Research and define files migration without losing git history

[pcapellan]

Main issue
wazuh/wazuh-packages#2904

Description

Currently, we are facing a migration process from different repositories to a new one, in a normal migration, all the changes history of the files are lost since they are just 'moved' from one repository to another.

This issue aims to find a way to avoid this history loss and document the process that must be followed to achieve it.

Tasks

• Research for a method/plugin/process to avoid files git history loss
• Define a process to implement this solution

---

Conclusion

After some research and testing the different solutions, I would like to highlight two different methods that we could use to accomplish the migration. None of these methods requires any plugin or installation apart from git itself:

Method 1 - Git filter-branch

Note

The CppServer team has defined a slightly different approach to this one. Link

1. Clone the source repository and navigate to it:

   git clone <source_repository>
   cd <path/to/source/repository>

2. Remove origin remote:

   git remote rm origin

3. Select the directories to 'migrate' and clean the repository:

   git filter-branch -f --prune-empty --index-filter   'git rm --cached -r -q -- . ; git reset -q $GIT_COMMIT -- <dir_to_migrate_1> <dir_to_migrate_2> <dir_to_migrate_x>' -- <source_branch>

   We can list all the directories we want to migrate.

4. Switch to the target repository's directory and add the source repository as a temporary local remote

   cd <path/to/target/repository>
   git remote add <source_name> <path/to/source/repository>

5. Pull the files and their history from the source, and remove the temporary local remote

   git pull <source_name> <source_branch> --allow-unrelated-histories --no-rebase
   git remote rm <source_name>

Real life example

1. Clone the source repository and navigate to it:

   git clone https://github.com/wazuh/wazuh-jenkins.git
   cd wazuh-jenkins

2. Remove origin remote:

   git remote rm origin

3. Select the directories to 'migrate' and clean the repository

   git filter-branch -f --prune-empty --index-filter   'git rm --cached -r -q -- . ; git reset -q $GIT_COMMIT -- bootstrap/ new-jenkins/' -- master
   
   git gc --aggressive

4. Switch to the target repository's directory and add the source repository as a temporary local remote

   cd ~/new_repo
   git remote add jenkins /home/wazuh-jenkins

5. Pull the files and their history from the source, and remove the temporary local remote

   git pull jenkins master  --allow-unrelated-histories --no-rebase
   git remote rm jenkins

Check result:

~/new_repo$ ls
bootstrap  new-jenkins

Pros:

• Just a few easy steps
• We can list all the directories to migrate in only one command

Cons:

• Slow when executing on a big source repository
• Affects files on the source repository folder

Method 2 - Git split

1. Clone the source repository and navigate to it:

   git clone <source_repository>
   cd <path/to/source/repository>

2. Remove origin remote:

   git remote rm origin

3. Select the directories to 'migrate' and assign a new split branch to each one:

   git subtree split -P <dir_to_migrate_1> -b <split_branch_1>
   git subtree split -P <dir_to_migrate_2> -b <split_branch_2>
   git subtree split -P <dir_to_migrate_x> -b <split_branch_x>

4. Switch to the target repository's directory and add the source repository as a local remote.

   cd <path/to/target/repository>
   git remote add <source_name> <path/to/source/repository>

5. Fetch and merge each split branch into a specific directory

   git fetch <source_name> <split_branch_1>
   git fetch <source_name> <split_branch_2>
   git fetch <source_name> <split_branch_3>
   
   git pull  <source_name> <split_branch_1> --no-rebase --allow-unrelated
   git pull  <source_name> <split_branch_2> --no-rebase --allow-unrelated
   git pull  <source_name> <split_branch_3> --no-rebase --allow-unrelated

6. Cleanup

   git remote rm <source_name>

Real life example

1. Clone the source repository and navigate to it:

   git clone https://github.com/wazuh/qa-integration-framework.git
   cd qa-integration-framework/

2. Remove origin remote:

   git remote rm origin

3. Select the directories to 'migrate' and assign a new split branch to each one:

   git subtree split -P src/wazuh_testing/tools/monitors/ -b integration_monitors
   git subtree split -P src/wazuh_testing/constants/ -b integration_constants

4. Switch to the target repository's directory and add the source repository as a local remote.

   cd ../new_repo
   git remote add integration /home/qa-integration-framework

5. Fetch and merge each split branch into a specific directory

   git fetch integration integration_monitors
   git fetch integration integration_constants
   
   git read-tree --prefix=monitors/ -u integration/integration_monitors
   git commit -am "Merge integration monitors to new repo"
   
   git read-tree --prefix=constants/ -u integration/integration_constants
   git commit -am "Merge integration constants to new repo"

6. Cleanup

   git remote rm <source_name>

Check results
bash ~/new_repo$ ls constants monitors

Pros:

• Steps are easy and easy to understand
• We can easily decide the destination folder for each directory/file migrated
• The source repository folder does not get any directories removed

Cons:

• Has more steps than the other option
• We have to create a split branch for each directory to migrate

[QU3B1M]

Update report

Research on git native capabilities to import code with its history from another repo, it seems like the tool is capable to do so, the option --allow-unrelated-histories exists for this purpose.

Testing the solution with a dummy repository:

1. Clone the source repository:

   git clone https://github.com/QU3B1M/hardhat-token-farm.git

2. Remove origin remote:

   git remote rm origin

3. Select the directory to 'migrate' and to keep the history from:

   git filter-branch --subdirectory-filter contracts/ -- --all

4. Clean up:

   git reset --hard
   git gc --aggressive
   git prune
   git clean -fd

5. Switch to the target repository's directory and add local remote to the source repository

   cd ../target
   git remote add contracts ../hardhat-token-farm/

6. Pull the files and its history from the source

   git pull contracts main --allow-unrelated-histories
   remote: Enumerating objects: 40, done.
   remote: Counting objects: 100% (40/40), done.
   remote: Compressing objects: 100% (28/28), done.
   remote: Total 40 (delta 10), reused 40 (delta 10), pack-reused 0
   Unpacking objects: 100% (40/40), 6.99 KiB | 1.40 MiB/s, done.
   From ../hardhat-token-farm
    * branch            main       -> FETCH_HEAD
    * [new branch]      main       -> contracts/main

7. Remove the local remote

   git remote rm contracts

8. Check the files and its history

   /target$ ls
   QBMToken.sol  TokenFarm.sol  test

   /target$ git log
   commit 3f68292c88864623b2148155823f485d964eaaa6 (HEAD -> master)
   Author: QU3B1M <[email protected]>
   Date:   Sat Apr 2 12:09:42 2022 -0300
   
       Issue tokens tests
   
   commit 1ea79f5b87e99a38bbda174c202175978d079b2e
   Author: QU3B1M <[email protected]>
   Date:   Sat Apr 2 11:46:24 2022 -0300
   
       Staking Tests
   
   commit d92e68634a02fc826fb1249731af379104fc1d2b

[jnasselle]

Review

Nice job @QU3B1M ! just a question: could you please check doing this with two repos? I mean, given repos A and B, create a new one C with some folders of A and some folders of B. This is just in case, because this is a more accurate scenario since we'll merge several repos into one

[QU3B1M]

On hold 🔐

This issue will remain on hold until 4.9.0 release testing is complete.

---

Issue resumed

[QU3B1M]

Update report

Working on a more realistic solution, having some troubles when migrating with "complex" paths. WIP

[damarisg]

We will work with Method 2 - Git split.

LGTM!