Git Partial Clones

I recently was introduced to Sparse Directories in SVN. In SVN, you can initially clone a repository and have it be empty until you checkout the specific files needed. I wondered if I can do the same with git. For the tl;dr skip to the conclusion section.

As a benchmark, we’re going to reference the size of a cryptographic library I helped author. As a baseline, let’s see how big the repository is before adding any flags.

git clone https://github.com/symcollab/cryptosolve
du -sh cryptosolve


90M cryptosolve

Using the --filter flag

With the filter flag, blobs that fall under a specified criteria do not get automatically downloaded during a clone. The blobs do, however, get downloaded whenever its associated files get checked out. By setting the flag to blob:none, we are telling git to not download any files initially. Though since the main branch gets checked out by default during a clone, git will still download the blobs associated with the main branch.

git clone --filter=blob:none https://github.com/symcollab/cryptosolve
du -sh cryptosolve


2.1M cryptosolve

Using the --no-checkout flag

We can then improve the last command by adding the no-checkout flag. This flag will not construct any of the files in the current branch. If you don’t include include the filter flag from before, then there really isn’t much of a space savings since all the information is stored in the git database.

git clone --no-checkout https://github.com/symcollab/cryptosolve
du -sh cryptosolve


89M cryptosolve

You can see that there are no files checked out with a ls -a.

. .. .git

Though with the filter flag, we can see the space savings!

git clone \
    --no-checkout \
    --filter=blob:none \
    https://github.com/symcollab/cryptosolve
du -sh cryptosolve


508K cryptosolve

Using the --sparse flag

The sparse flag makes it so that when we checkout a reference, only the immediate files in the root directory are constructed. Additional commands then need to be issued in order to checkout other directories. With the sparse flag by itself, there isn’t much savings since all the information is still downloaded and stored in the git database.

git clone --sparse https://github.com/symcollab/cryptosolve
du -sh cryptosolve


89M cryptosolve

Conclusion

The power comes from when we combine all these flags together.

git clone \
    --filter=blob:none \
    --sparse \
    --no-checkout \
    https://github.com/symcollab/cryptosolve
du -sh cryptosolve


516K cryptosolve

It’s not different from the --filter=blob:none --no-checkout command initially, but when we checkout a branch we can see that not all the blobs get downloaded.

cd cryptosolve
git checkout main
cd ..
du -sh cryptosolve


644K cryptosolve

You can fetch folders as you please with the following command:

git sparse-checkout add FOLDERNAME

You can even set it so that a specific folder is shown at the root of your directory.

git sparse-checkout set FOLDERNAME