
features.num_lines_added = len(added_lines)
features.num_lines_removed = len(removed_lines)
features.num_hunks = self._count_hunks(patch)
# --- Code structure (AST-based for Python) ---
if self._is_python_patch(patch):
features.added_functions = self._extract_added_functions(added_lines)
features.modified_functions = self._extract_modified_functions(
patch, example.repo_context
)
features.has_error_handling = self._check_error_handling(added_lines)
features.has_edge_case_handling = self._check_edge_cases(added_lines)
features.cyclomatic_complexity_delta = self._estimate_complexity_delta(
added_lines, removed_lines
)
features.nesting_depth_max = self._estimate_max_nesting(added_lines)
# --- Issue-patch alignment ---
issue_keywords = self._extract_issue_keywords(example.problem_statement)
patch_text = '\n'.join(added_lines + removed_lines)
addressed = self._match_keywords(issue_keywords, patch_text)
features.issue_keywords_addressed = addressed
features.issue_components_mentioned = self._extract_components(
example.problem_statement
)
if issue_keywords:
features.keyword_coverage_ratio = len(addressed) / len(issue_keywords)
# --- Code quality signals ---
added_text = '\n'.join(added_lines)
features.has_todos = self._check_patterns(added_text, self.TODO_PATTERNS)
features.has_hardcoded_values = self._check_patterns(added_text, self.HARDCODE_PATTERNS)
features.has_debug_statements = self._check_patterns(added_text, self.DEBUG_PATTERNS)
features.style_violations = self._check_style_basic(added_lines)
features.follows_project_style = len(features.style_violations) == 0
# --- Risk signals ---
changed_files = self._get_changed_files(patch)
features.modifies_core_files = self._check_core_files(changed_files)
features.change_scope = self._assess_scope(
features.num_files_changed,
features.num_lines_added + features.num_lines_removed
)
new_imports = self._extract_new_imports(added_lines)
features.has_imports_added = len(new_imports) > 0
features.new_imports = new_imports
features.touches_tests = any(
'test' in f.lower() for f in changed_files
)
return features
# =========================================================================
# Diff parsing
# =========================================================================
def _parse_diff_lines(self, diff: str) -> tuple[list[str], list[str]]:
"""Parse unified diff into added and removed lines (without +/- prefix)."""
added = []
removed = []
for line in diff.split('\n'):
if line.startswith('+') and not line.startswith('+++'):
added.append(line[1:])
elif line.startswith('-') and not line.startswith('---'):
removed.append(line[1:])
return added, removed
def _count_files_changed(self, diff: str) -> int:
"""Count number of files changed in the diff."""
return len(set(
m.group(1)
for m in re.finditer(r'^diff --git a/.+ b/(.+)$', diff, re.MULTILINE)
))
def _count_hunks(self, diff: str) -> int:
"""Count number of hunks (@@ markers) in the diff."""
return len(re.findall(r'^@@\s', diff, re.MULTILINE))
def _get_changed_files(self, diff: str) -> list[str]:
"""Get list of changed file paths."""
return list(set(
m.group(1)
for m in re.finditer(r'^diff --git a/.+ b/(.+)$', diff, re.MULTILINE)
))
def _is_python_patch(self, diff: str) -> bool:
"""Check if the patch modifies Python files."""
files = self._get_changed_files(diff)
return any(f.endswith('.py') for f in files)
# =========================================================================
# Code structure analysis
# =========================================================================
def _extract_added_functions(self, added_lines: list[str]) -> list[str]:
"""Find function/method definitions in added lines."""
functions = []
for line in added_lines:
match = re.match(r'\s*def\s+(\w+)\s*\(', line)
if match:
functions.append(match.group(1))
# Also check for class definitions
match = re.match(r'\s*class\s+(\w+)', line)
if match:
functions.append(f"class:{match.group(1)}")
return functions
def _extract_modified_functions(
self, diff: str, repo_context: dict
) -> list[str]:
"""Find functions that were modified (existed before, changed now)."""
modified = []
# Parse hunk headers to find function context
for match in re.finditer(
r'^@@\s+.*\s+@@\s*(.*)$', diff, re.MULTILINE
):
context = match.group(1).strip()
# Hunk headers often contain the function name
func_match = re.match(r'def\s+(\w+)', context)
if func_match:
modified.append(func_match.group(1))
class_match = re.match(r'class\s+(\w+)', context)
if class_match:
modified.append(f"class:{class_match.group(1)}")
return list(set(modified))
def _check_error_handling(self, added_lines: list[str]) -> bool:
"""Check if added code includes error/exception handling."""
