feat: Create PatchCoder for applying patch-formatted code edits

This commit is contained in:
Paul Gauthier (aider) 2025-04-14 14:36:24 -07:00
parent ffb743e108
commit 8cc8027b40

282
aider/coders/patch_coder.py Normal file
View file

@ -0,0 +1,282 @@
import pathlib
import re
from dataclasses import dataclass, field
from enum import Enum
from typing import Dict, List, Optional, Tuple
from ..dump import dump # noqa: F401
from .base_coder import Coder
# Adapted structures and types from apply_patch.py for parsing and applying
class ActionType(str, Enum):
ADD = "Add"
DELETE = "Delete"
UPDATE = "Update"
@dataclass
class Chunk:
orig_index: int = -1
del_lines: List[str] = field(default_factory=list)
ins_lines: List[str] = field(default_factory=list)
context_before: List[str] = field(default_factory=list) # Store context for validation/application
@dataclass
class PatchAction:
type: ActionType
path: str
new_content: Optional[str] = None # For Add
chunks: List[Chunk] = field(default_factory=list) # For Update
move_path: Optional[str] = None # For Update
class PatchCoder(Coder):
"""
A coder that uses a custom patch format for code modifications,
inspired by the format described in tmp.gpt41edits.txt.
"""
edit_format = "patch"
gpt_prompts = None # Prompts to be added later
def get_edits(self) -> List[PatchAction]:
"""
Parses the LLM response content (containing the patch) into a list of PatchAction objects.
"""
content = self.partial_response_content
if not content or not content.strip():
return []
try:
parsed_edits = self._parse_patch_content(content)
return parsed_edits
except Exception as e:
raise ValueError(f"Error parsing patch content: {e}")
def _parse_patch_content(self, content: str) -> List[PatchAction]:
"""
Parses the patch content string into a list of PatchAction objects.
This is a simplified parser based on the expected format. A more robust
implementation would adapt the full parser logic from apply_patch.py,
including context finding and validation against current file content.
"""
edits = []
lines = content.splitlines()
i = 0
in_patch = False
current_action = None
while i < len(lines):
line = lines[i]
i += 1
if line.strip() == "*** Begin Patch":
in_patch = True
continue
if not in_patch:
continue
if line.strip() == "*** End Patch":
if current_action:
edits.append(current_action)
in_patch = False
break # End of patch found
# Match Action lines (Update, Add, Delete)
match = re.match(r"\*\*\* (Update|Add|Delete) File: (.*)", line)
if match:
if current_action:
edits.append(current_action) # Save previous action
action_type_str, path = match.groups()
action_type = ActionType(action_type_str)
path = path.strip()
current_action = PatchAction(type=action_type, path=path)
# Check for optional Move to line immediately after Update
if action_type == ActionType.UPDATE and i < len(lines):
move_match = re.match(r"\*\*\* Move to: (.*)", lines[i])
if move_match:
current_action.move_path = move_match.group(1).strip()
i += 1 # Consume the move line
continue
if not current_action:
# Skip lines before the first action inside the patch
continue
# Handle content for Add action
if current_action.type == ActionType.ADD:
if current_action.new_content is None:
current_action.new_content = ""
# Assuming ADD content starts immediately and uses '+' prefix
if line.startswith("+"):
current_action.new_content += line[1:] + "\n"
else:
# Or maybe ADD content is just raw lines until next ***?
# This part needs clarification based on exact format spec.
# Assuming '+' prefix for now. If not, adjust logic.
pass # Ignore lines not starting with '+' in ADD? Or raise error?
continue
# Handle chunks for Update action
if current_action.type == ActionType.UPDATE:
# This simplified parser doesn't handle @@ context or chunk boundaries well.
# It assumes a simple sequence of context, '-', '+' lines per chunk.
# A real implementation needs the state machine from apply_patch.py's peek_next_section.
# Placeholder: treat consecutive -,+ blocks as single chunk for simplicity.
if not current_action.chunks:
current_action.chunks.append(Chunk()) # Start first chunk
chunk = current_action.chunks[-1]
if line.startswith("-"):
chunk.del_lines.append(line[1:])
elif line.startswith("+"):
chunk.ins_lines.append(line[1:])
elif line.startswith("@@"):
# Context line - ignored by this simplified parser
pass
elif line.strip() == "*** End of File":
# EOF marker - ignored by this simplified parser
pass
else:
# Assume it's context line if not +/-/@@
# This simplified parser doesn't store context properly.
pass
continue
if in_patch and not current_action:
# Started patch but no actions found before end?
pass # Or raise error?
if in_patch and current_action:
# Reached end of content without *** End Patch
edits.append(current_action) # Append the last action
# Consider raising a warning or error about missing End Patch sentinel
return edits
