RedBlueStrategy
Posts Tags About
RedBlueStrategy

Contents

Contents

How Git Manages Changes

Alexander Myasnikov avatarAlexander Myasnikov
     1495 words   7 minutes 

Git is a content-addressable storage system: all data is stored as objects inside the .git directory.
Each object in Git has its hash, which is used for identification.

Let’s see how common commands use Git’s structure to perform various operations.

We will experiment on the httpexpect repository at tag v2.16.0.

The commit hash that a branch points to is stored in .git/refs/heads/<branch>:

❯ cat .git/refs/heads/master

9be446356b6eddc852a20b420cbf19c3c53acca3

Verify with Git:

❯ git rev-parse master

9be446356b6eddc852a20b420cbf19c3c53acca3

The commit hash that a tag points to is stored in .git/refs/tags/<tag>.
For example:

❯ cat .git/refs/tags/v2.16.0

e6879c0c3e358e8400f3fc5e9677a48ceb661740

Verify with Git:

❯ git rev-parse v2.16.0

e6879c0c3e358e8400f3fc5e9677a48ceb661740

There is a special file .git/HEAD that contains a pointer to the current branch or commit.
For example:

❯ cat .git/HEAD

e6879c0c3e358e8400f3fc5e9677a48ceb661740

If we switch to the master branch, HEAD looks like this:

❯ cat .git/HEAD

ref: refs/heads/master

All objects in Git are immutable and stored in .git/objects.
Each object’s name is a hash computed from its content.

Objects are stored in the compressed format <type> <size>\0<content>, where type can be:

  • blob - file content
  • tree - list of files and directories
  • commit - snapshot of the project with metadata
  • tag - annotated tag with metadata

Compute the hash for the existing file formatter.go:

❯ {
 echo -n "blob "
 echo -n $(cat formatter.go | wc -c)
 echo -ne "\0"
 cat formatter.go
} | sha1

02985f31c63a87435b23c9cdaa4837b355300446

Verify the object exists and hashes match:

❯ cat .git/objects/02/985f31c63a87435b23c9cdaa4837b355300446 | perl -MCompress::Zlib -0777 -e 'print uncompress <>' | sha1

02985f31c63a87435b23c9cdaa4837b355300446

Currently, HEAD points to commit e6879c0c3e358e8400f3fc5e9677a48ceb661740.
Let’s find the object with this hash and inspect its structure:

❯ cat .git/objects/e6/879c0c3e358e8400f3fc5e9677a48ceb661740 | perl -MCompress::Zlib -0777 -e 'print uncompress <>' | tr '\0' '\n'

commit 234
tree c2635674529d78a11624302cc23480a4d00e6984
parent 420f3aeeaa0c45bfac885856ad24dd9c2569d14b
author Victor Gaydov <victor@enise.org> 1696324180 +0400
committer Victor Gaydov <victor@enise.org> 1696324220 +0400

Refine colorhttp func

Explanation:

  • commit – object type.
  • tree – pointer to the file tree at the time of the commit.
  • parent – previous commit, needed for tracking project history.
  • author – commit author and date.
  • committer – the person who applied the commit and date.
  • Then comes the commit message.

Verify with Git command:

❯ git log -1 e6879c0c3e358e8400f3fc5e9677a48ceb661740

commit e6879c0c3e358e8400f3fc5e9677a48ceb661740 (HEAD, tag: v2.16.0, origin/v2)
Author: Victor Gaydov <victor@enise.org>
Date:   2023-10-03 13:09:40 +0400

    Refine colorhttp func

git log gives a nicer format, but all information is present.

Check all files in .git/refs (and .git/HEAD) and compare hashes.
If hashes match, the filename/path contains the branch/tag name:

❯ grep -rl "e6879c0c3e358e8400f3fc5e9677a48ceb661740" .git/refs .git/HEAD

.git/refs/tags/v2.16.0
.git/refs/remotes/origin/v2
.git/HEAD

Let’s see which files existed at the time of commit e6879c0c3e358e8400f3fc5e9677a48ceb661740.
The file tree for this commit is in object c2635674529d78a11624302cc23480a4d00e6984.

