github version rollback (reproduced)

This article is reproduced from https://www.liaoxuefeng.com/wiki/0013739516305929606dd18361248578c67b8067c8c017b000/0013744142037508cf42e51debf49668810645e02887691000

Personally recommend this blogger: Liao Xuefeng.

version rollback

Now that you have learned to modify files, and then submit the changes to the Git repository, now, practice again, modify the readme.txt file as follows:

Git is a distributed version control system.
Git is free software distributed under the GPL.

Then try to commit:

$ git add readme.txt
$ git commit -m "append GPL"
[master 3628164] append GPL
 1 file changed, 1 insertion(+), 1 deletion(-)

Like this, you keep making changes to the files, and then keep submitting the changes to the repository, just like when you play an RPG game, every time you pass a level, the game state will be automatically saved. Take the state of the previous level. Sometimes, before a boss fight, you will manually save it so that in case the boss fight fails, you can start over from the nearest place. The same is true for Git, whenever you feel that a file has been modified to a certain extent, you can "save a snapshot", which is called a snapshot in Git commit. Once you mess up your files, or delete them by mistake, you can commitrestore from the most recent one and continue working instead of losing months of work.

Now, let's review how many versions of the readme.txt file have been committed to the Git repository:

Version 1: write a readme file

Git is a version control system.
Git is free software.

Version 2: add distributed

Git is a distributed version control system.
Git is free software.

Version 3: append GPL

Git is a distributed version control system.
Git is free software distributed under the GPL.

Of course, in actual work, how can we remember in our minds what has been changed in a file with thousands of lines each time, otherwise what would a version control system do? The version control system must have some command that can tell us the history, in Git, we use the git logcommand to view:

$ git log
commit 3628164fb26d48395383f8f31179f24e0882e1e0
Author: Michael Liao <[email protected]>
Date:   Tue Aug 20 15:11:49 2013 +0800

    append GPL

commit ea34578d5496d7dd233c827ed32a8cd576c5ee85
Author: Michael Liao <[email protected]>
Date:   Tue Aug 20 14:53:12 2013 +0800

    add distributed

commit cb926e7ea50ad11b8f9e909c05226233bf755030
Author: Michael Liao <[email protected]>
Date:   Mon Aug 19 17:51:55 2013 +0800

    wrote a readme file

git logThe command shows the commit log from the most recent to the furthest, we can see 3 commits, the most recent is append GPL, the last is add distributed, the oldest is wrote a readme file. If you think that the output information is too much and you are dazzled, you can try adding --pretty=onelineparameters:

$ git log --pretty=oneline
3628164fb26d48395383f8f31179f24e0882e1e0 append GPL
ea34578d5496d7dd233c827ed32a8cd576c5ee85 add distributed
cb926e7ea50ad11b8f9e909c05226233bf755030 wrote a readme file

需要友情提示的是，你看到的一大串类似3628164...882e1e0的是commit id（版本号），和SVN不一样，Git的commit id不是1，2，3……递增的数字，而是一个SHA1计算出来的一个非常大的数字，用十六进制表示，而且你看到的commit id和我的肯定不一样，以你自己的为准。为什么commit id需要用这么一大串数字表示呢？因为Git是分布式的版本控制系统，后面我们还要研究多人在同一个版本库里工作，如果大家都用1，2，3……作为版本号，那肯定就冲突了。

每提交一个新版本，实际上Git就会把它们自动串成一条时间线。如果使用可视化工具查看Git历史，就可以更清楚地看到提交历史的时间线：

git-log-timeline

好了，现在我们启动时光穿梭机，准备把readme.txt回退到上一个版本，也就是“add distributed”的那个版本，怎么做呢？

首先，Git必须知道当前版本是哪个版本，在Git中，用HEAD表示当前版本，也就是最新的提交3628164...882e1e0（注意我的提交ID和你的肯定不一样），上一个版本就是HEAD^，上上一个版本就是HEAD^^，当然往上100个版本写100个^比较容易数不过来，所以写成HEAD~100。

现在，我们要把当前版本“append GPL”回退到上一个版本“add distributed”，就可以使用git reset命令：

$ git reset --hard HEAD^
HEAD is now at ea34578 add distributed

--hard参数有啥意义？这个后面再讲，现在你先放心使用。

看看readme.txt的内容是不是版本add distributed：

$ cat readme.txt
Git is a distributed version control system.
Git is free software.

果然。

还可以继续回退到上一个版本wrote a readme file，不过且慢，然我们用git log再看看现在版本库的状态：

$ git log
commit ea34578d5496d7dd233c827ed32a8cd576c5ee85
Author: Michael Liao <[email protected]>
Date:   Tue Aug 20 14:53:12 2013 +0800

    add distributed

commit cb926e7ea50ad11b8f9e909c05226233bf755030
Author: Michael Liao <[email protected]>
Date:   Mon Aug 19 17:51:55 2013 +0800

    wrote a readme file

最新的那个版本append GPL已经看不到了！好比你从21世纪坐时光穿梭机来到了19世纪，想再回去已经回不去了，肿么办？

办法其实还是有的，只要上面的命令行窗口还没有被关掉，你就可以顺着往上找啊找啊，找到那个append GPL的commit id是3628164...，于是就可以指定回到未来的某个版本：

$ git reset --hard 3628164
HEAD is now at 3628164 append GPL

版本号没必要写全，前几位就可以了，Git会自动去找。当然也不能只写前一两位，因为Git可能会找到多个版本号，就无法确定是哪一个了。

再小心翼翼地看看readme.txt的内容：

$ cat readme.txt
Git is a distributed version control system.
Git is free software distributed under the GPL.

果然，我胡汉三又回来了。

Git的版本回退速度非常快，因为Git在内部有个指向当前版本的HEAD指针，当你回退版本的时候，Git仅仅是把HEAD从指向append GPL：

git-head

改为指向add distributed：

git-head-move

然后顺便把工作区的文件更新了。所以你让HEAD指向哪个版本号，你就把当前版本定位在哪。

现在，你回退到了某个版本，关掉了电脑，第二天早上就后悔了，想恢复到新版本怎么办？找不到新版本的commit id怎么办？

在Git中，总是有后悔药可以吃的。当你用$ git reset --hard HEAD^回退到add distributed版本时，再想恢复到append GPL，就必须找到append GPL的commit id。Git提供了一个命令git reflog用来记录你的每一次命令：

$ git reflog
ea34578 HEAD@{0}: reset: moving to HEAD^
3628164 HEAD@{1}: commit: append GPL
ea34578 HEAD@{2}: commit: add distributed
cb926e7 HEAD@{3}: commit (initial): wrote a readme file

终于舒了口气，第二行显示append GPL的commit id是3628164，现在，你又可以乘坐时光机回到未来了。

小结

现在总结一下：

HEAD指向的版本就是当前版本，因此，Git允许我们在版本的历史之间穿梭，使用命令git reset --hard commit_id。
穿梭前，用git log可以查看提交历史，以便确定要回退到哪个版本。
要重返未来，用git reflog查看命令历史，以便确定要回到未来的哪个版本。