index.add('.') adds even .git directory into the index, and adds files also with ./ prefix

[yarikoptic]

Compare how stock git does it (related: there was this noise about security issue of having case-insensitive filesystems and .Git file in a repo):

$> rm -rf /tmp/123; mkdir /tmp/123; cd /tmp/123; git init; echo 1 > 1; git add 1; git status                                                                      
Initialized empty Git repository in /tmp/123/.git/
On branch master

Initial commit

Changes to be committed:
  (use "git rm --cached <file>..." to unstage)

    new file:   1

and how GitPython (as of 1.0.1-74-gc8b8379 as in Debian):

$> rm -rf /tmp/123; mkdir /tmp/123; cd /tmp/123; git init; echo 1 > 1; python -c 'import git; r=git.Repo("/tmp/123"); r.index.add(".")'; git status
Initialized empty Git repository in /tmp/123/.git/
On branch master

Initial commit

Changes to be committed:
  (use "git rm --cached <file>..." to unstage)

    new file:   ./.git/HEAD
    new file:   ./.git/config
    new file:   ./.git/description
    new file:   ./.git/hooks/applypatch-msg.sample
    new file:   ./.git/hooks/commit-msg.sample
    new file:   ./.git/hooks/post-update.sample
    new file:   ./.git/hooks/pre-applypatch.sample
    new file:   ./.git/hooks/pre-commit.sample
    new file:   ./.git/hooks/pre-push.sample
    new file:   ./.git/hooks/pre-rebase.sample
    new file:   ./.git/hooks/prepare-commit-msg.sample
    new file:   ./.git/hooks/update.sample
    new file:   ./.git/info/exclude
    new file:   ./.git/objects/49/8b267a8c7812490d6479839c5577eaaec79d62
    new file:   ./.git/objects/51/5f4836297fdf7567c066983c16e5eff598f7bd
    new file:   ./.git/objects/cb/089cd89a7d7686d284d8761201649346b5aa1c
    new file:   ./.git/objects/d0/0491fd7e5bb6fa28c517a0bb32b8b506539d4d
    new file:   ./1

Untracked files:
  (use "git add <file>..." to include in what will be committed)

    1

not sure why git says that 1 is untracked but ./1 was added... I guess also GitPython intricacy

[Byron]

Thanks for letting me know !

As I see the IndexFile implementation, it's definitely located in the plumbing area within the git ecosystem. Therefore I would expect it to do exactly as you say, which is what it does in this case as well.

The actual bug here seems to be left-over untracked file. I could imagine that for some reason, it added the ./ prefix to the filename into the index, which should in fact just be 1, not ./1.

In my own tests, I get a different, but also incorrect result:

Changes to be committed:
  (use "git rm --cached <file>..." to unstage)

    new file:   ./.git/HEAD
    new file:   ./.git/config
    new file:   ./.git/description
    new file:   ./.git/hooks/applypatch-msg.sample
        [...]
    new file:   ./.git/index
    new file:   ./.git/info/exclude
    new file:   ./.git/objects/41/42082bcb939bbc17985a69ba748491ac6b62a5
        [...]
    new file:   ./1
    new file:   1

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git checkout -- <file>..." to discard changes in working directory)

    modified:   ./.git/index

The latter should be fixed !

Meta

• git-python@7b85487
• git version 2.5.4 (Apple Git-61)

[yarikoptic]

@Byron Thanks for looking into it. what is the proper way to stage content '.' for a commit then?

[Byron]

I think you would have to filter this path out by yourself. It is valid to pass in iterators, and a filter could look like (p for p in paths if not is_git_dir(p)).
Of course, this is something people would probably get wrong the first time, as it is somewhat unexpected, yet I believe plumbing shouldn't impose any limits, even if that means it's a loaded footgun.

[rgeorgeoff]

This is still a thing v.v

[kakulukia]

Just ran into the same issue. Why is the .git folder pushed into the index? Thats totally weird!

[kakulukia]

okay there is some kind of workaround - thank god - but its still a very weird bug!
#292

[Byron]

Just for completeness, here is the link to the comment describing the workaround.