Blog 1

Random Talk on Random Thoughts

Git Object ID Generation (2): Trees With One Single Blob

| Comments | ∞

Background

In the first post in this series, I’ve claimed that the generation of object IDs in Git is the SHA-1 hash of the string

<object type name> SP <len> NUL <data>

,where

  • <data> stands for the output of git cat-file -p {hash}
  • <len> means the length of <data>. It can be measured with the command wc -c.

Problem

To verify my claim, I followed the steps in Chapter 4 of Version Control with Git.

  1. Create a folder named hello and go to that directory.
  2. Initialise an empty Git repository.
  3. Create the file hello.txt with one single line “hello world”.
  4. Add the file to Git’s object storage.
  5. Get a tree object from the index.
  6. Capture the contents of the tree object in test.txt.
  7. Count the number of bytes in test.txt.

  8. Create the file len.txt consisting of

     "tree" SP <result in item 6> NUL

    without the line terminator.

  9. Concatenate the contents of the files len.txt and test.txt and compute its SHA-1 hash.
$ mkdir hello && cd hello
$ git init
$ echo "hello world" > hello.txt
$ git add hello.txt
$ git write-tree
68aba62e560c0ebc3396e8ae9335232cd93a3f60
$ git cat-file -p 68aba62e560c0ebc3396e8ae9335232cd93a3f60 | tee test.txt
100644 blob 3b18e512dba79e4c8300dd08aeb37f8e728b8dad	hello.txt
$ wc -c test.txt
63 test.txt
$ printf "tree 63\0" > len.txt
$ cat len.txt test.txt | shasum
10bd0f0350027c25edc4ce72aba60e641f55596d  -

As can be seen above, I’ve a wrong SHA-1 hash. How can I get back the right SHA-1 hash?

Method

I googled “git tree hash id”, and I found the chosen answer of this Stack Overflow question very explanative. One may shorten the included command by replacing echo -en with printf.

Get the object size right

In this case, the object size of the tree 68aba62e560c0ebc3396e8ae9335232cd93a3f60 containg the blob 3b18e512dba79e4c8300dd08aeb37f8e728b8dad which corresponds to the file hello.txt should be 37 (= 6 + 1 + 9 + 1 + 20).

Get the object content right

As the blob ID is stored as binary value, I copied it and pasted it Vim so that I could easily insert \x in front of a pair of hex digits in the blob ID. I then put the things together in a command.

# Contents of `testing.sh' as seen inside Vim
# Note that there's NO newline character in the following command
printf "tree 37\x00100644 hello.txt\x00\x3b\x18\xe5\x12\xdb\xa7\x9e\x4
c\x83\x00\xdd\x08\xae\xb3\x7f\x8e\x72\x8b\x8d\xad" | shasum

Result

By executing the above command, I got the right SHA-1 hash: 68aba62e560c0ebc3396e8ae9335232cd93a3f60.

Fact learnt: formatting printf’s output

In the Stack Overflow question, there’s a command

find .git/objects/ -type f -printf "%h%f %s\n"
  • The flag -type f stands for files. Without this flag, directories like .git/objects will be displayed.
  • The flag -printf formats the output.
    • %h means the head of the file name without the last component of the file name. Thus, it expands to a the path of a directory without the trailing /.
    • %f means the last component of the file name. As a result, the / inside the displayed SHA-1 hashes are taken away.
    • %s means the file size

Comments