Re: GIT_CEILING_DIRECTORY

Barry Warsaw [off-list ref] writes:

I put my home directory under git (recently converted from bzr), but since I
have some subdirectories under $HOME that are not under git (and some that
are) I want to stop e.g. `git status` from traversing up into $HOME.

Let me understand the use case.  You have $HOME/.git that governs
everything under $HOME, but there are parts of $HOME/, such as
$HOME/projects/*, that will never be controled by $HOME/.git?

Two obvious reactions are:

 - hopefully $HOME/.gitignore covers these non-git parts by having
   entries like '/projects/'; this should not affect the behaviour
   of CEILING though.

 - typing "git status" inside $HOME/projects/ does not make much
   sense in the first place.

I _think_ the "are we in a Git-managed working tree and if so, then
where is the .git directory?" discovery works like this:

    - Are we sitting inside a subdirectory of one of the CEILING
      list elements?  For the purpose of this determination,
      directory 'foo' is not considered a subdirectory of 'foo'
      itself.  If we are, remember where the closest CEILING is.

    - Set the "directory we are checking" to the current directory.

    - Iterate:

      - Does the "directory we are checking" look like the root of a
        working tree managed by Git?  I.e. has ".git" directly in
        it, etc.  If so, we found the Git-managed working tree and
        its ".git".  Return.

      - Truncate one level from "directory we are checking",
        i.e. chdir(..);

      - Are we at a filesystem boundary (unless an environment tells
        us otherwise), or have we hit the closest CEILING we
        determined earlier?  We are not allowed to check if we are
        in a Git-managed working tree at higher level than this
        level.  Return.

      - Otherwise, keep checking.

So setting $HOME/projects as one of the elements in the CEILING list
would not stop us going up if you are actually at $HOME/projects,
but we would stop if you started from $HOME/projects/python.

This somehow sounds a bit inconsistent to me, but I say "a bit
inconsistent" because "Why do we give different answer to 'is
$HOME/projects/python governed by $HOME/.git?' depending on where we
start the discovery process?" is a non-argument (i.e. that is not
the question CEILING is answering).

I have a feeling that we must have done that for a reason.  It may
be interesting to see what breaks in t1504 if the above logic is
updated to stop when you start at a CEILING (unlike the current code
where it stops only when you start below a CEILING).

On Mar 09, 2016, at 09:29 AM, Junio C Hamano wrote:

Let me understand the use case.  You have $HOME/.git that governs
everything under $HOME, but there are parts of $HOME/, such as
$HOME/projects/*, that will never be controled by $HOME/.git?

Correct.

Two obvious reactions are:

- hopefully $HOME/.gitignore covers these non-git parts by having
  entries like '/projects/'; this should not affect the behaviour
  of CEILING though.

Correct.  In this case, $HOME/.gitignore has `projects` so `git status`
etc. in $HOME does the right thing.

- typing "git status" inside $HOME/projects/ does not make much
  sense in the first place.

True, and normally I wouldn't do this explicitly, but it comes up because I
have a bash prompt that shows me the status of the current directory
($GIT_PS_*) so even when I just cd to ~/projects I see status for $HOME.

I _think_ the "are we in a Git-managed working tree and if so, then
where is the .git directory?" discovery works like this:

[...]

So setting $HOME/projects as one of the elements in the CEILING list
would not stop us going up if you are actually at $HOME/projects,
but we would stop if you started from $HOME/projects/python.

And indeed, that works great.

This somehow sounds a bit inconsistent to me, but I say "a bit
inconsistent" because "Why do we give different answer to 'is
$HOME/projects/python governed by $HOME/.git?' depending on where we
start the discovery process?" is a non-argument (i.e. that is not
the question CEILING is answering).

I have a feeling that we must have done that for a reason.  It may
be interesting to see what breaks in t1504 if the above logic is
updated to stop when you start at a CEILING (unlike the current code
where it stops only when you start below a CEILING).

That would be interesting; it seems like it would solve my use case.

Cheers,
-Barry