Re: [PATCH] path_treatment: also ignore $GIT_DIR if it's not .git

> On Mon, Dec 2, 2013 at 3:01 PM, Dennis Kaarsemaker
> <dennis@kaarsemaker.net> wrote:
> > On ma, 2013-12-02 at 07:38 +0700, Duy Nguyen wrote:
> >> On Mon, Dec 2, 2013 at 6:38 AM, Dennis Kaarsemaker
> >> <dennis@kaarsemaker.net> wrote:
> >> > On ma, 2013-12-02 at 00:08 +0100, Thomas Rast wrote:
> >> >> Duy Nguyen <pclouds@gmail.com> writes:
> >> >>
> >> >> > On Mon, Dec 2, 2013 at 2:04 AM, Dennis Kaarsemaker
> >> >> > <dennis@kaarsemaker.net> wrote:
> >> >> >> We always ignore anything named .git, but we should also ignore the git
> >> >> >> directory if the user overrides it by setting $GIT_DIR
> >> >> [...]
> >> >> >> +       if (simplify_away(path->buf, path->len, simplify) || is_git_directory(path->buf))
> >> >> >>                 return path_none;
> >> >> >
> >> >> > this adds 2 access, 1 lstat, 1 open, 1 read, 1 close to _every_ path
> >> >> > we check. Is it worth the cost?
> >> >>
> >> >> Moreover it is a much more inclusive check than what the commit message
> >> >> claims: it will ignore anything that looks like a .git directory,
> >> >> regardless of the name.  In particular GIT_DIR doesn't have anything to
> >> >> do with it.
> >> >
> >> > Ah, yes thanks, that's rather incorrect indeed. How about the following
> >> > instead? Passes all tests, including the new one.
> >> >
> >> > --- a/dir.c
> >> > +++ b/dir.c
> >> > @@ -1198,7 +1198,7 @@ static enum path_treatment treat_path(struct dir_struct *dir,
> >> >                 return path_none;
> >> >         strbuf_setlen(path, baselen);
> >> >         strbuf_addstr(path, de->d_name);
> >> > -       if (simplify_away(path->buf, path->len, simplify))
> >> > +       if (simplify_away(path->buf, path->len, simplify) || !strncmp(get_git_dir(), path->buf, path->len))
> >> >                 return path_none;
> >>
> >> get_git_dir() may return a relative or absolute path, depending on
> >> GIT_DIR/GIT_WORK_TREE. path->buf is always relative. You'll pass one
> >> case with this (relative vs relative) and fail another. It might be
> >> simpler to just add get_git_dir(), after converting to relative path
> >> and check if it's in worktree, to the exclude list and let the current
> >> exclude mechanism handle it.
> >
> > This type of invocation really only works from the root of the workdir
> > anyway and both a relative and absolute path work just fine:
> >
> > dennis@lightning:~/code/git$ GIT_DIR=$(pwd)/.foo ./git status
> > On branch master
> > nothing to commit, working directory clean
> > dennis@lightning:~/code/git$ GIT_DIR=./.foo ./git status
> > On branch master
> > nothing to commit, working directory clean
> >
> > Well, unless you set GIT_WORK_TREE as well, but then it still works:
> >
> > dennis@lightning:~/code/git/t$ GIT_DIR=$(pwd)/../.foo GIT_WORK_TREE=.. ../git status
> > On branch master
> > nothing to commit, working directory clean
> > dennis@lightning:~/code/git/t$ GIT_DIR=../.foo GIT_WORK_TREE=.. ../git status
> > On branch master
> > nothing to commit, working directory clean
> >
> > So I'm wondering when you think this will fail. Because then I can add a
> > test for that case too.
> 
> ~/w/git $ cd t
> ~/w/git/t $ GIT_TRACE_SETUP=1 ../git --git-dir=../.git --work-tree=..
> --no-pager status
> setup: git_dir: /home/pclouds/w/git/.git
> setup: worktree: /home/pclouds/w/git
> setup: cwd: /home/pclouds/w/git
> setup: prefix: t/
> On branch exclude-pathspec
> Your branch and 'origin/master' have diverged,
> and have 2 and 5 different commits each, respectively.
> 
> I can't say this is the only case though. One has to audit to all
> possible setup cases in setup_git_directory() to make that claim.