previous episode

Version Control with Git

next episode

Exploring History

Overview

Teaching: 25 min
Exercises: 0 min
Questions

  • How can I identify old versions of files?

  • How do I review my changes?

  • How can I recover old versions of files?

Objectives

  • Explain what the HEAD of a repository is and how to use it.

  • Identify and use Git commit numbers.

  • Compare various versions of tracked files.

  • Restore old versions of files.

As we saw in the previous lesson, we can refer to commits by their identifiers. You can refer to the most recent commit of the working directory by using the identifier HEAD.

We’ve been adding one line at a time to README.md, so it’s easy to track our progress by looking, so let’s do that using our HEADs. Before we start, let’s make a change to README.md.

$ nano README.md
$ cat README.md

We want to classify the digits 0-9 using machine learning
We will start by training a Support Vector Machine
And only use Deep Learning if we really need to
An ill-considered change

Now, let’s see what we get.

$ git diff HEAD README.md

diff --git a/README.md b/README.md
index b36abfd..0848c8d 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,4 @@
 We want to classify the digits 0-9 using machine learning
 We will start by training a Support Vector Machine
 And only use Deep Learning if we really need to
+An ill-considered change.

which is the same as what you would get if you leave out HEAD (try it). The real goodness in all this is when you can refer to previous commits. We do that by adding ~1 to refer to the commit one before HEAD.

$ git diff HEAD~1 README.md

If we want to see the differences between older commits we can use git diff again, but with the notation HEAD~1, HEAD~2, and so on, to refer to them:

$ git diff HEAD~2 README.md

diff --git a/README.md b/README.md
index df0654a..b36abfd 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,4 @@
 We want to classify the digits 0-9 using machine learning
+We will start by training a Support Vector Machine
+And only use Deep Learning if we really need to
+An ill-considered change

We could also use git show which shows us what changes we made at an older commit as well as the commit message, rather than the differences between a commit and our working directory that we see by using git diff.

$ git show HEAD~2 README.md

commit 34961b159c27df3b475cfe4415d94a6d1fcd064d
Author: Bob <bob@github.com>
Date:   Thu Aug 22 10:07:21 2013 -0400

    Inital notes on classification

diff --git a/README.md b/README.md
new file mode 100644
index 0000000..df0654a
--- /dev/null
+++ b/README.md
@@ -0,0 +1 @@
+We want to classify the digits 0-9 using machine learning

In this way, we can build up a chain of commits. The most recent end of the chain is referred to as HEAD; we can refer to previous commits using the ~ notation, so HEAD~1 (pronounced “head minus one”) means “the previous commit”, while HEAD~123 goes back 123 commits from where we are now.

We can also refer to commits using those long strings of digits and letters that git log displays. These are unique IDs for the changes, and “unique” really does mean unique: every change to any set of files on any computer has a unique 40-character identifier. Our first commit was given the ID f22b25e3233b4645dabd0d81e651fe074bd8e73b, so let’s try this:

$ git diff f22b25e3233b4645dabd0d81e651fe074bd8e73b README.md

diff --git a/README.md b/README.md
index df0654a..93a3e13 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,4 @@
 We want to classify the digits 0-9 using machine learning
+We will start by training a Support Vector Machine
+And only use Deep Learning if we really need to
+An ill-considered change

That’s the right answer, but typing out random 40-character strings is annoying, so Git lets us use just the first few characters:

$ git diff f22b25e README.md

diff --git a/README.md b/README.md
index df0654a..93a3e13 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,4 @@
 We want to classify the digits 0-9 using machine learning
+We will start by training a Support Vector Machine
+And only use Deep Learning if we really need to
+An ill-considered change

All right! So we can save changes to files and see what we’ve changed—now how can we restore older versions of things? Let’s suppose we accidentally overwrite our file:

$ nano README.md
$ cat README.md

We will need more labeled data

git status now tells us that the file has been changed, but those changes haven’t been staged:

$ git status

On branch master
Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git checkout -- <file>..." to discard changes in working directory)

	modified:   README.md

no changes added to commit (use "git add" and/or "git commit -a")

We can put things back the way they were by using git checkout:

$ git checkout HEAD README.md
$ cat README.md

We want to classify the digits 0-9 using machine learning
We will start by training a Support Vector Machine
And only use Deep Learning if we really need to

As you might guess from its name, git checkout checks out (i.e., restores) an old version of a file. In this case, we’re telling Git that we want to recover the version of the file recorded in HEAD, which is the last saved commit. If we want to go back even further, we can use a commit identifier instead:

$ git checkout f22b25e README.md

$ cat README.md

We want to classify the digits 0-9 using machine learning

$ git status

# On branch master
Changes to be committed:
  (use "git reset HEAD <file>..." to unstage)
# Changes not staged for commit:
#   (use "git add <file>..." to update what will be committed)
#   (use "git checkout -- <file>..." to discard changes in working directory)
#
#	modified:   README.md
#
no changes added to commit (use "git add" and/or "git commit -a")

Notice that the changes are on the staged area. Again, we can put things back the way they were by using git checkout:

$ git checkout HEAD README.md

Don’t Lose Your HEAD

Above we used

$ git checkout f22b25e README.md

to revert README.md to its state after the commit f22b25e. If you forget README.md in that command, Git will tell you that “You are in ‘detached HEAD’ state.” In this state, you shouldn’t make any changes. You can fix this by reattaching your head using git checkout master

It’s important to remember that we must use the commit number that identifies the state of the repository before the change we’re trying to undo. A common mistake is to use the number of the commit in which we made the change we’re trying to get rid of. In the example below, we want to retrieve the state from before the most recent commit (HEAD~1), which is commit f22b25e:

Git Checkout

So, to put it all together, here’s how Git works in cartoon form:

https://figshare.com/articles/How_Git_works_a_cartoon/1328266

Simplifying the Common Case

If you read the output of git status carefully, you’ll see that it includes this hint:

(use "git checkout -- <file>..." to discard changes in working directory)

As it says, git checkout without a version identifier restores files to the state saved in HEAD. The double dash -- is needed to separate the names of the files being recovered from the command itself: without it, Git would try to use the name of the file as the commit identifier.

