Re: Strangeness with git-add and nested repositories

From: Junio C Hamano <gitster@pobox.com>
Date: Apr 28, 2016, 16:48 UTC
Message-ID: <xmqqtwil65nx.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <xmqq1t5p7kmp.fsf@gitster.mtv.corp.google.com>

Junio C Hamano <gitster@pobox.com> writes:

>  * "git add -- path/to/subdir/file", when any of path/, path/to/,
>    path/to/subdir/ is a Git repository that is different from the
>    current Git repository, must fail.

If any of the leading directories in that long path is actually a git repository that is different from the current one, it shouldn't have added it. IOW, adding the path is a bug.

"git add A/B/C" usually enforces this rule by

 - check A; if A is a separate repository (A/.git exists, etc.),
   do not add this path;

   - check A/B; if A/B is a separate repository, ignore;

     - check A/B/C; if A/B/C is a separate repository, ignore.
       otherwise add it.

When given multiple paths, e.g. "git add A/B/C A/D", it tries to "optimize" things by first finding common leading directory (in this case "A/") and doing something slightly different, and I think the bug Andrew saw lies in that codepath. It is likely that the code is forgetting to make sure that there is no top of enclosed working tree in the common leading directory part of the path.