The file contains hashes in binary format, so to display them nicely:

❯ cat .git/objects/c2/635674529d78a11624302cc23480a4d00e6984 | perl -MCompress::Zlib -0777 -e '
    $_ = uncompress(<STDIN>);
    s/^tree \d+\0//;
    while (/(.*?)\0(.{20})/sg) {
      my ($header, $sha) = ($1, $2);
      $header =~ /^(\d+) (.*)$/;
      my $hex = unpack("H*", $sha);
      print "$1\t$hex\t$2\n";
    }
  '

40000	d3f3c0e53b33d211697bea88a56f7e62deb6d115	.github
100644	6bcd33c7f8945c6526bc2b3442fc591946448eff	.gitignore
100644	4f0aa540e6980fd3fe27f6921b923541a9b5f469	.golangci.yml
100644	f4e3cec654057e6b7d011f9d004fc17e412393d4	.ignore
100644	da361dcc087c3d081a5ceae48ae064f2e6df9260	.spelling
100644	c72b02ee8e98654ae8b92732a0c8429a17e1ba51	HACKING.md
100644	a022050415f901d9e2bb76880f7e14a879c70404	LICENSE
100644	6f31bfaa909a0f435076e73140b029e49750428b	Makefile
100644	6a9971e0d3ae6df648ac98e61deb32d0e8d9ebd8	README.md
40000	7ccc58aa1fb590b1f94a3279c48b1b6b706ef46d	_examples
40000	b8eeb9c418ccc558c14b1fe3da6fac0ce3cd5234	_images
100644	c1314d2f943727a232fd6eefe442132374653ccd	array.go
100644	bde4500abf7dcc8c6bc5f425aac33a8bf0a3816c	array_test.go
100644	d8dcb77078e1fa667793adb5096f180c42e21210	assertion.go
100644	17c863cbc3e9f82ae39158eb1a8c859ed54cf9f9	assertion_test.go
100644	0880e663ddaff0cd07664a080682e9f2bb07b7b2	assertion_validation.go
100644	0b0a6ddd1667bcf0b71c1bc1a8b0d233c1da744d	assertionseverity_string.go
100644	9ed2e4f0749aa8aaa54813624c41fe3b4b2022b4	assertiontype_string.go
100644	f1b3d171938aa8ba1b69141388519c1cb35763b1	binder.go
... and other changes

To verify, you can use git cat-file -p c2635674529d78a11624302cc23480a4d00e6984.

Or use git ls-tree to list files:

❯ git ls-tree -r e6879c0c3e358e8400f3fc5e9677a48ceb661740
100644 blob d276992856053ee54e26f7bd1fbe237ad1e08db6	.github/FUNDING.yml
100644 blob dd2452e4e151b3bd3cea38918d66153561ecf68a	.github/workflows/build.yaml
100644 blob 25e11982853085f09bc1c6161bab7b793ff48873	.github/workflows/detect_conflicts.yml
100644 blob 6bcd33c7f8945c6526bc2b3442fc591946448eff	.gitignore
100644 blob 4f0aa540e6980fd3fe27f6921b923541a9b5f469	.golangci.yml
100644 blob f4e3cec654057e6b7d011f9d004fc17e412393d4	.ignore
100644 blob da361dcc087c3d081a5ceae48ae064f2e6df9260	.spelling
100644 blob c72b02ee8e98654ae8b92732a0c8429a17e1ba51	HACKING.md
100644 blob a022050415f901d9e2bb76880f7e14a879c70404	LICENSE
100644 blob 6f31bfaa909a0f435076e73140b029e49750428b	Makefile
100644 blob 6a9971e0d3ae6df648ac98e61deb32d0e8d9ebd8	README.md
100644 blob 5812292bb2a4d0db578a4a2eb9740549585de472	_examples/.golangci.yml
100644 blob 0ce3624bfde7c2531957bbf752ed3ca0dfea6f70	_examples/doc.go
100644 blob d70ab083953bf89ae9d0314f273e63c37efd6abd	_examples/echo.go
100644 blob 548c51cfc20ac67bac5d19636eaf2beb0991c0f0	_examples/echo_test.go
100644 blob f2ab214fe185e1d93235d15852e2609d4bff34a5	_examples/fasthttp.go
100644 blob fefacee3dbb47ab5a5f74adaf9cd282d13dfe347	_examples/fasthttp_test.go
100644 blob fa15533fd7ea55ccc41717eff0ab3804e7439f6c	_examples/formatter_test.go
... and other changes

