mirror of
https://github.com/jxxghp/MoviePilot.git
synced 2026-10-31 04:14:38 +08:00
711 lines
29 KiB
Python
711 lines
29 KiB
Python
# 版权所有 (c) 2025 Nous Research；授权条款见随附的 MIT 许可证。
|
||
"""
|
||
对模型生成的文本修改执行分级模糊查找与替换。
|
||
|
||
STRATEGIES 按从严格到宽松的顺序匹配，容忍工具参数中的空白、缩进、
|
||
转义和 Unicode 字形差异，同时将替换位置映射回原始文本。
|
||
"""
|
||
|
||
import bisect
|
||
import re
|
||
from difflib import SequenceMatcher
|
||
from typing import Callable, Optional
|
||
|
||
Span = tuple[int, int]
|
||
|
||
IDENTICAL_STRINGS_ERROR = (
|
||
"No edit was applied because old_string and new_string are identical. "
|
||
"Provide the existing text to replace in old_string and the changed "
|
||
"replacement text in new_string.")
|
||
|
||
UNICODE_MAP = {
|
||
# 排版双引号
|
||
"\u201c": '"', "\u201d": '"',
|
||
# 排版单引号
|
||
"\u2018": "'", "\u2019": "'",
|
||
# 长短破折号
|
||
"\u2014": "--", "\u2013": "-",
|
||
# 省略号与不换行空格
|
||
"\u2026": "...", "\u00a0": " ",
|
||
# 数学排版负号
|
||
"\u2212": "-",
|
||
# Unicode 空格分隔符，避免仅因字形不同而进入相似度回退
|
||
"\u2000": " ", "\u2001": " ", "\u2002": " ", "\u2003": " ",
|
||
"\u2004": " ", "\u2005": " ", "\u2006": " ", "\u2007": " ",
|
||
"\u2008": " ", "\u2009": " ", "\u200a": " ", "\u202f": " ",
|
||
"\u205f": " ", "\u3000": " "}
|
||
|
||
|
||
def _unicode_normalize(text: str) -> str:
|
||
"""
|
||
将排版用 Unicode 变体映射为对应的 ASCII 字符。
|
||
"""
|
||
for char, repl in UNICODE_MAP.items():
|
||
text = text.replace(char, repl)
|
||
return text
|
||
|
||
|
||
|
||
def _calculate_line_positions(content_lines: list[str], start_line: int,
|
||
end_line: int, content_length: int) -> Span:
|
||
"""
|
||
计算指定行切片覆盖的字符范围，结束位置不包含在范围内。
|
||
"""
|
||
start_pos = sum(len(line) + 1 for line in content_lines[:start_line])
|
||
end_pos = sum(len(line) + 1 for line in content_lines[:end_line]) - 1
|
||
return start_pos, min(content_length, end_pos)
|
||
|
||
|
||
def _window_spans(content: str, content_lines: list[str], n: int,
|
||
accept: Callable[[int], bool]) -> list[Span]:
|
||
"""
|
||
返回通过 accept 判定的各个连续 n 行窗口在原始文本中的字符范围。
|
||
"""
|
||
return [
|
||
_calculate_line_positions(content_lines, i, i + n, len(content))
|
||
for i in range(len(content_lines) - n + 1) if accept(i)]
|
||
|
||
|
||
def _match_transformed_lines(content: str, pattern: str,
|
||
transform: Callable[[list[str]], list[str]]) -> list[Span]:
|
||
"""
|
||
对每个行块应用指定变换后，比较待查片段与文件内容。
|
||
"""
|
||
content_lines = content.split('\n')
|
||
pattern_norm = transform(pattern.split('\n'))
|
||
n = len(pattern_norm)
|
||
return _window_spans(content, content_lines, n,
|
||
lambda i: transform(content_lines[i:i + n]) == pattern_norm)
|
||
|
||
|
||
def _strip_boundary(lines: list[str]) -> list[str]:
|
||
"""
|
||
仅移除首行和末行的边界空白，保留中间行。
|
||
"""
|
||
lines = list(lines)
|
||
lines[0] = lines[0].strip()
|
||
if len(lines) > 1:
|
||
lines[-1] = lines[-1].strip()
|
||
return lines
|
||
|
||
|
||
def _build_orig_to_norm_map(original: str) -> list[int]:
|
||
"""
|
||
将原始字符位置映射到 Unicode 规范化后的字符位置。
|
||
|
||
单个字符可能展开成多个字符；返回表长度为原文长度加 1，末项表示文本结束位置。
|
||
"""
|
||
result: list[int] = []
|
||
norm_pos = 0
|
||
for char in original:
|
||
result.append(norm_pos)
|
||
repl = UNICODE_MAP.get(char)
|
||
norm_pos += len(repl) if repl is not None else 1
