better error when output is dvcignored

[jorgeorpinel]

UPDATE: Skip to #3538 (comment)

---

$ dvc version
DVC version: 0.90.2
Python version: 3.7.5
Platform: Windows-10-10.0.18362-SP0
Binary: True
Package: exe
Supported remotes: azure, gdrive, gs, hdfs, http, https, s3, ssh, oss
Filesystem type (workspace): ('NTFS', 'C:\\')

Having data in .dvcignore makes DVC commands ignore all file names starting with that string e.g data2, dataset. https://dvc.org/doc/user-guide/dvcignore links to gitignore patterms but Git doesn't do that.

UPDATE: Skip to #3538 (comment)

[efiop]

@jorgeorpinel So having data, right? Git does the same thing, hence why we use /data in .gitignore when we dvc add data.

[jorgeorpinel]

Maybe it behaves differently in distinct OSs? I'm on Windows (Cmder/ Git Bash):

λ git init
λ touch data data2 data3 data4
λ echo data > .gitignore
λ ls
data  data2  data3  data4
λ git status
...
Untracked files:
  (use "git add <file>..." to include in what will be committed)
        .gitignore
        data2
        data3
        data4
λ git add data
The following paths are ignored by one of your .gitignore files:
data
Use -f if you really want to add them.
λ git add .
λ git status
...
        new file:   .gitignore
        new file:   data2
        new file:   data3
        new file:   data4

[efiop]

@jorgeorpinel Oh, I misunderstood your report. Need to check that, indeed. @pared Maybe you have any ideas?

[pared]

Will check this, sorry for late response

[pared]

@jorgeorpinel that was happening on windows right?
following test is passing on my linux machine for 0.90.2:

def test_ignore_data(tmp_dir, dvc):
    tmp_dir.gen({"data": "data", "data1": "data1", "data2": "data2",
                 "data3":"data3", ".dvcignore": "data"})


    files = []
    for r, d, fs in dvc.tree.walk("."):
        files.extend([os.path.join(r, f) for f in fs])

    assert set(files) == set(["./data1", "./data2", "./data3", "./.dvcignore"])

[jorgeorpinel]

Sorry guys. I don't remember which DVC commands I noticed originally that don't respect files in .dvcignore. It wasn't a great report... But I'm QAing this and I think there's definitely some funny stuff happening:

$ echo data > data
$ echo data > .dvcignore
$ dvc add data
100% Add|██...  # Works and puts it in .gitignore. Should ignore? E.g.:

$ git add data
The following paths are ignored by one of your .gitignore files:
data
Use -f if you really want to add them.

I'm on 0.93.0 from exe installer now. Yes, Windows (Git Bash on Cmder)

Also, the first example in https://dvc.org/doc/user-guide/dvcignore#examples doesn't work for me (all files inside data/ dir are added and cached). I didn't check the others.

[pared]

@jorgeorpinel hmm, that does not seem right.
I think we should behave in a similar way as git does, which is informing the user that his output collides with an entry of some .dvcignore file.
This functionality can be implemented probably after we introduce something like git check-ignore for .dvcignore. I think there was an issue for that already, but I am unable to find it.