The only difference is that git ls-tree displays only a list of files.
We have displayed a list of files and directories located in the root directory.
To get all the files, you just need to recursively go through all the directories in the same way.

The structure of the objects is clear, so for brevity, we will use git cat-file.

To do this, you need to compare the file trees of the previous commit with the file tree of the current commit.
From the parent field 420f3aeeaa0c45bfac885856ad24dd9c2569d14b, we get the parent commit tree - d4259cdf526369da146cf6195148e2309a9a08c6

We compare the trees using the git diff command and see that the contents of the formatter.go file have changed:

❯ git diff --text --no-index <(git cat-file -p d4259cdf526369da146cf6195148e2309a9a08c6) <(git cat-file -p c2635674529d78a11624302cc23480a4d00e6984) | cat

diff --git a/dev/fd/13 b/dev/fd/15
--- a/dev/fd/13
+++ b/dev/fd/15
@@ -38,7 +38,7 @@
 100644 blob 585a0a1b341ddc9baebf6868f932b4c17f08fd5e	expect_test.go
-100644 blob 7b8439619f48224b440dda08c3058a1ce9bafe3d	formatter.go
+100644 blob 02985f31c63a87435b23c9cdaa4837b355300446	formatter.go
 100644 blob a78d1d2555fcb3486a95fc2e2439a750243efdde	formatter_test.go

Compare two blob objects:

❯ git diff --text --no-index <(git cat-file -p 7b8439619f48224b440dda08c3058a1ce9bafe3d) <(git cat-file -p 02985f31c63a87435b23c9cdaa4837b355300446) | cat

diff --git a/dev/fd/13 b/dev/fd/15
--- a/dev/fd/13
+++ b/dev/fd/15
@@ -1009,93 +1009,67 @@ var defaultTemplateFuncs = template.FuncMap{
 	},
-	"colorhttp": func(enable bool, colorName string, isResponse bool, input string) string {
+	"colorhttp": func(enable bool, isResponse bool, input string) string {
 		if !enable {

... and other changes

To verify this, we run git log -p and see that these were indeed the changes made:

❯ git log -p | head -n 20

commit e6879c0c3e358e8400f3fc5e9677a48ceb661740
Author: Victor Gaydov <victor@enise.org>
Date:   2023-10-03 13:09:40 +0400

    Refine colorhttp func

diff --git a/formatter.go b/formatter.go
index 7b84396..02985f3 100644
--- a/formatter.go
+++ b/formatter.go
@@ -1009,93 +1009,67 @@ var defaultTemplateFuncs = template.FuncMap{
 		}
 		return color.New(colorAttr).Sprint(input)
 	},
-	"colorhttp": func(enable bool, colorName string, isResponse bool, input string) string {
+	"colorhttp": func(enable bool, isResponse bool, input string) string {
 		if !enable {
 			return input
 		}

Make a small change in chain_test.go and add it to the index.