|
||
result.append(norm_pos)
|
||
return result
|
||
|
||
|
||
def _invert_norm_map(orig_to_norm: list[int]) -> dict[int, int]:
|
||
"""
|
||
查找映射到指定规范化位置的最早原始字符位置。
|
||
"""
|
||
inverted: dict[int, int] = {}
|
||
for orig_pos, norm_pos in enumerate(orig_to_norm[:-1]):
|
||
inverted.setdefault(norm_pos, orig_pos)
|
||
return inverted
|
||
|
||
|
||
def _norm_end_to_orig(orig_to_norm: list[int], orig_start: int, norm_end: int) -> int:
|
||
"""
|
||
从原始起点向后查找，直到映射位置到达规范化结束位置。
|
||
"""
|
||
orig_len = len(orig_to_norm) - 1
|
||
orig_end = orig_start
|
||
while orig_end < orig_len and orig_to_norm[orig_end] < norm_end:
|
||
orig_end += 1
|
||
return orig_end
|
||
|
||
|
||
def _map_positions_norm_to_orig(orig_to_norm: list[int], norm_matches: list[Span]) -> list[Span]:
|
||
"""
|
||
将规范化文本中的匹配范围转换为原始文本范围。
|
||
"""
|
||
norm_to_orig_start = _invert_norm_map(orig_to_norm)
|
||
results: list[Span] = []
|
||
for norm_start, norm_end in norm_matches:
|
||
if norm_start in norm_to_orig_start:
|
||
orig_start = norm_to_orig_start[norm_start]
|
||
results.append((orig_start, _norm_end_to_orig(orig_to_norm, orig_start, norm_end)))
|
||
return results
|
||
|
||
|
||
def _map_normalized_positions(original: str, normalized: str,
|
||
normalized_matches: list[Span]) -> list[Span]:
|
||
"""
|
||
为连续空格或制表符折叠后的匹配结果尽力恢复原始字符范围。
|
||
"""
|
||
# 记录每个原始字符位置在规范化文本中的对应位置。
|
||
orig_to_norm = []
|
||
orig_idx = norm_idx = 0
|
||
while orig_idx < len(original) and norm_idx < len(normalized):
|
||
if original[orig_idx] == normalized[norm_idx]:
|
||
orig_to_norm.append(norm_idx)
|
||
orig_idx += 1
|
||
norm_idx += 1
|
||
elif original[orig_idx] in ' \t' and normalized[norm_idx] == ' ':
|
||
# 整段空白消费完毕后才推进规范化位置，保持折叠前后的映射一致。
|
||
orig_to_norm.append(norm_idx)
|
||
orig_idx += 1
|
||
if orig_idx < len(original) and original[orig_idx] not in ' \t':
|
||
norm_idx += 1
|
||
else:
|
||
# 原文多余空白或不应出现的规范化差异都映射到当前规范化位置。
|
||
orig_to_norm.append(norm_idx)
|
||
orig_idx += 1
|
||
orig_to_norm.extend([len(normalized)] * (len(original) - orig_idx))
|
||
|
||
norm_to_orig_start: dict[int, int] = {}
|
||
norm_to_orig_end = {}
|
||
for orig_pos, norm_pos in enumerate(orig_to_norm):
|
||
norm_to_orig_start.setdefault(norm_pos, orig_pos)
|
||
norm_to_orig_end[norm_pos] = orig_pos
|
||
|
||
original_matches = []
|
||
for norm_start, norm_end in normalized_matches:
|
||
if norm_start in norm_to_orig_start:
|
||
orig_start = norm_to_orig_start[norm_start]
|
||
else:
|
||
orig_start = min(i for i, n in enumerate(orig_to_norm) if n >= norm_start)
|
||
if norm_end - 1 in norm_to_orig_end:
|
||
orig_end = norm_to_orig_end[norm_end - 1] + 1
|
||
else:
|
||
orig_end = orig_start + (norm_end - norm_start)
|
||
# 仅当规范化匹配以空格结尾时吸收末尾折叠空白；否则保留匹配后作为单词边界的首个空白。
|
||
if norm_end < len(normalized) and normalized[norm_end - 1] == ' ':
|
||
while orig_end < len(original) and original[orig_end] in ' \t':
|
||
orig_end += 1
|
||
original_matches.append((orig_start, min(orig_end, len(original))))
|
||
return original_matches
|
||
|
||
|
||
# 每种策略接受原文和待查片段，返回原始文本中的起止字符范围。
|
||
|
||
def _strategy_exact(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
使用不重叠的精确匹配，保持 str.replace 的替换语义。
