# Git - internals
# ressources
4:48 les fichiers objects (de
.git/objects/) sont compressés avec zlib.
The anatomy of a Git commit : Christoph Burgdorf - 20141118
# objects files
written in .git/objects/
list all objects files in .git/objects/ :
$ find .git/objects -type f
.git/objects/56/a6051ca2b02b04ef92d5150c9ef600403cb1de
.git/objects/ff/e298c3ce8bb07326f888907996eaa48d266db4
.git/objects/2e/65efe2a145dda7ee51d1741299f848e5bf752e
.git/objects/7b/35c3cb173e25102b0c26600ed2d45361c7a44e
.git/objects/27/4c0052dd5408f8ae2bc8440029ff67d79bc5c3
.git/objects/0e/ed1217a2947f4930583229987d90fe5e8e0b74
2
3
4
5
6
7
8
9
# 4 types of objects files
- commit (the commit description)
- tree (directory state)
- blob (content files state)
- annotated tag
# warning objects files creation
Theses files are created once the git add command is executed. It means you could have a lot of orphans objects (not linked with a ref)
if you have added to the index a lot of files or directories without putting them into a commit.
The command git prune -n (or git fsck --unreachable) display the list of orphans objects.
The command git prune -v delete the orphans objects then print the deleted list.
# immutability
Git objects files are immutable, once created, Git reads them, and that's all. There is no modification ever on it.
They are only deleted when the user execute the prune command if they are unreachable from any refs.
# lire les fichiers objects
La commande git cat-file <object-hash> -p permet d'afficher le contenu d'un fichier object.
Ex commit :
$ git cat-file 87a68b76ea470d7ad0f6ecbc5b46eda45e628e43 -p
tree 3852c9e63207a9ca2ccf00a5ba5233661ee6ff00
author Robert DUCHMOUL <robert.duchmoul@domain.tld> 1524570143 +0200
committer Robert DUCHMOUL <robert.duchmoul@domain.tld> 1524570143 +0200
test
2
3
4
5
6
Ex tree :
$ git cat-file 3852c9e63207a9ca2ccf00a5ba5233661ee6ff00 -p
100644 blob 274c0052dd5408f8ae2bc8440029ff67d79bc5c3 number.txt
2
Ex blob :
$ git cat-file 274c0052dd5408f8ae2bc8440029ff67d79bc5c3 -p
1234
2
La même commande en remplaçant -p par -t pour avoir le type de l'object.
$ git cat-file 87a68b76ea470d7ad0f6ecbc5b46eda45e628e43 -t
commit
$ git cat-file 3852c9e63207a9ca2ccf00a5ba5233661ee6ff00 -t
tree
$ git cat-file 274c0052dd5408f8ae2bc8440029ff67d79bc5c3 -t
blob
2
3
4
5
6
7
8
# lire le fichier index
$ git ls-files --stage
100644 274c0052dd5408f8ae2bc8440029ff67d79bc5c3 0 number.txt
2
What does the Git index contain EXACTLY?
# files structure
# blob file
A blob store the content of a file. Structure is header + content.
Header is : the file type (blob), a space, the length value in byte, a null char to terminate.
Ex : blob 16\u0000
A blob for a file containing what is up, doc? would output this object file content blob 16\u0000what is up, doc?.
Then Git compute the SHA1 for the object file content bd9dbf5aae1a3862dd1526723246b20206e5fc37.
Then Git compress the content with zlib.
Then Git create a new object file in .git/objects/.
The first two SHA1 hash chars are the directory (bd), the 38 last the object file name (9dbf5aae1a3862dd1526723246b20206e5fc37).
# tree file
Exactly the same as blob, but the content is not a file content but a list of the represented directory :
100644 blob 2e65efe2a145dda7ee51d1741299f848e5bf752e letter.txt
100644 blob 56a6051ca2b02b04ef92d5150c9ef600403cb1de number.txt
2
# commit file
Same but the content is different, it contains the referenced base tree, the author, the committer (could be different), a blank line, then the commit message (could be on several lines) :
tree 83207f0274383b4a79ff6d6c297e95204ba961bc
author Matthew Brett <matthew.brett@gmail.com> 1487004984 +0000
committer Matthew Brett <matthew.brett@gmail.com> 1487004984 +0000
An example commit
2
3
4
5
# index file
It is a lot more complicated and support structure versions.
Cf appropriate documentation in the Git repo itself.