IPADS Tutorial - Git


本次 Tutorial 分享的是 Git 的数据模型和基本用法。以 Git 的内部的数据模型(data model)入手,讲解 Git 内部对于项目、文件数据、历史等的抽象和管理,然后围绕科研和开发过程中的具体案例讲解 Git 的常用命令,帮助刚加入实验室的新同学快速了解 Git。


Version Control System

Git 是一个版本控制工具,关键点是 跟踪修改

Why version control?

Working by yourself

  • Look at old versions of a project

  • Keep a log of why certain changes were made

  • Work on parallel branches of development

Working with others

  • See what other people have changed, learn and review
  • Resolve conflicts in concurrent development

How to “learn” Git?

  • Git’s interface is a leaky abstraction, learning Git top-down (starting with its interface / command-line interface) can lead to a lot of confusion
  • Its underlying design and ideas are beautiful
  • Bottom-up explanation of Git, starting with its data model and later covering the command-line interface

Git 的接口抽象又复杂,但内部的设计非常的简洁,所以我们推荐自底向上学习它内部的逻辑,再来看接口如何映射到逻辑。

Thinking of history: story of snapshots

图中展示了线性版本之间的关系,早期的版本控制大家也确实是这么用的。但是 Git 没有用这个模型。

Git 使用的模型是有向无环图(DAG),它允许一个 snapshot 有多个父亲,Git 通过有向无环图这种方式维护历史。

Commit/Snapshot: who are you?

Snapshot is a collection of files and folders within some top-level directory

File is called a “blob”: a bunch of bytes.

A directory is called a “tree”: maps names to blobs or trees

  • directories can contain other directories

Commit 和 Snapshot 又是什么呢?在 Git 中他们把 文件目录 的结合,其中文件被称为“blob”(一堆字符);目录是一棵树,它可以包含 blobs 和 tree。

<root> (tree)
+—— foo (tree)
| |
| + bar.txt (blob, contents = "hello world")
+—— baz.txt (blob, content = "git is wonderful")

Data models

Data model as Code

// a file is a bunch of bytes
type blob = array<byte>
// a directory contains named files and directories
type tree = map<string, tree | blob>
// a commit has parents, metadata, and the top-level tree
type commit = struct {
parents: array<commit>
author: string
message: string
snapshot: tree

Objects and content-addressing

type object = blob | tree | commit
objects = map<string, object>

def store(object):
id = sha1(object)
objects[id] = object

def load(id):
return objects[id]

刚才提到的 blob、tree 和 commit 都可以归为 object,在 Git 中所有的 object 都是通过 SHA 哈希定位。

SHA-1 is not for Human, References are

Human-readable names for SHA-1 hashes, called references

  • References are mutable

  • E.g., the master/main references usually point to the latest commit in the main branch of development

Git 为了易用性引入了 references 概念,可以简单视为指针。我们平常使用的 master/main 其实都是 SHA-1 的代名词。

References as Code

references = map<string, string>

def update_reference(name, id):
references[name] = id

def read_reference(name):
return references[name]
def load_reference(name_or_id):
if name_or_id in references:
return load(references[name_or_id])
return load(name_or_id)

The last piece: Repositories & Staging Area

A Git repository: objects and references

Why staging area?

  • Clean snapshots
  • Git: allowing you to specify which modifications should be included in the next snapshot through a mechanism called the “staging area”.

Git 仓库就是 objects 和 reference,我们下载下来后可能会有 master 或者 main 分支,分支本身又是一个 reference,reference 会产生一个 id 指向 object。通过这样的方式 Git 维护整个项目。


Scenario-1: work on a local project

  • Start a new project with git init
  • Check status using git status
git init
git status

echo "hello git" >> hello.txt

git status
git add hello.txt
git status

git commit -m 'init commit'

git status

Check history using git log

git log

Switch to an older version: git checkout [commit_id]

Show changes on staging : git checkout [commit_id]

cd hello.txt
echo "new line" >> hello.txt
cat hello.txt

git diff hello.txt

Scenario-1: summary

  • Tracking history
  • A better way to manage your project
    • A single commit to implement a single functionalities
    • Easily roll-back to a workable version

简单总结我们发现 Git 的基本功能可以让你非常方便的去管理你的本地项目,可以让你做一些修改保存更新,也可以回滚到之前的状态做一些测试。

Tips: How to write a “useful” commit msg?

