Dvc.org: cmd-ref: add a note that gc might try to pull missing .dir cache files

[efiop]

https://discordapp.com/channels/485586884165107732/563406153334128681/690239392094486641

[jorgeorpinel]

e.g. for directories if corresponding .dir cache file is missing, then gc is not able to tell which files from that dir are needed, so it tries to download them.

@efiop any other cases in which GC may hit the remote? Thanks

[efiop]

@jorgeorpinel I think that's all.

[imhardikj]

@jorgeorpinel I tried to implement this on example-get-started repo.
For instance after running dvc pull prepare.dvc the content of cache was:

├── 58
│   └── 245acfdc65b519c44e37f7cce12931
├── 68
│   ├── 36f797f3924fb46fcfd6b9f6aa6416.dir
│   └── 36f797f3924fb46fcfd6b9f6aa6416.dir.unpacked
│       ├── test.tsv
│       └── train.tsv
└── 9d
    └── 603888ec04a6e75a560df8678317fb

Then I ran rm -Rf .dvc/cache to delete the cache and used dvc gc -w. It did pull some .dir cache files from remote as dvc gc isn't able to tell which files in directory are missing. This is also mentioned in the discord discussion

├── 42
│   └── c7025fc0edeb174069280d17add2d4.dir
└── 68
     └── 36f797f3924fb46fcfd6b9f6aa6416.dir

Should I create a PR and add a note for this in gc.md?

[jorgeorpinel]

Where did the 42 one come from @imhardikj ?

[imhardikj]

First I deleted the cache directory and then I did dvc gc -w. So I am sure gc fetched 68 as well as 42 from remote.

[jorgeorpinel]

Oh OK you cloned example-get-started to its latest commit, but only pulled prepare.dvc at first right (when you ran tree .dvc/cache)? That would explain it. 42 must be from a later stage.