text = '\n'.join(added_lines)
patterns = [
r'\btry\s*:',
r'\bexcept\b',
r'\braise\b',
r'\bValueError\b',
r'\bTypeError\b',
r'\bKeyError\b',
r'\bAssertionError\b',
r'\bRuntimeError\b',
r'\bif\s+.*\bis\s+None\b',
r'\bif\s+not\b',
]
return any(re.search(p, text) for p in patterns)
def _check_edge_cases(self, added_lines: list[str]) -> bool:
"""Check if the patch handles edge cases."""
text = '\n'.join(added_lines)
patterns = [
r'\bif\s+len\(', # Length checks
r'\bif\s+not\s+\w+\s*:', # Empty checks
r'\bif\s+\w+\s+is\s+None', # None checks
r'\bif\s+.*<=?\s*0', # Zero/negative checks
r'\bif\s+isinstance\(', # Type checks
r'\bif\s+hasattr\(', # Attribute checks
r'\bor\s+\[\]', # Default empty list
r'\bor\s+\{\}', # Default empty dict
r'\bor\s+""', # Default empty string
r'\.get\(', # Dict .get() with default
]
return sum(1 for p in patterns if re.search(p, text)) >= 2
def _estimate_complexity_delta(
self, added_lines: list[str], removed_lines: list[str]
) -> int:
"""Estimate change in cyclomatic complexity."""
complexity_keywords = [
'if', 'elif', 'else', 'for', 'while', 'try', 'except',
'and', 'or', 'with', 'assert'
]
def count_complexity(lines):
count = 0
for line in lines:
stripped = line.strip()
for kw in complexity_keywords:
if re.search(rf'\b{kw}\b', stripped):
count += 1
break # Count each line only once
return count
return count_complexity(added_lines) - count_complexity(removed_lines)
def _estimate_max_nesting(self, added_lines: list[str]) -> int:
"""Estimate maximum nesting depth in added code."""
max_depth = 0
for line in added_lines:
if line.strip():
# Count leading spaces (assume 4-space indent)
stripped = line.lstrip()
indent = len(line) - len(stripped)
depth = indent // 4
max_depth = max(max_depth, depth)
return max_depth
# =========================================================================
# Issue-patch alignment
# =========================================================================
def _extract_issue_keywords(self, problem_statement: str) -> list[str]:
"""Extract meaningful keywords from the issue description."""
# Remove code blocks
text = re.sub(r'```[\s\S]*?```', '', problem_statement)
text = re.sub(r'`[^`]+`', '', text)
# Remove URLs
text = re.sub(r'https?://\S+', '', text)
# Extract potential identifiers (CamelCase, snake_case, etc.)
identifiers = re.findall(r'\b[A-Z][a-z]+(?:[A-Z][a-z]+)+\b', text) # CamelCase
identifiers += re.findall(r'\b\w+_\w+\b', text) # snake_case
identifiers += re.findall(r'\b[A-Z]{2,}\b', text) # CONSTANTS
# Extract error message keywords
errors = re.findall(r'\b\w*(?:Error|Exception|Warning|Failure)\b', text)
# Extract method/function names mentioned
methods = re.findall(r'\.(\w+)\(', text)
methods += re.findall(r'def\s+(\w+)', text)
methods += re.findall(r'class\s+(\w+)', text)
# Combine and deduplicate
keywords = list(set(identifiers + errors + methods))
# Filter out common words
stopwords = {
'the', 'is', 'in', 'it', 'to', 'and', 'or', 'not', 'this',
'that', 'with', 'for', 'are', 'was', 'has', 'have', 'had',
'when', 'would', 'should', 'could', 'will', 'can', 'may',
'one', 'two', 'use', 'used', 'using', 'see', 'also',
}
keywords = [k for k in keywords if k.lower() not in stopwords and len(k) > 2]
return keywords[:30] # Cap at 30 keywords
def _match_keywords(self, keywords: list[str], patch_text: str) -> list[str]:
"""Check which issue keywords appear in the patch."""
patch_lower = patch_text.lower()
return [k for k in keywords if k.lower() in patch_lower]
def _extract_components(self, problem_statement: str) -> list[str]:
"""Extract software component names from the issue."""