|
||
|
||
重叠范围会破坏 replace_all 的文本位置，因此不予返回。
|
||
"""
|
||
return [m.span() for m in re.finditer(re.escape(pattern), content)]
|
||
|
||
|
||
def _strategy_line_trimmed(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
去除每行首尾空白后比较。
|
||
"""
|
||
return _match_transformed_lines(content, pattern, lambda ls: [line.strip() for line in ls])
|
||
|
||
|
||
def _strategy_whitespace_normalized(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
把连续空格或制表符折叠为单个空格后比较。
|
||
"""
|
||
def normalize(s: str) -> str:
|
||
"""
|
||
折叠水平方向空白，同时保留行边界。
|
||
"""
|
||
return re.sub(r'[ \t]+', ' ', s)
|
||
|
||
content_normalized = normalize(content)
|
||
matches_in_normalized = _strategy_exact(content_normalized, normalize(pattern))
|
||
if not matches_in_normalized:
|
||
return []
|
||
return _map_normalized_positions(content, content_normalized, matches_in_normalized)
|
||
|
||
|
||
def _strategy_indentation_flexible(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
忽略各行开头的缩进后比较。
|
||
"""
|
||
return _match_transformed_lines(content, pattern, lambda ls: [line.lstrip() for line in ls])
|
||
|
||
|
||
def _strategy_escape_normalized(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
把待查片段中的字面换行、制表和回车转义序列解释为控制字符后比较。
|
||
"""
|
||
pattern_unescaped = pattern.replace('\\n', '\n').replace('\\t', '\t').replace('\\r', '\r')
|
||
if pattern_unescaped == pattern:
|
||
return []
|
||
return _strategy_exact(content, pattern_unescaped)
|
||
|
||
|
||
def _strategy_trimmed_boundary(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
仅去除首尾行的边界空白后比较。
|
||
"""
|
||
return _match_transformed_lines(content, pattern, _strip_boundary)
|
||
|
||
|
||
def _strategy_unicode_normalized(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
两侧字符均规范化为 ASCII 后，尝试精确匹配及逐行去空白匹配。
|
||
"""
|
||
norm_pattern = _unicode_normalize(pattern)
|
||
norm_content = _unicode_normalize(content)
|
||
if norm_content == content and norm_pattern == pattern:
|
||
return []
|
||
norm_matches = (_strategy_exact(norm_content, norm_pattern)
|
||
or _strategy_line_trimmed(norm_content, norm_pattern))
|
||
if not norm_matches:
|
||
return []
|
||
return _map_positions_norm_to_orig(_build_orig_to_norm_map(content), norm_matches)
|
||
|
||
|
||
def _strategy_block_anchor(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
以首尾行作为锚点，按中间行的相似度选择匹配块。
|
||
"""
|
||
pattern_lines = _unicode_normalize(pattern).split('\n')
|
||
if len(pattern_lines) < 2:
|
||
return []
|
||
first_line = pattern_lines[0].strip()
|
||
last_line = pattern_lines[-1].strip()
|
||
n = len(pattern_lines)
|
||
|
||
# 在规范化行上匹配，但从原始行计算偏移，避免单字符展开造成位置偏移。
|
||
norm_content_lines = _unicode_normalize(content).split('\n')
|
||
potential_matches = {
|
||
i for i in range(len(norm_content_lines) - n + 1)
|
||
if norm_content_lines[i].strip() == first_line
|
||
and norm_content_lines[i + n - 1].strip() == last_line}
|
||
# 更宽松的阈值曾匹配到无关文本块，因此保留当前相似度下限。
|
||
threshold = 0.50 if len(potential_matches) == 1 else 0.70
|
||
pattern_middle = '\n'.join(pattern_lines[1:-1])
|
||
|
||
def similar(i: int) -> bool:
|
||
"""
|
||
按锚点之间的正文相似度阈值判断匹配。
|
||
"""
|
||
if i not in potential_matches:
|
||
return False
|
||
if n <= 2:
|
||
