成为git砖家(13): 使用git am清理EasyCNN仓库

先前研究过 https://github.com/xylcbd/EasyCNN 仓库的git提交历史，发现有很多重复的commit，当时是用 smartgit 这一 GUI 软件查看 commit tree，手工删除重复 commit ，挺累的。

最近发现了 git format-patch 和 git am 命令，对于清理 EasyCNN 的重复的 git 提交历史很有帮助。这里记录下，真的挺有意思。

大体思路：生成原仓库每一 commit 的patch，按编号扫描 patch 文件，相邻 patch 文件如果名字相同（不考虑patch编号），并且 commit 时间相同，并且git diff相同，那就认为是重复的commit，仅留下第一个; 在新建目录创建仓库，逐一应用留下来的patch。

具体命令是：

bash 复制代码

cd EasyCNN
# 生成 patch
git format-patch --root

cd ..
mkdir ecnn
git am ../EasyCNN/0001-initialize-repo.patch
git am ../EasyCNN/0003-init-cnn-framework.patch
...
git am ../EasyCNN/0103-Update-PoolingLayer.cpp.patch

发现逐个patch上执行 git am 的过程很重复，于是让千问写了python脚本，也便于后人重复执行:

其中检查的时间复杂度是O(N), 只考虑相邻commit是相同commit的情况

python 复制代码

#!/usr/bin/env python3
import sys
import re
import subprocess
from pathlib import Path

def extract_subject_from_filename(filename: str) -> str:
    stem = Path(filename).stem
    return re.sub(r'^\d+-', '', stem)

def extract_diff_content(patch_file):
    with open(patch_file, 'r', encoding='utf-8', errors='replace') as f:
        lines = f.readlines()
    diff_lines = []
    in_diff = False
    for line in lines:
        if line.startswith('diff --git'):
            in_diff = True
        if in_diff:
            diff_lines.append(line)
    return ''.join(diff_lines).strip()

def deduplicate_patches_linear(patch_dir: Path, output_dir: Path):
    # 按文件名排序（0001, 0002, ...）
    patches = sorted(patch_dir.glob("*.patch"), key=lambda x: x.name)
    if not patches:
        print("❌ 没有找到 .patch 文件")
        return []

    unique_patches = []
    prev_subject = None
    prev_diff = None

    for p in patches:
        try:
            subject = extract_subject_from_filename(p.name)
            diff = extract_diff_content(p)

            # 如果和前一个 patch 主题相同 且 diff 相同 → 跳过
            if subject == prev_subject and diff == prev_diff:
                print(f"⏭️  跳过重复 patch: {p.name}")
                continue

            # 否则保留
            unique_patches.append(p)
            prev_subject = subject
            prev_diff = diff

        except Exception as e:
            print(f"⚠️ 跳过损坏文件 {p}: {e}")

    # 输出去重结果
    output_dir.mkdir(exist_ok=True)
    final_list = []
    for i, p in enumerate(unique_patches, 1):
        new_name = f"{i:04d}_{p.name}"
        dst = output_dir / new_name
        dst.write_text(p.read_text(encoding='utf-8', errors='replace'), encoding='utf-8')
        final_list.append(dst)

    print(f"✅ 原始 {len(patches)} 个 patch，去重后剩 {len(final_list)} 个")
    return final_list

def apply_patches_in_new_repo(patch_files, target_repo_dir):
    target = Path(target_repo_dir)
    if target.exists():
        print(f"❌ 目标目录 {target} 已存在，请先删除")
        return False

    subprocess.run(["git", "init", str(target)], check=True)

    for patch in patch_files:
        print(f"📦 应用 {patch.name} ...")
        result = subprocess.run(
            ["git", "-C", str(target), "am", str(patch)],
            capture_output=True,
            text=True
        )
        if result.returncode != 0:
            print(f"❌ 应用失败: {patch}")
            print(result.stderr)
            return False

    print(f"🎉 全部 {len(patch_files)} 个 patch 应用成功！仓库位于: {target}")
    return True

def main():
    if len(sys.argv) != 3:
        print("用法: python3 dedup.py <patch_dir> <new_repo_dir>")
        sys.exit(1)

    patch_dir = Path(sys.argv[1]).resolve()
    new_repo_dir = Path(sys.argv[2]).resolve()

    if not patch_dir.is_dir():
        print(f"❌ patch 目录不存在: {patch_dir}")
        sys.exit(1)

    dedup_dir = patch_dir.parent / "deduplicated_patches"
    unique_patches = deduplicate_patches_linear(patch_dir, dedup_dir)
    if not unique_patches:
        sys.exit(1)

    apply_patches_in_new_repo(unique_patches, new_repo_dir)

if __name__ == "__main__":
    main()