Blog 1

Random Talk on Random Thoughts

Git Object ID Generation (1): Blobs and Commits

| Comments | ∞

The SHA-1 hash for blobs isn’t so hard to generate, and the process is now well-known. I don’t repeat it here. The one for commits can be similarly generated.

For example, 5b7b566b8a07d4813ba9f08a326e169cf38ca20f is a hash of the repository of this blog.

Remark: The email shown below is fake, so the SHA-1 hash of HEAD isn’t real. For the reason of displaying that email, you may refer to the remark in Git Object Id Generation (4): General Trees.

$ git rev-parse head
5b7b566b8a07d4813ba9f08a326e169cf38ca20f
$ git cat-file -p 5b7b566b8a07d4813ba9f08a326e169cf38ca20f | tee test.txt
tree 2d864bcb7e4944e9d98b663649c79084692873c1
parent afcb4d97cb447112bd2e930159966d92b8e4754a
author vincent tam <demo@example.com> 1438859683 +0800
committer vincent tam <demo@example.com> 1438864942 +0800

a new post on some git low level commands

I *don't* go over the details of the book *Pro Git*, which is available
online.  I just select some important ones for quick reference.
$ wc -c test.txt
379 test.txt
$ printf "commit 379\0" > len.txt
$ cat len.txt test.txt | shasum
5b7b566b8a07d4813ba9f08a326e169cf38ca20f  -

We get the same SHA-1 hash. Therefore, the ID for Git commit objects is just the SHA-1 hash of the contents of the Git commits with the string commit {len}\0 inserted at the beginning, where {len} stands for the number of bytes of the Git commit object (which is stored in test.txt in the above situation).

Facts learnt

The printf command

The commands printf "\0" and printf "\000" doesn’t differ. Therefore, if a digit zero follows the null character, one may indicate the null character using hex digits: printf "\x000". printf will interpret \x00 as a null character and the trailing digit 0 won’t be mixed up with the characters on its left.

The wc command

Apart from counting words, the wc command provide various flags which return the number of lines, characters, etc. In the past, I knew three flags only: -l, -w, and -c. I thought that they told wc to count the number of lines, words and characters respectively. After running the wc command on strings that include accents (e.g. “café”, “resumé”, etc), I realised that I misunderstood the function of the -c flag, which actually counts the number of bytes of the input. To count the number of characters, use the -m flag instead.

Comments