The fact that files can be reverted one by one tends to change the way people organize their work. If everything is in one large document, it’s hard (but not impossible) to undo changes to the introduction without also undoing changes made later to the conclusion. If the introduction and conclusion are stored in separate files, on the other hand, moving backward and forward in time becomes much easier.

Recovering Older Versions of a File

Jennifer has made changes to the Python script that she has been working on for weeks, and the modifications she made this morning “broke” the script and it no longer runs. She has spent ~ 1hr trying to fix it, with no luck…

Luckily, she has been keeping track of her project’s versions using Git! Which commands below will let her recover the last committed version of her Python script called data_cruncher.py?

  1. $ git checkout HEAD

  2. $ git checkout HEAD data_cruncher.py

  3. $ git checkout HEAD~1 data_cruncher.py

  4. $ git checkout <unique ID of last commit> data_cruncher.py

Solution

Both 2 and 4

Reverting a Commit

Jennifer is collaborating on her Python script with her colleagues and realizes her last commit to the group repository is wrong and wants to undo it. Jennifer needs to undo correctly so everyone in the group repository gets the correct change. git revert [wrong commit ID] will make a new commit that undoes Jennifer’s previous wrong commit. Therefore git revert is different than git checkout [commit ID] because checkout is for local changes not committed to the group repository. Below are the right steps and explanations for Jennifer to use git revert, what is the missing command?

  1. ____ # Look at the git history of the project to find the commit ID

  2. Copy the ID (the first few characters of the ID, e.g. 0b1d055).

  3. git revert [commit ID]

  4. Type in the new commit message.

  5. Save and close

    Solution

    git log

Understanding Workflow and History

What is the output of cat svm/classify.py at the end of this set of commands?

$ cd digits-classifier
$ nano svm/classify.py #input the following text: import matplotlib
$ git add svm/classify.py
$ nano svm/classify.py #add the following text: import tensorflow as tf
$ git commit -m "Imports tensorflow"
$ git checkout HEAD svm/classify.py
$ cat svm/classify.py #this will print the contents of svm/classify.py to the screen

1.

import tensorflow as tf

2.

import matplotlib

3.

import matplotlib
import tensorflow as tf

4.

Error because you have changed svm/classify.py without committing the changes

Solution

Line by line:

$ cd digits-classifier

Enters into the ‘digits-classifier’ directory

$ nano svm/classify.py #input the following text: import matplotlib

We created a new file and wrote a sentence in it, but the file is not tracked by git.

$ git add svm/classify.py

Now the file is staged. The changes that have been made to the file until now will be committed in the next commit.

$ nano svm/classify.py #add the following text: import tensorflow as tf

The file has been modified. The new changes are not staged because we have not added the file.

$ git commit -m "Imports tensorflow"

The changes that were staged (import matplotlib) have been committed. The changes that were not staged (import tensorflow as tf) have not. Our local working copy is different than the copy in our local repository.

$ git checkout HEAD svm/classify.py

With checkout we discard the changes in the working directory so that our local copy is exactly the same as our HEAD, the most recent commit.

$ cat svm/classify.py #this will print the contents of svm/classify.py to the screen

If we print svm/classify.py we will get answer 2.

Checking Understanding of git diff

Consider this command: git diff HEAD~3 README.md. What do you predict this command will do if you execute it? What happens when you do execute it? Why?

Try another command, git diff [ID] README.md, where [ID] is replaced with the unique identifier for your most recent commit. What do you think will happen, and what does happen?

Getting Rid of Staged Changes

git checkout can be used to restore a previous commit when unstaged changes have been made, but will it also work for changes that have been staged but not committed? Make a change to README.md, add that change, and use git checkout to see if you can remove your change.

Explore and Summarize Histories

Exploring history is an important part of git, often it is a challenge to find the right commit ID, especially if the commit is from several months ago.

Imagine the digits-classifier project has more than 50 files. You would like to find a commit with specific text in README.md is modified. When you type git log, a very long list appeared, How can you narrow down the search?

Recall that the git diff command allow us to explore one specific file, e.g. git diff README.md. We can apply a similar idea here.

$ git log README.md

Unfortunately some of these commit messages are very ambiguous e.g. update files. How can you search through these files?

Both git diff and git log are very useful and they summarize a different part of the history for you. Is it possible to combine both? Let’s try the following:

$ git log --patch README.md

You should get a long list of output, and you should be able to see both commit messages and the difference between each commit.

Question: What does the following command do?

$ git log --patch HEAD~3 *.txt

Key Points

  • git diff displays differences between commits.

  • git checkout recovers old versions of files.

previous episode

next episode