Flexible version control for files and folders.

The simplest way is to get it from PyPi:

pip install bev

ls
# image.png ids.json some-folder
bev add image.png
ls
# image.png.hash ids.json some-folder
bev add ids.json some-folder
ls
# image.png.hash ids.json.hash some-folder.hash

git add image.png.hash ids.json.hash some-folder.hash
git commit -m "added new files"

ls
# image.png.hash ids.json.hash some-folder.hash
bev pull image.png.hash --mode copy
ls
# image.png ids.json.hash some-folder.hash
bev pull some-folder.hash --mode copy
ls
# image.png ids.json.hash some-folder

In this recipe we "expand" the hashed folder and fill it with the hashes of the files it contains. This is much faster than copying back the entire folder.

ls
# image.png.hash ids.json.hash some-folder.hash
bev pull some-folder.hash --mode hash
ls
# image.png.hash ids.json.hash some-folder
ls some-folder
# photo.jpg.hash some-text-file.txt.hash nested-folder

Afterwards you can add the folder back

bev add some-folder
ls
# image.png.hash ids.json.hash some-folder.hash

1. Choose a folder for your repository and create a basic config (.bev.yml):

main:
  storage: /path/to/storage/folder

meta:
  hash: sha256

2. Run init

bev init

3. Add files to bev

bev add /path/to/some/file.json
# also can provide several paths
bev add /path/to/some/folder/ /path/to/some/image.png

4. ... and to git

git add file.json.hash folder.hash image.png.hash
git commit -m "added files"

5. Access the files from python

import imageio
from bev import Repository

# `version` can be a commit hash or a git tag 
repo = Repository('/path/to/repo', version='8a7fe6')
image = imageio.imread(repo.resolve('image.png'))

6. Or from cli

# replace the folder's hash by the hashes of its files
bev pull folder.hash --mode hash
# entirely restore the folder (inverse of `bev add folder`)
bev pull folder.hash --mode copy
# same for files
bev pull image.png.hash --mode copy

Here are some tutorials that cover more advanced configuration, including multiple storage locations and machines:

1. Create a repository - needed only at first time setup
2. Adding files
3. Accessing files

DVC is a great project, and we took inspiration from it while designing bev. However, out lab has several requirements that DVC doesn't meet:

1. Our data caches are spread across multiple HDDs - we need support for multiple cache locations
2. We have multiple machines, and each of them has a different storage configuration: locations, number of HDDs, their volumes - we need a flexible way of choosing the right config depending on the machine
3. Often we simultaneously conduct experiments on different versions of the same data - we need easy access to multiple version of the same data
4. The need for dvc checkout after git checkout is error-prone, because it can lead to situations when the data is not consistent with the current commit - we need a more constrained relation between data and git

bev supports all four out of the box!

However, if these requirements are not essential to your project, you may want to stick with DVC - its community and tests coverage is much larger.