Compare .git/index with a previous copy:

❯ git diff --text --no-index <(xxd /tmp/old_index) <(xxd ./.git/index) | cat

diff --git a/dev/fd/13 b/dev/fd/15
--- a/dev/fd/13
+++ b/dev/fd/15
@@ -258,10 +258,10 @@
 00001010: 21a3 f429 0000 81a4 0000 01f5 0000 0000  !..)............
 00001020: 0000 31e0 94fc 11ee 2e63 9aac 4d09 5213  ..1......c..M.R.
 00001030: f026 b4e7 44ab 4656 0008 6368 6169 6e2e  .&..D.FV..chain.
-00001040: 676f 0000 6916 3bed 26c1 a7a8 6916 3bed  go..i.;.&...i.;.
-00001050: 26c1 a7a8 0100 0010 21a4 f54d 0000 81a4  &.......!..M....
-00001060: 0000 01f5 0000 0000 0000 53a3 b767 65b3  ..........S..ge.
-00001070: 3d7f ae0f 59fc 10e2 423c e7ea 4581 16bc  =...Y...B<..E...
+00001040: 676f 0000 6916 3c31 000d 44d6 6916 3c30  go..i.<1..D.i.<0
+00001050: 3b60 3632 0100 0010 21a4 f54d 0000 81a4  ;`62....!..M....
+00001060: 0000 01f5 0000 0000 0000 53a6 d0bd f288  ..........S.....
+00001070: e37d 62c7 7e6c 2ea6 3254 8044 2b46 a941  .}b.~l..2T.D+F.A
 00001080: 000d 6368 6169 6e5f 7465 7374 2e67 6f00  ..chain_test.go.
 00001090: 0000 0000 6916 2983 3712 b6d4 6916 2983  ....i.).7...i.).
 000010a0: 3712 b6d4 0100 0010 21a3 f42a 0000 81a4  7.......!..*....

I don’t know how to make it clearer, but here you can see that the hash has changed from b76765b33d7fae0f59fc10e2423ce7ea458116bc to d0bdf288e37d62c77e6c2ea6325480442b46a941.
Let’s compare the files with these hashes:

❯ git diff --text --no-index <(git cat-file -p b76765b33d7fae0f59fc10e2423ce7ea458116bc) <(git cat-file -p d0bdf288e37d62c77e6c2ea6325480442b46a941) | cat
diff --git a/dev/fd/13 b/dev/fd/15
--- a/dev/fd/13
+++ b/dev/fd/15
@@ -822,7 +822,7 @@ func TestChain_TestingTB(t *testing.T) {
 			want: true,
 		},
 		{
-			name: "AssertReporter",
+			name: "AssertReporterNew",
 			args: args{
 				handler: &DefaultAssertionHandler{
 					Formatter: newMockFormatter(t),

The changed hash corresponds to the updated file, which git diff HEAD also shows.

❯ git diff HEAD | cat
diff --git a/chain_test.go b/chain_test.go
index b76765b..d0bdf28 100644
--- a/chain_test.go
+++ b/chain_test.go
@@ -822,7 +822,7 @@ func TestChain_TestingTB(t *testing.T) {
 			want: true,
 		},
 		{
-			name: "AssertReporter",
+			name: "AssertReporterNew",
 			args: args{
 				handler: &DefaultAssertionHandler{
 					Formatter: newMockFormatter(t),

A lightweight tag is just a pointer to a commit:

❯ cat .git/refs/tags/v2.16.0

e6879c0c3e358e8400f3fc5e9677a48ceb661740

Let’s create an annotated tag with the command git tag -a v2.16.0-1 -m “version v2.16.0-1” and see what it points to:

❯ cat .git/refs/tags/v2.16.0-1

fd8a701b59285ffd3b143cf7973ae2ba67b1f9fd

An annotated tag is a new object of type tag that contains metadata and a pointer to commit e6879c0c3e358e8400f3fc5e9677a48ceb661740:

❯ cat .git/objects/fd/8a701b59285ffd3b143cf7973ae2ba67b1f9fd | perl -MCompress::Zlib -0777 -e 'print uncompress <>' | tr '\0' '\n'

tag 171
object e6879c0c3e358e8400f3fc5e9677a48ceb661740
type commit
tag v2.16.0-1
tagger Alexander Myasnikov <myasnikov.alexander.s@gmail.com> 1763062383 +0300

version v2.16.0-1
Updated on 2025-11-13
 Git, Question
 | Home