return True
|
||
content_middle = '\n'.join(norm_content_lines[i + 1:i + n - 1])
|
||
return SequenceMatcher(None, content_middle, pattern_middle).ratio() >= threshold
|
||
|
||
return _window_spans(content, content.split('\n'), n, similar)
|
||
|
||
|
||
def _strategy_context_aware(content: str, pattern: str) -> list[Span]:
|
||
"""
|
||
最后尝试有锚点的逐行相似匹配，要求每个非空行的相似度至少为 0.80。
|
||
|
||
首尾锚点先缩小扫描范围，再逐行检查以减少偶然匹配。
|
||
"""
|
||
pattern_lines = pattern.split('\n')
|
||
content_lines = content.split('\n')
|
||
n = len(pattern_lines)
|
||
if n > len(content_lines):
|
||
return []
|
||
first_pat = pattern_lines[0].strip()
|
||
last_pat = pattern_lines[-1].strip()
|
||
|
||
def _sim(a: str, b: str) -> float:
|
||
"""
|
||
比较去除首尾空白后的行，精确相同的行保持满分。
|
||
"""
|
||
return 1.0 if a == b else SequenceMatcher(None, a, b).ratio()
|
||
|
||
def accept(i: int) -> bool:
|
||
"""
|
||
要求首尾锚点及所有非空行均通过匹配检查。
|
||
"""
|
||
block_lines = content_lines[i:i + n]
|
||
if _sim(first_pat, block_lines[0].strip()) < 0.80:
|
||
return False
|
||
if _sim(last_pat, block_lines[-1].strip()) < 0.80:
|
||
return False
|
||
return all(
|
||
not p_line.strip() or _sim(p_line.strip(), c_line.strip()) >= 0.80
|
||
for p_line, c_line in zip(pattern_lines, block_lines))
|
||
|
||
return _window_spans(content, content_lines, n, accept)
|
||
|
||
|
||
# 优先精确匹配，最后才尝试基于相似度的策略。
|
||
STRATEGIES: list[tuple[str, Callable[[str, str], list[Span]]]] = [
|
||
("exact", _strategy_exact),
|
||
("line_trimmed", _strategy_line_trimmed),
|
||
("whitespace_normalized", _strategy_whitespace_normalized),
|
||
("indentation_flexible", _strategy_indentation_flexible),
|
||
("escape_normalized", _strategy_escape_normalized),
|
||
("trimmed_boundary", _strategy_trimmed_boundary),
|
||
("unicode_normalized", _strategy_unicode_normalized),
|
||
("block_anchor", _strategy_block_anchor),
|
||
("context_aware", _strategy_context_aware)]
|
||
|
||
# 这些策略仅近似匹配旧片段，可用于唯一位置的替换，不能用于批量替换。
|
||
SIMILARITY_STRATEGIES = frozenset({"block_anchor", "context_aware"})
|
||
|
||
|
||
|
||
def is_already_applied(content: str, old_string: str, new_string: str) -> bool:
|
||
"""
|
||
保守判断修改是否已经应用，避免重复提交同一次替换。
|
||
|
||
:param content: 当前文件文本
|
||
:param old_string: 原本需要替换的旧片段
|
||
:param new_string: 期望写入的新片段
|
||
:return: 新片段至少 8 个字符且已精确存在、旧片段已不存在时为 True
|
||
"""
|
||
if not new_string or len(new_string.strip()) < 8 or new_string not in content:
|
||
return False
|
||
return old_string == new_string or old_string not in content
|
||
|
||
|
||
def _matched_regions(content: str, matches: list[Span]) -> str:
|
||
"""
|
||
仅收集当前匹配策略选中的原始文本区域。
|
||
"""
|
||
return "".join(content[start:end] for start, end in matches)
|
||
|
||
|
||
def _format_match_locations(content: str, matches: list[Span], cap: int = 5) -> str:
|
||
"""
|
||
将最多 cap 个匹配位置渲染为带行号的短片段。
|
||
"""
|
||
rows = []
|
||
for start, _end in matches[:cap]:
|
||
line_no = content.count("\n", 0, start) + 1
|
||
line_start = content.rfind("\n", 0, start) + 1
|
||
line_end = content.find("\n", line_start)
|
||
if line_end == -1:
|
||
line_end = len(content)
|
||
snippet = content[line_start:line_end].strip()
|
||
if len(snippet) > 80:
|
||
snippet = snippet[:77] + "..."
