dvc get from subdir

[dmpetrov]

$ dvc -V
0.70.0+38be14

Regular pull works just fine for subdirs:

$ git clone https://github.com/dmpetrov/dataset
$ cd dataset
$ dvc pull dir1.dvc
$ ls dir1
file1 file2

dvc get does not work:

$ dvc get https://github.com/dmpetrov/dataset dir1/file1
ERROR: failed to get 'dir1/file1' from 'https://github.com/dmpetrov/dataset' - Output 'dir1/file1' not found in target repository 'https://github.com/dmpetrov/dataset'


Having any troubles? Hit us up at https://dvc.org/support, we are always happy to help!

This is quite a basic scenario - high priority bug.

• Please make sure it covered by unit tests

EDITED: It worked before - regression.

[dmpetrov]

It seems like it is not a regression. It fails only on data dirs, not in data files in a regular dir (which worked):

$ dvc get https://github.com/dmpetrov/dataset mydir/myfile

I'm removing high-pri tag.

[shcheklein]

@dmpetrov is dataset public to reproduce? It looks like I don't have access to it.

[shcheklein]

If I get it right.

Yes, I think it was never implemented. DVC in general never provided file-level granularity for directories. E.g. you can't dvc pull dir1/file1 you have to pull the whole dir1. More or less for the same reason dvc get operates with the same assumptions - directory is a single manageable entity, you can't split it.

So, I would consider this as a feature request. That makes sense to me, even though would be the first time we go to intra-directory level of data management with DVC.

Initial implementation would be inefficient due to the reasons I described - we don't have a mean to fetch into cache a single file for a directory.

Also, should we support subdirectories? wildcards? any other considerations?

[efiop]

Related #2458

[efiop]

Closing as a duplicate of #2458