# Look for file paths
files = re.findall(r'[\w/]+\.py\b', problem_statement)
# Look for module/class references
modules = re.findall(r'(?:from|import)\s+([\w.]+)', problem_statement)
# Look for class.method patterns
class_methods = re.findall(r'(\w+\.\w+)\(', problem_statement)
return list(set(files + modules + class_methods))[:20]
# =========================================================================
# Code quality signals
# =========================================================================
def _check_patterns(self, text: str, patterns: list[str]) -> bool:
"""Check if any of the patterns match in the text."""
return any(re.search(p, text, re.IGNORECASE) for p in patterns)
def _check_style_basic(self, added_lines: list[str]) -> list[str]:
"""Basic style checks without external tools."""
violations = []
for i, line in enumerate(added_lines):
# Line too long (PEP 8: 79 chars, flexible to 100)
if len(line) > 120:
violations.append(f"line_too_long:{i}")
# Trailing whitespace
if line != line.rstrip():
violations.append(f"trailing_whitespace:{i}")
# Mixed tabs and spaces
if '\t' in line and ' ' in line:
violations.append(f"mixed_indentation:{i}")
# Multiple statements on one line (except for comprehensions)
if ';' in line and 'for' not in line and 'import' not in line:
violations.append(f"multiple_statements:{i}")
# Deduplicate by type
types_seen = set()
unique = []
for v in violations:
vtype = v.split(':')[0]
if vtype not in types_seen:
types_seen.add(vtype)
unique.append(v)
return unique
def _check_core_files(self, changed_files: list[str]) -> bool:
"""Check if any changed files match core/infrastructure patterns."""
for f in changed_files:
for pattern in self.CORE_FILE_PATTERNS:
if re.search(pattern, f):
return True
return False
def _assess_scope(self, num_files: int, total_lines: int) -> str:
"""Assess the scope of changes."""
if num_files <= 1 and total_lines <= 20:
return "minimal"
elif num_files <= 3 and total_lines <= 100:
return "moderate"
else:
return "extensive"
def _extract_new_imports(self, added_lines: list[str]) -> list[str]:
"""Extract newly added import statements."""
imports = []
for line in added_lines:
stripped = line.strip()
if stripped.startswith('import ') or stripped.startswith('from '):
imports.append(stripped)
return imports
def extract_features_batch(
examples: list[PatchExample],
show_progress: bool = True,
) -> list[tuple[PatchExample, PatchFeatures]]:
"""Extract features for a batch of examples."""
extractor = FeatureExtractor()
results = []
for i, ex in enumerate(examples):
if show_progress and (i + 1) % 50 == 0:
print(f" Extracted features for {i+1}/{len(examples)} examples")
features = extractor.extract(ex)
results.append((ex, features))
if show_progress:
print(f" Done: {len(results)} examples processed")
return results
if __name__ == "__main__":
import json
from patchjudge.data_loader import SWEBenchLoader
loader = SWEBenchLoader()
examples = loader.build_dataset(sources=["coderforge"])
extractor = FeatureExtractor()
# Extract features for first 5 examples
for ex in examples[:5]:
features = extractor.extract(ex)
print(f"\n{'='*60}")
print(f"Instance: {ex.instance_id}")
print(f"Agent: {ex.agent_name}")
print(f"Test passed: {ex.test_passed}")
print(f"Files changed: {features.num_files_changed}")
print(f"Lines +{features.num_lines_added}/-{features.num_lines_removed}")
print(f"Hunks: {features.num_hunks}")
print(f"Scope: {features.change_scope}")
print(f"Error handling: {features.has_error_handling}")
print(f"Edge cases: {features.has_edge_case_handling}")
print(f"TODOs: {features.has_todos}")
print(f"Debug stmts: {features.has_debug_statements}")
print(f"Hardcoded: {features.has_hardcoded_values}")
print(f"Core files: {features.modifies_core_files}")
print(f"New imports: {features.new_imports}")
print(f"Issue keywords addressed: {features.issue_keywords_addressed[:5]}")
print(f"Keyword coverage: {features.keyword_coverage_ratio:.2f}")
print(f"Style violations: {features.style_violations}")
print(f"Added functions: {features.added_functions}")
print(f"Modified functions: {features.modified_functions}")
|