|
||
rows.append(f" L{line_no}: {snippet}")
|
||
extra = len(matches) - cap
|
||
if extra > 0:
|
||
rows.append(f" ... and {extra} more")
|
||
return "\n".join(rows)
|
||
|
||
|
||
def fuzzy_find_and_replace(content: str, old_string: str, new_string: str,
|
||
replace_all: bool = False) -> tuple[str, int, Optional[str], Optional[str]]:
|
||
"""
|
||
依次尝试匹配策略并替换选中的文本区域。
|
||
|
||
:param content: 待修改的原始文件文本
|
||
:param old_string: 用于定位的非空旧片段，不接受纯空白
|
||
:param new_string: 替换内容，空字符串表示删除
|
||
:param replace_all: 是否替换全部匹配；近似相似策略不允许批量替换
|
||
:return: 新文本、匹配数量、策略名称及错误说明；失败时保留原文并返回零匹配
|
||
"""
|
||
if not old_string:
|
||
# 给出可执行的恢复建议，避免模型在空旧片段错误后不断重复同一调用；参见 cline/cline#13970。
|
||
return content, 0, None, (
|
||
"old_string is empty — nothing to match. Set old_string to the exact "
|
||
"existing text the replacement should replace (read the file first if "
|
||
"unsure). To create a new file or fully rewrite one, use write_file "
|
||
"instead. Do not re-send this call unchanged.")
|
||
if not old_string.strip():
|
||
# 纯空白锚点容易命中大量区域或产生歧义，不能可靠定位编辑目标。
|
||
return content, 0, None, (
|
||
"old_string is only whitespace — provide non-blank text to match. Set it "
|
||
"to the exact existing text the replacement should replace (read the file "
|
||
"first if unsure). Do not re-send this call unchanged.")
|
||
if old_string == new_string:
|
||
return content, 0, None, IDENTICAL_STRINGS_ERROR
|
||
|
||
for strategy_name, strategy_fn in STRATEGIES:
|
||
matches = strategy_fn(content, old_string)
|
||
if not matches:
|
||
continue
|
||
|
||
if len(matches) > 1 and not replace_all:
|
||
locations = _format_match_locations(content, matches)
|
||
return content, 0, None, (
|
||
f"Found {len(matches)} matches for old_string. "
|
||
f"Provide more context to make it unique, or use replace_all=True. "
|
||
f"Matches:\n{locations}")
|
||
if replace_all and len(matches) > 1 and strategy_name in SIMILARITY_STRATEGIES:
|
||
return content, 0, None, (
|
||
f"Found {len(matches)} approximate matches via the "
|
||
f"'{strategy_name}' strategy; replace_all only applies to exact "
|
||
f"matches. Provide the precise text (whitespace included) so an "
|
||
f"exact/line-trimmed match can be made.")