def apply_edits(self, edits: List[PatchAction]):
"""
Applies the parsed PatchActions to the corresponding files.
"""
if not edits:
return
for action in edits:
full_path = self.abs_root_path(action.path)
path_obj = pathlib.Path(full_path)
try:
if action.type == ActionType.ADD:
if path_obj.exists():
# According to apply_patch.py, Add should fail if file exists.
# This check should ideally happen during parsing with file content access.
raise ValueError(f"ADD Error: File already exists: {action.path}")
if action.new_content is None:
raise ValueError(f"ADD change for {action.path} has no content")
# Ensure parent directory exists
path_obj.parent.mkdir(parents=True, exist_ok=True)
self.io.write_text(full_path, action.new_content.rstrip('\n') + '\n') # Ensure single trailing newline
elif action.type == ActionType.DELETE:
if not path_obj.exists():
# Allow deleting non-existent files (idempotent)
pass
else:
path_obj.unlink()
elif action.type == ActionType.UPDATE:
if not path_obj.exists():
# Update should fail if file doesn't exist (checked in apply_patch.py parser)
raise ValueError(f"UPDATE Error: File does not exist: {action.path}")
current_content = self.io.read_text(full_path)
if current_content is None:
raise ValueError(f"Could not read file for UPDATE: {action.path}")
# Apply the update logic using the parsed chunks
new_content = self._apply_update(current_content, action.chunks, action.path)
target_full_path = self.abs_root_path(action.move_path) if action.move_path else full_path
target_path_obj = pathlib.Path(target_full_path)
# Ensure parent directory exists for target
target_path_obj.parent.mkdir(parents=True, exist_ok=True)
self.io.write_text(target_full_path, new_content)
if action.move_path and full_path != target_full_path:
# Remove original file after successful write to new location
path_obj.unlink()
else:
raise ValueError(f"Unknown action type encountered: {action.type}")
except Exception as e:
# Raise a ValueError to signal failure, consistent with other coders.
raise ValueError(f"Error applying action '{action.type}' to {action.path}: {e}")
def _apply_update(self, text: str, chunks: List[Chunk], path: str) -> str:
"""
Applies UPDATE chunks to the given text content.
Requires accurate chunk information (indices, lines) from a robust parser.
This simplified version assumes chunks are sequential and indices are correct.
"""
if not chunks:
return text # No changes specified
orig_lines = text.splitlines() # Use splitlines() to match apply_patch.py behavior
dest_lines = []
last_orig_line_idx = -1 # Track the end of the last applied chunk in original lines
# apply_patch.py finds context during parsing. Here we assume indices are pre-validated.
# A robust implementation would re-validate context here or rely entirely on parser validation.
# Sort chunks? apply_patch.py implies they are processed in order found in patch.
# Chunks need accurate `orig_index` relative to the start of their *context* block.
# This simplified implementation lacks proper context handling and index calculation.
# It assumes `orig_index` is the absolute line number from start of file, which is incorrect
# based on apply_patch.py.
# --> THIS METHOD NEEDS REWRITING BASED ON A CORRECT PARSER <--
# For demonstration, let's process sequentially, assuming indices are somewhat meaningful.
current_orig_line_num = 0
for chunk in chunks:
# Placeholder: Assume chunk application logic here.
# This needs the sophisticated context matching and index handling from apply_patch.py.
# The current simplified parser doesn't provide enough info (like validated indices).
# Raising NotImplementedError until a proper parser/applier is integrated.
raise NotImplementedError(
"_apply_update requires a robust parser and context handling, similar to apply_patch.py"
)
# --- Hypothetical logic assuming correct indices ---
# chunk_start_index = chunk.orig_index # Needs correct calculation based on context
# if chunk_start_index < current_orig_line_num:
# raise ValueError(f"{path}: Overlapping or out-of-order chunk detected.")
#
# # Add lines between the last chunk and this one
# dest_lines.extend(orig_lines[current_orig_line_num:chunk_start_index])
#
# # Verify deleted lines match (requires normalization)
# num_del = len(chunk.del_lines)
# actual_deleted = orig_lines[chunk_start_index : chunk_start_index + num_del]
# # if normalized(actual_deleted) != normalized(chunk.del_lines):
# # raise ValueError(f"{path}: Mismatch in deleted lines for chunk at index {chunk_start_index}")
#
# # Add inserted lines
# dest_lines.extend(chunk.ins_lines)
#
# # Advance index past the deleted lines
# current_orig_line_num = chunk_start_index + num_del
# --- End Hypothetical ---
# Add remaining lines after the last chunk
dest_lines.extend(orig_lines[current_orig_line_num:])
return "\n".join(dest_lines) + "\n" # Ensure trailing newline