dvc: support .dvcignore

[efiop]

Same as .gitignore, but for dvc. It will tell dvc which paths to ignore when caching data. Would be extremely useful for ignoring auto-generated/temporary files/directories that are created as a side effect and don't carry anything useful.

First iteration should make patterns listed in dvcignore be ignored on Repo.stages() when collecting stages.

echo '/directory_with_millions_of_files' > .dvcignore
dvc status # should not enter `directory_with_millons_of_files`

Second iteration should support dvc add/run. We have a separate issue for it at #1876

[dmpetrov]

@efiop could you please clarify a bit more? Is it for users mostly or for internal DVC usage? Should we include .dvc/ into .dvcignore in the case of internal usage?

[efiop]

It is for users mostly as a convenient way to ignore some paths/files/dirs from being cached/tracked by dvc. Good point about .dvc/ though, we will act as if it is in .dvcignore by default :)

[ghost]

Is this somehow related #1471 ?

I saw on the chat that the problem was that the MD5 of a directory changed because the file system created a .DS_Store on it and detected it as changed.

I can see the same stuff happening for other things if you are not careful enough (e.g. vim swap files, file locking mechanisms that create dotfiles, IDE specific files / .tags, etc.)

If we are introducing .dvcignore it would also be great to have a global .dvcignore the same way you can have a global .gitignore

[efiop]

@MrOutis Not related to #1471 , it is exclusively about dvc tracking files.

Great idea about a global .dvcignore! I didn't know .gitignore could be global.

[mhham]

I confirm that .DS_Store files are a pain when dvc adding directories on mac Os, and completely agree with the need of a .dvcignore file.

[alguevara7]

I think an equally pressing issue is that having a large (4 million files) un-cached folder slows down dvc, as it needs to traverse the whole folder before executing any command.

Adding support for .dvcignore would add the required capability to address this issue.

[shcheklein]

Please, create a ticket or a page to document the changes.