|
||
|
||
# 非精确匹配经过规范化，新片段可能携带原文件没有的序列化转义偏差。
|
||
if strategy_name != "exact":
|
||
drift_err = _detect_escape_drift(content, matches, old_string, new_string)
|
||
if drift_err:
|
||
return content, 0, None, drift_err
|
||
|
||
effective_new = _maybe_unescape_new_string(new_string, content, matches)
|
||
if strategy_name == "unicode_normalized":
|
||
effective_new = _preserve_unicode_in_replacement(content, matches, old_string, effective_new)
|
||
new_content = _apply_replacements(
|
||
content, matches, effective_new,
|
||
old_string=old_string if strategy_name != "exact" else None)
|
||
return new_content, len(matches), strategy_name, None
|
||
|
||
return content, 0, None, "Could not find a match for old_string in the file"
|
||
|
||
|
||
|
||
def _detect_newline_literal_drift(content: str, matches: list[Span],
|
||
old_string: str, new_string: str) -> Optional[str]:
|
||
"""
|
||
检测工具参数被额外 JSON 转义，导致字面换行序列代替文件中的真实换行。
|
||
|
||
逐个匹配区域比较字面转义数量，避免 replace_all 把多个区域计数叠加。
|
||
只有旧片段比对应文件区域多出字面换行序列，且新片段也包含该序列时才报告错误。
|
||
只在新片段中出现的转义无法与有意新增字面文本区分；回车由后续解码逻辑处理。
|
||
|
||
:return: 可确认转义漂移时的错误说明，否则为 None
|
||
"""
|
||
if "\\n" not in new_string:
|
||
return None
|
||
old_literals = old_string.count("\\n")
|
||
for start, end in matches:
|
||
region = content[start:end]
|
||
if "\n" in region and old_literals > region.count("\\n"):
|
||
return (
|
||
"Escape-drift detected: old_string contains more literal "
|
||
"'\\\\n' sequences than the matched region of the file, which "
|
||
"has real line breaks there instead. This is almost always a "
|
||
"tool-call serialization artifact where a line break got escaped "
|
||
"one extra time; new_string would write it as backslash + n. "
|
||
"Re-read the file with read_file and pass old_string/new_string "
|
||
"with actual line breaks, keeping only the '\\\\n' sequences "
|
||
"that appear literally in the file.")
|
||
return None
|
||
|
||
|
||
def _detect_escape_drift(content: str, matches: list[Span],
|
||
old_string: str, new_string: str) -> Optional[str]:
|
||
"""
|
||
检查新片段是否携带额外引号转义、加倍反斜杠或代替真实换行的字面序列。
|
||
|
||
:return: 发现工具参数转义漂移时的错误说明，否则为 None
|
||
"""
|
||
has_quote_suspects = "\\'" in new_string or '\\"' in new_string
|
||
if not has_quote_suspects and "\\" not in old_string:
|
||
return None
|
||
|
||
matched_regions = _matched_regions(content, matches)
|
||
if has_quote_suspects:
|
||
for suspect in ("\\'", '\\"'):
|
||
if suspect in new_string and suspect in old_string and suspect not in matched_regions:
|
||
plain = suspect[1]
|
||
return (
|
||
f"Escape-drift detected: old_string and new_string contain "
|
||
f"the literal sequence {suspect!r} but the matched region of "
|
||
f"the file does not. This is almost always a tool-call "
|
||
f"serialization artifact where an apostrophe or quote got "
|
||
f"prefixed with a spurious backslash. Re-read the file with "
|
||
f"read_file and pass old_string/new_string without "
|
||
f"backslash-escaping {plain!r} characters.")
|
||
newline_drift = _detect_newline_literal_drift(content, matches, old_string, new_string)
|
||
if newline_drift:
|
||
return newline_drift
|
||
return _detect_backslash_doubling(matched_regions, old_string, new_string)
|
||
|
||
|
||
def _backslash_runs(s: str) -> list[int]:
|
||
"""
|
||
按出现顺序返回各段连续反斜杠的长度。
|
||
"""
|
||
return [len(run) for run in re.findall(r"\\+", s)]
|
||
|
||
|
||
def _detect_backslash_doubling(matched_regions: str, old_string: str,
|
||
new_string: str) -> Optional[str]:
|
||
"""
|
||
检测旧片段中每段反斜杠是否均为文件对应长度的两倍。
|
||
|
||
要求连续段数量一致，至少有足够明确的加倍信号，且新片段尚未符合文件中的长度。
|
||
"""
|
||
old_runs = _backslash_runs(old_string)
|
||
file_runs = _backslash_runs(matched_regions)
|
||
if (not old_runs or not file_runs or len(old_runs) != len(file_runs)
|
||
or old_runs == file_runs
|
||
or any(o != f * 2 for o, f in zip(old_runs, file_runs))
|
||
or not (any(f >= 2 for f in file_runs) or len(file_runs) >= 2)
|
||
or _backslash_runs(new_string) == file_runs):
|
||
return None
|
||
return (
|
||
"Escape-drift detected: every backslash run in old_string is exactly "
|
||
"twice as long as in the matched region of the file (e.g. the file "
|
||
"has `\\\\` where old_string has `\\\\\\\\`). The tool-call arguments "
|
||
"were JSON-escaped one extra time; applying new_string verbatim would "
|
||
"double every backslash in the file. Re-read the file with read_file "
|
||
"and resend old_string/new_string with the backslash counts exactly "
|
||
"as they appear in the file.")