可以参考 Linux 社区的 commit 格式。

Command (finally…[2]

之前提到 Git 使用 DAG 模型,所以会有分支。在实际应用中我们的主要问题是:

  • 如何创建分支
  • 如何合并多个分支

Scenario-2: Debugging

  • You find a bug in your project
  • You need to add many logs to debug
  • Create and switch to a new branch: git checkout -b
  • Chekc the current branch: git branch

假设你在项目中发现了一个 bug,然后你希望在项目中加很多 log 和 printf 去 debug。我们可以选择切换到一个新的分支。

git status
git checkout -b debug

Merge debug branch into main: git merge

git commit -asm "debug: add debug info"

Merge debug branch into main: git merge

git checkout main
git merge debug

When you rush papers, you may have many branches, implementing features, test cases, debug infos

git rebase: Rebase is thought as one of the most complicated part in Git

简单来说,rebase 是让你在 git 维护的历史 DAG 上调整他们的结构*/*关系的

Case-1: you want to keep master and topic branches, but applies commits in topic branches based on latest master commits

git rebase master topic

Rebase vs. Merge

Rebase 和 Merge 最大的区别在于 merge 会创建一个新的 commit(如图所示的 M)以继承多个状态,而 rebase 则会把 E 消掉,改变其中的顺序关系。

  • Case-2: More branches rebase!
  • How to make topic based on master (without next’s commits)
git rebase --onto master next topic

  • Case-2: More branches rebase!
  • Similiar cases
git rebase --onto master topicA topicB

git rebase --onto topicA~5 topicA~3 topicA

Command (finally…3


  • git remote: list remotes

  • *git remote add: add a remote

  • *git push:: send objects to remote, and update remote reference

  • git branch –set-upstream-to=/**: set up correspondence between local and remote branch

  • git fetch: retrieve objects/references from a remote

  • git pull: same as git fetch; git merge

  • git clone: download repository from remote**

Scenario-3: Gitlab/Gitee/Github

基于 Git 的代码托管平台

  • Github(网络不一定好)
  • Gitee(国内用还是很靠谱的)
  • Gitlab(实验室项目)

定期的 pull/push 是个好习惯


  • 在代码仓库平台上合并修改
  • 代码 Review

Command (finally…4


  • git commit –amend: edit a commit’s contents/message
  • git reset HEAD: unstage a file
  • git checkout –: discard changes

Scenario-4: You will make mistakes, sometimes

You made a commit, but with wrong msg: git commit —amend

git commit --amend

You mistakenly add a file into stage area: git reset HEAD

git status
git add hello.txt
git status

git reset HEAD hello.txt
git status

You want to discard changes on some files: git checkout —

git status
git checkout -- hello.txt

git status

Command (finally…5


  • git config: Git is highly customizable
  • git clone –depth=1: shallow clone, without entire version history
  • git add -p: interactive staging
  • git rebase -i: interactive rebasing
  • git blame: show who last edited which line
  • git stash: temporarily remove modifications to working directory
  • git bisect: binary search history (e.g. for regressions)
  • .gitignore: specify intentionally untracked files to ignore

Scenario-5: Git can do more for you

Working in a team, who write the bug code?: git blame

git blame README.md

  • .gitignore: ignore the matched files

因为 Git 使用的是文件快照来保存版本历史,而二进制文件在压缩上几乎没有效果,所以,二进制文件只要有一点点修改,保存的就是整个文件内容。

所以大的二进制文件是禁止放到 Git 里面去管理的。那么多大才算大呢?一般的标准是单个二进制文件的大小不要超过 100kb。

Summary and Q&A?

  • Basic knowledge about git is necessary
  • More “advanced” tools (e.g., vscode) may help you use Git
  • Try to read Pro-Git (https://git-scm.com/book/en/v2) if you want to know more
  • Thx