mirror of
https://github.com/val1813/kwcode.git
synced 2026-09-03 06:34:30 +08:00
c32e423d04
核心思路：32B模型无法有效整合raw pytest输出，工程把反馈解析成 LLM最容易理解的形式（结构化诊断句），同时加入执行反馈内循环、 docstring注入、批次拆解等机制提升通过率。 9项改动： 1. generate_diagnosis() 结构化诊断句 2. _run_execution_feedback() 内循环 3. usage_finder.py 调用关系注入 4. _extract_docstrings() + _inject_docstrings() 5. _maybe_create_missing_module() LLM生成缺失模块 6. _run_whole_file_refactor() 批次拆解 7. _build_retry_hint() delta反馈 8. _clean_code_output() 格式清理（已覆盖） 9. _inject_skill_context() SKILL.md注入 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
93 lines
3.0 KiB
Python
93 lines
3.0 KiB
Python
"""
|
|
AST-based usage finder: deterministically locates all call sites of given functions.
|
|
No LLM involved — pure static analysis.
|
|
"""
|
|
|
|
import ast
|
|
import glob
|
|
import os
|
|
import logging
|
|
from typing import Dict, List
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def find_all_usages(project_root: str, func_names: list) -> Dict[str, List[str]]:
|
|
"""
|
|
Find all call sites of the given function names across the project.
|
|
Returns {func_name: ["relative/path.py:lineno: code_snippet"]}
|
|
|
|
Only searches .py files, skips test files and __pycache__.
|
|
"""
|
|
if not func_names:
|
|
return {}
|
|
|
|
# Normalize: extract bare function name from "Class.method" format
|
|
lookup = {}
|
|
for name in func_names:
|
|
bare = name.split(".")[-1] if "." in name else name
|
|
lookup[bare] = name
|
|
|
|
result = {name: [] for name in func_names}
|
|
|
|
py_files = glob.glob(os.path.join(project_root, "**", "*.py"), recursive=True)
|
|
|
|
for py_file in py_files:
|
|
rel = os.path.relpath(py_file, project_root).replace("\\", "/")
|
|
|
|
# Skip test files and cache
|
|
if "__pycache__" in rel:
|
|
continue
|
|
basename = os.path.basename(py_file).lower()
|
|
if "test" in basename:
|
|
continue
|
|
|
|
try:
|
|
with open(py_file, encoding="utf-8", errors="ignore") as f:
|
|
src = f.read()
|
|
tree = ast.parse(src)
|
|
lines = src.split("\n")
|
|
except Exception:
|
|
continue
|
|
|
|
for node in ast.walk(tree):
|
|
if not isinstance(node, ast.Call):
|
|
continue
|
|
|
|
called = None
|
|
if isinstance(node.func, ast.Attribute):
|
|
called = node.func.attr
|
|
elif isinstance(node.func, ast.Name):
|
|
called = node.func.id
|
|
|
|
if called and called in lookup:
|
|
original_name = lookup[called]
|
|
lineno = getattr(node, "lineno", 0)
|
|
snippet = lines[lineno - 1].strip() if 0 < lineno <= len(lines) else ""
|
|
entry = f"{rel}:{lineno}: {snippet[:80]}"
|
|
if entry not in result[original_name]:
|
|
result[original_name].append(entry)
|
|
|
|
# Log summary
|
|
total = sum(len(v) for v in result.values())
|
|
if total:
|
|
logger.debug("[usage_finder] Found %d call sites for %d functions", total, len(func_names))
|
|
|
|
return result
|
|
|
|
|
|
def format_usages_for_prompt(usages: Dict[str, List[str]], max_per_func: int = 5) -> str:
|
|
"""Format usage results into a prompt-injectable string."""
|
|
parts = []
|
|
for func_name, sites in usages.items():
|
|
if not sites:
|
|
continue
|
|
truncated = sites[:max_per_func]
|
|
lines = "\n".join(f" {s}" for s in truncated)
|
|
suffix = f"\n ... 还有{len(sites) - max_per_func}处" if len(sites) > max_per_func else ""
|
|
parts.append(f"### {func_name}() 的调用点\n{lines}{suffix}")
|
|
|
|
if not parts:
|
|
return ""
|
|
return "## 调用关系（修改函数签名时必须同步更新这些调用点）\n" + "\n\n".join(parts)
|