|
||
|
||
|
||
def _maybe_unescape_new_string(new_string: str, content: str, matches: list[Span]) -> str:
|
||
"""
|
||
仅在匹配区域已包含真实控制字符时，将新片段中的对应制表或回车序列还原。
|
||
|
||
不处理字面换行序列，以免破坏源码中有意保留的转义文本。
|
||
"""
|
||
if "\\t" not in new_string and "\\r" not in new_string:
|
||
return new_string
|
||
matched_regions = _matched_regions(content, matches)
|
||
for literal, control in (("\\t", "\t"), ("\\r", "\r")):
|
||
if literal in new_string and control in matched_regions:
|
||
new_string = new_string.replace(literal, control)
|
||
return new_string
|
||
|
||
|
||
|
||
def _leading_whitespace(line: str) -> str:
|
||
"""
|
||
取得替换时需要保留的行首缩进。
|
||
"""
|
||
return line[:len(line) - len(line.lstrip(" \t"))]
|
||
|
||
|
||
def _first_meaningful_line(text: str) -> Optional[str]:
|
||
"""
|
||
选择第一个非空行作为缩进基准。
|
||
"""
|
||
return next((line for line in text.split("\n") if line.strip()), None)
|
||
|
||
|
||
def _reindent_replacement(file_region: str, old_string: str, new_string: str) -> str:
|
||
"""
|
||
非精确匹配后，将新片段的基础缩进对齐到文件中的实际缩进。
|
||
|
||
保留相对嵌套层级；浅于旧片段基准的行也对齐到文件基准。
|
||
"""
|
||
if not new_string:
|
||
return new_string
|
||
old_first = _first_meaningful_line(old_string)
|
||
file_first = _first_meaningful_line(file_region)
|
||
if old_first is None or file_first is None:
|
||
return new_string
|
||
old_indent = _leading_whitespace(old_first)
|
||
file_indent = _leading_whitespace(file_first)
|
||
if old_indent == file_indent:
|
||
return new_string
|
||
|
||
out_lines: list[str] = []
|
||
for line in new_string.split("\n"):
|
||
if not line.strip():
|
||
out_lines.append(line)
|
||
elif _leading_whitespace(line).startswith(old_indent):
|
||
out_lines.append(file_indent + line[len(old_indent):])
|
||
else:
|
||
out_lines.append(file_indent + line.lstrip(" \t"))
|
||
return "\n".join(out_lines)
|
||
|
||
|
||
def _preserve_unicode_in_replacement(content: str, matches: list[Span],
|
||
old_string: str, new_string: str) -> str:
|
||
"""
|
||
仅把旧片段到新片段的真实编辑应用到文件，保留原文未修改的 Unicode 排版字符。
|
||
"""
|
||
file_region = _matched_regions(content, matches)
|
||
norm_old = _unicode_normalize(old_string)
|
||
if norm_old != _unicode_normalize(file_region):
|
||
# 规范化结果未能匹配时回退，避免错误应用该策略。
|
||
return new_string
|
||
|
||
file_orig_to_norm = _build_orig_to_norm_map(file_region)
|
||
|
||
result_parts: list[str] = []
|
||
for tag, i1, i2, j1, j2 in SequenceMatcher(None, norm_old, new_string).get_opcodes():
|
||
if tag == "equal":
|
||
# 定位规范化位置对应的原始字符，包括处在单字符展开序列内部的位置。
|
||
orig_start = bisect.bisect_right(file_orig_to_norm, i1) - 1
|
||
orig_end = _norm_end_to_orig(file_orig_to_norm, orig_start, i2)
|
||
result_parts.append(file_region[orig_start:orig_end])
|
||
elif tag != "delete":
|
||
result_parts.append(new_string[j1:j2])
|
||
return "".join(result_parts)
|
||
|
||
|
||
