Chapter_10

10 - Git Internals

1 - Plumbing and Porcelain

• Files and folders we care about:
  • HEAD
  • index
  • objects
  • refs

2 - Git Objects

1 - Objects

• Git is a content-addressable filesystem
• Key-value store
• Any type of file can be inserted
• git hash-object can add the file to the db
• git cat-file can display the data

2 - Tree Objects

• Tree objects contain a list of sub objects and their names
• Can be used to store commit info as well as directory structure and file name
• Even be used for just a single file
• git cat-file -p master^{tree} shows the master tree for the HEAD commit

3 - Commit Objects

• git commit-tree adds your tree to the index
• Commit objects specify the top level tree for the snapshot, the parent commits, the author and commiter information, a blank line, finally the commit message

4 - Object Storage

• Objects have a header with the size of the content and then the content
• And SHA1 is deterministically generated from the data and the header
• Finally the blob is compressed with zlib

3 - Git References

References

• Git stores references to SHA1's in the refs folder
• git update-ref can change references
• A branch is a simple reference

The HEAD

• Head is a symbolic link to a branch
• If you are in a the "detached head" state then it will actually have data in it

Tags

• Simple tags are a dumb ref to a tree
• Annotated tags contain commit like data and a reference to the commit object that is tagged
• Technically you can tag any object
  • Public keys for instance can be stored and tagged

Remotes

• Stored in the refs subdirectory
• refer to the remote you have saved

4 - Packfiles

1 - Packfiles

• git gc will force packing files
• Git can look at blobs that have similar patterns and store one copy and the deltas

5 - The Refspec

1 - Refspec

• Tells git how to track the remote
• Format is (+?)<source location>:<destination location>
  • +refs/heads/*:refs/remotes/origin/* as a default example for origin
• You can use the refspec to delete stuff on origin if you want, by leaving off the source

6 - Transfer Protocols

1 - Dumb Protocol

• Used by a clone without auth
• Uses the info/refs file which is created with the post-receive hook
• Uses basic HTTP

2 - Smart Protocol

• send-pack and receive-pack are used to work with the smart protocol to upload data
  • Work over SSH
• fetch-pack and upload-pack are used to download data

7 - Maintenance and Data Recovery

1 - Maintenance

• git gc --auto only does anything when you have a lot of loose files or packs
  • Configurable but around 7k loose files or 50 packs
• gc will pack up loose files into a pack

2 - Data Recovery

• git reflog records what HEAD has pointed to
• You can use this to find the lost commit hash
• git branch <name> <hash you want to save> will create a branch pointing to your lost commits
• git fsck --full will show you all dangling objects if you for instance lost your reflog
  • This probably implies that if you do a gc you will lose your data, that you have already gone out of your way to lose

3 - Removing Objects

• You can use filter-branch to rewrite history by git rm'ing the file you want gone
• Will have to inspect packs to find the sha1 of the file you care about and then find the commits that touched the file and then work backward from there

8 - Environment Variables

• Not super useful right now, but bookmark the page for when it comes in handy
• The Debugging env vars are going to be great