Re: git rev-parse --show-toplevel inside `.git` returns 0 and prints nothing

[Anthony Sottile <asottile@umich.edu>]

On Mon, Nov 18, 2019 at 7:33 PM Jeff King <peff@peff.net> wrote:
>
> On Tue, Nov 19, 2019 at 11:52:54AM +0900, Junio C Hamano wrote:
>
> > If I were designing the feature today, with today's rest-of-git in
> > mind, I would say
> >
> >  - In a bare repository, exit with non-zero status after giving an
> >    error message "no working tree".
> >
> >  - In a repository that has a single associated working tree, show
> >    the path to the top-level of that working tree and exit with zero
> >    status.
>
> Do you mean to do this even in when the cwd is inside .git?
>
> I think that's confusing, because you don't actually have a working tree
> at all. E.g.:
>
>   $ git rev-parse --show-toplevel
>   /home/peff/tmp
>   $ git status -b --short
>   ## No commits yet on master
>
>   $ cd .git
>   $ git rev-parse --show-toplevel
>   $ git status -b --short
>   fatal: this operation must be run in a work tree
>
> So internal commands like status accept that we have no working tree in
> this situation. But "--show-toplevel" just prints nothing. I'd amend
> your second point to be "If we are in the working tree of a repository,
> show the path to the top-level of that working tree and exit with zero
> status".
>
> And then that leaves another case: we are not in the working tree of the
> repository. In which case I think it should be the same as the bare
> repository.
>
> And from that, your multi-working-tree case falls out naturally:
>
> > In a repository that has more than one working trees (which is one
> > of the things "todasy's rest-of-git" has that did not exist back
> > when --show-prefix/--show-toplevel etc. were invented), then what?
> > Would it make sense to show the primary working tree?  What if the
> > worktree(s) were made off of a bare repository, in which case nobody
> > is the primary?
>
> There may be multiple working trees, but we can only be in one of them
> at a time. So that's the one that we show.
>
> And the only real change here is that "--show-toplevel" prints an error
> and exits non-zero when we won't have a working tree. Something like:
>
> diff --git a/builtin/rev-parse.c b/builtin/rev-parse.c
> index 3857fd1b8a..81161f2dfb 100644
> --- a/builtin/rev-parse.c
> +++ b/builtin/rev-parse.c
> @@ -805,6 +805,8 @@ int cmd_rev_parse(int argc, const char **argv, const char *prefix)
>                                 const char *work_tree = get_git_work_tree();
>                                 if (work_tree)
>                                         puts(work_tree);
> +                               else
> +                                       die("this operation must be run in a work tree");
>                                 continue;
>                         }
>                         if (!strcmp(arg, "--show-superproject-working-tree")) {
>
>
> I think the reason this hasn't come up until now is callers are expected
> to use require_work_tree() or "rev-parse --is-inside-work-tree" first.
>
> It would probably make sense for the rev-parse documentation to also
> clarify what "the top-level directory" is.
>
> -Peff

I realize I forgot to include the X to my Y :) -- this was a totally
silly case that I got as a bug report:
https://github.com/pre-commit/pre-commit/issues/1219

I *expected* an error case but didn't get one

Anthony