def _apply_replacements(content: str, matches: list[Span],
|
||
new_string: str, old_string: Optional[str] = None) -> str:
|
||
"""
|
||
从后向前应用替换以保持偏移有效，非精确匹配时按每个区域重排缩进。
|
||
"""
|
||
result = content
|
||
for start, end in sorted(matches, key=lambda x: x[0], reverse=True):
|
||
adjusted = new_string
|
||
if old_string is not None:
|
||
adjusted = _reindent_replacement(content[start:end], old_string, new_string)
|
||
result = result[:start] + adjusted + result[end:]
|
||
return result
|
||
|
||
|
||
|
||
def _visualize_whitespace(line: str) -> str:
|
||
"""
|
||
可视化行首空白：箭头表示制表符，中点表示空格。
|
||
"""
|
||
stripped = line.lstrip(" \t")
|
||
prefix = line[:len(line) - len(stripped)]
|
||
return prefix.replace("\t", "→").replace(" ", "·") + stripped
|
||
|
||
|
||
def find_closest_lines(old_string: str, content: str, context_lines: int = 2, max_results: int = 3) -> str:
|
||
"""
|
||
查找最接近旧片段锚点行的文件行，返回带行号的片段；没有候选时返回空字符串。
|
||
"""
|
||
if not old_string or not content:
|
||
return ""
|
||
old_lines = old_string.splitlines()
|
||
content_lines = content.splitlines()
|
||
if not old_lines or not content_lines:
|
||
return ""
|
||
|
||
anchor = old_lines[0].strip() or next((line.strip() for line in old_lines if line.strip()), "")
|
||
if not anchor:
|
||
return ""
|
||
|
||
scored = sorted(((SequenceMatcher(None, anchor, line.strip()).ratio(), i)
|
||
for i, line in enumerate(content_lines) if line.strip()), key=lambda x: -x[0])
|
||
top = [s for s in scored if s[0] > 0.3][:max_results]
|
||
if not top:
|
||
return ""
|
||
|
||
parts = []
|
||
seen_ranges = set()
|
||
for _, line_idx in top:
|
||
start = max(0, line_idx - context_lines)
|
||
end = min(len(content_lines), line_idx + len(old_lines) + context_lines)
|
||
if (start, end) in seen_ranges:
|
||
continue
|
||
seen_ranges.add((start, end))
|
||
parts.append("\n".join(
|
||
f"{start + j + 1:4d}| {content_lines[start + j]}" for j in range(end - start)))
|
||
result = "\n---\n".join(parts)
|
||
|
||
# 去除空白后已相同的候选行需可视化缩进，使模型能按文件实际缩进修正输入。
|
||
best_line = content_lines[top[0][1]]
|
||
if best_line.strip() == anchor and best_line != old_lines[0]:
|
||
result += (
|
||
"\n\nWhitespace difference detected (→ = tab, · = space):\n"
|
||
f" file has: {_visualize_whitespace(best_line)}\n"
|
||
f" you sent: {_visualize_whitespace(old_lines[0])}\n"
|
||
"Use the exact whitespace shown in 'file has'.")
|
||
return result
|
||
|
||
|
||
def format_no_match_hint(error: Optional[str], match_count: int,
|
||
old_string: str, content: str) -> str:
|
||
"""
|
||
仅为普通未匹配错误提供相似行提示。
|
||
|
||
歧义、转义漂移和内容相同等错误虽然也返回零匹配，但相似行提示会误导重试，因此不追加。
|
||
"""
|
||
if match_count != 0 or not error or not error.startswith("Could not find"):
|
||
return ""
|
||
hint = find_closest_lines(old_string, content)
|
||
return "\n\nDid you mean one of these sections?\n" + hint if hint else ""
|