From: Justin Tobler Date: Wed, 10 Jun 2026 22:22:04 GMT Subject: Re: [PATCH 5/7] environment: split up concerns of `is_bare_repository_cfg` Message-ID: In-Reply-To: <20260610-b4-pks-setup-drop-global-state-v1-5-5dff3eec8f06@pks.im> On 26/06/10 08:56AM, Patrick Steinhardt wrote: > The `is_bare_repository_cfg` variable tracks two different pieces of > information: > > - It tracks whether the user has invoked git with the "--bare" flag, > which makes us treat any discovered Git repository as if it was a > bare repository. > > - Otherwise it tracks whether the discovered `the_repository` is bare. > > This makes the flag extremely confusing and creates a bit of a challenge > when handling multiple repositories in the same process. Indeed, this does seem rather confusing. > Split up the concerns of this variable into two pieces: > > - `startup_info.force_bare_repository` tracks whether the user has > passed the "--bare" flag. This is used as a hint to treat newly set > up repositories as bare regardless of whether or not they have a > worktree. > > - `struct repository::bare_cfg` tracks whether or not a repository is > considered bare. This takes into account both whether the user has > passed "--bare" and the discovered state of the repository itself. > > Whether or not a repository is bare is now resolved when checking the > repository's format, and is then later applied to the repository itself > via `apply_repository_format()`. Nice. These seem like they should be tracked separately, so splitting them up sounds like a good idea. > This enables a subsequent change where we make `is_bare_repository()` > not depend on global state anymore. > > Signed-off-by: Patrick Steinhardt > --- > builtin/init-db.c | 2 +- > environment.c | 5 ++--- > environment.h | 1 - > git.c | 2 +- > repository.c | 1 + > repository.h | 7 +++++++ > setup.c | 21 ++++++++++++++------- > setup.h | 6 ++++++ > worktree.c | 2 +- > 9 files changed, 33 insertions(+), 14 deletions(-) > > diff --git a/builtin/init-db.c b/builtin/init-db.c > index 52aa92fb0a..566732c9f4 100644 > --- a/builtin/init-db.c > +++ b/builtin/init-db.c > @@ -81,7 +81,7 @@ int cmd_init_db(int argc, > const char *template_dir = NULL; > char *template_dir_to_free = NULL; > unsigned int flags = 0; > - int bare = is_bare_repository_cfg; > + int bare = startup_info->force_bare_repository ? 1 : -1; Any particular reason to continue mapping `force_bare_repository=false` to -1? Or was this to just minimize changes? > const char *object_format = NULL; > const char *ref_format = NULL; > const char *initial_branch = NULL; [snip] > diff --git a/repository.h b/repository.h > index 36e2db2633..7d649e32e7 100644 > --- a/repository.h > +++ b/repository.h > @@ -117,6 +117,13 @@ struct repository { > bool worktree_initialized; > bool worktree_config_is_bogus; > > + /* > + * Whether the repository is bare, as set by "core.bare" config or > + * inferred during repository discovery. -1 means unset/unknown, 0 > + * means non-bare, 1 means bare. > + */ > + int bare_cfg; Now we track whether a repository is bare in `struct repository` and removes the need to a separate global to track this state. > + > /* > * Path from the root of the top-level superproject down to this > * repository. This is only non-NULL if the repository is initialized > diff --git a/setup.c b/setup.c > index 71fc6b33da..2b690da8ca 100644 > --- a/setup.c > +++ b/setup.c > @@ -795,10 +795,16 @@ static int check_repository_format_gently(const char *gitdir, > has_common = 0; > } > > - if (!has_common) { > - if (candidate->is_bare != -1) > - is_bare_repository_cfg = candidate->is_bare; > - } else { > + if (startup_info->force_bare_repository) { > + candidate->is_bare = 1; > + FREE_AND_NULL(candidate->work_tree); > + } else if (has_common) { > + /* > + * When sharing a common dir with another repository (e.g. a > + * linked worktree), do not let this repository's config > + * dictate bareness; it is inherited from the main worktree. > + */ > + candidate->is_bare = -1; > FREE_AND_NULL(candidate->work_tree); Previously, when there was a common dir, `candidate->work_tree` was left untouched, but now we are expclicitly setting it. I'm not sure I fully understand this change. -Justin