Re: [PATCH] Preserve cwd in setup_git_directory()

On 7/24/08, Johannes Schindelin [off-list ref] wrote:

Hi,

 On Thu, 24 Jul 2008, Nguyen Thai Ngoc Duy wrote:

 > On 7/24/08, Johannes Schindelin [off-list ref] wrote:
 >

quoted

quoted

 On Thu, 24 Jul 2008, Nguyen Thai Ngoc Duy wrote:

 > >
 > >  > When GIT_DIR is not set, cwd is used to determine where .git is. If
 > >  > core.worktree is set, setup_git_directory() needs to jump back to
 > >  > the original cwd in order to calculate worktree, this leads to
 > >  > incorrect .git location later in setup_work_tree().
 > >
 > > I do not understand.  core.worktree is either absolute, in which case
 > > there is no problem.  Or it is relative to where the config lives, no?
 >
 > The problem is GIT_DIR is not absolute in this case. So cwd must stay
 > where git dir is until it is absolute-ized by setup_work_tree().


I do not see GIT_DIR being set in your test case at all.

 I do not see how get_git_work_tree() ro get_relative_cwd() should ever be
 allowed to chdir().

 _If_ they were (which I strongly doubt), they should chdir() back
 themselves.

 I now wasted easily 30 minutes just trying to make sense of your patch and
 your response.  And I am still puzzled.

 Your commit message was of no help.

Alright, let's look at the code.

 1. cwd is moved to toplevel working directory by setup_git_directory_gently()
 2. setup_git_env() by default will set git_dir variable as ".git" as
part of check_repository_format_gently()
 3. now in setup_git_directory() finds out core.worktree, it chdir()
to get relative prefix
 4. setup_work_tree() sees that git_dir is not absolute path, it makes
git_dir absolute

If in step 3, it does not chdir(), step 4 will be right. In this case,
step 3 does chdir() and not going back, access to git repository will
fail as Geoff Russell discovered.
-- 
Duy