patch-fixer 0.5.1

Fixes erroneous git apply patches to the best of its ability.

patch-fixer

So you asked an LLM to generate a code diff, tried to apply it with git apply, and got a bunch of malformed patch errors? Well fear no more, patch-fixer is here to save the day... more or less.

This tool can also split patches into separate files based on file lists, making it easy to selectively apply changes.

Installation

# Make sure you're using at least python 3.10
python -m venv .venv/
source .venv/bin/activate
pip install patch-fixer

Usage

Command Line Interface

After installation, patch-fixer provides a unified command-line interface:

Fixing broken patches:

patch-fixer fix original broken.patch fixed.patch

where:

• original is the file or directory you were trying to patch
• broken.patch is the malformed patch generated by the LLM
• fixed.patch is the output file containing the (hopefully) fixed patch

Options:

• --fuzzy: enable fuzzy string matching for better context matching (experimental)
• --add-newline: add final newlines when processing "No newline at end of file" markers

Splitting patches by file:

# Split with files specified on command line
patch-fixer split input.patch included.patch excluded.patch -f file1.py file2.py

# Split using a file list
patch-fixer split input.patch included.patch excluded.patch -i files_to_include.txt

where:

• input.patch is the patch file to split
• included.patch will contain changes for the specified files
• excluded.patch will contain changes for all other files
• -f allows specifying files directly on the command line
• -i reads the file list from a text file (one file per line)

Python API

Fixing patches:

from patch_fixer import fix_patch

patch_file = "/path/to/broken.patch"
original = "/path/to/original/state"    # file or directory being patched
with open(patch_file, encoding="utf-8") as f:
    patch_lines = f.readlines()
    
# basic usage
fixed_lines = fix_patch(patch_lines, original)

# with fuzzy matching enabled
fixed_lines = fix_patch(patch_lines, original, fuzzy=True)

# with final newline addition
fixed_lines = fix_patch(patch_lines, original, add_newline=True)

output_file = "/path/to/fixed.patch"
with open(output_file, 'w', encoding='utf-8') as f:
    f.writelines(fixed_lines)

Splitting patches:

from patch_fixer import split_patch

with open("input.patch", encoding="utf-8") as f:
    patch_lines = f.readlines()

# split to include only specific files
files_to_include = ["./src/main.py", "./src/utils.py"]
included, excluded = split_patch(patch_lines, files_to_include)

# write the split patches
with open("included.patch", 'w', encoding='utf-8') as f:
    f.writelines(included)
    
with open("excluded.patch", 'w', encoding='utf-8') as f:
    f.writelines(excluded)

Advanced Features

Patch Analysis

Analyze patches without modifying them:

from patch_fixer.analyzer import analyze_patch, get_patch_summary

with open("patch.diff") as f:
    lines = f.readlines()

# Get detailed information
info = analyze_patch(lines)
print(f"Files: {info.total_files}, Added: {info.total_additions}, Removed: {info.total_deletions}")

# Get human-readable summary
print(get_patch_summary(lines))

Custom Processing Hooks

Add custom logic to patch processing:

def log_hunks(hunk_lines, context):
    print(f"Processing hunk in {context['current_file']}")
    return hunk_lines

fixed = fix_patch(patch_lines, original, pre_hunk_hook=log_hunks)

Known Limitations

• When fixing patches with missing index lines, the tool requires the files to be in a git repository to regenerate the index. This is only needed for file deletions and renames.
• patch-fixer assumes the patch follows git's unified diff format.
• Current implementation is not very robust to corrupted hunk content
  • Much more comprehensive fuzzy string matching is planned

Local Testing

git clone https://github.com/ajcm474/patch-fixer.git
cd patch-fixer
pip install -e .[test]
pytest

From version 0.3.0 onward (at least until version 1.0), some test failures are expected in bugfix versions as I like to use test-driven development to build out new features. Please only report test failures if the same test existed and passed in the most recent 0.x.0 version.

License

This is free and open source software, released under the Apache 2.0 License. See LICENSE for details.