Re: [RFC GSoC PATCH v3 0/2] Make read_attr() repository-aware by introducing a lazy bare state

From: Phillip Wood <phillip.wood123@gmail.com>
Date: Feb 19, 2026, 10:42 UTC
Message-ID: <0d6c5b14-1c53-4372-8395-7c7cba9fbf0b@gmail.com>
In-Reply-To: <20260208062949.596-1-kumarayushjha123@gmail.com>

Hi Ayush

On 08/02/2026 06:29, Ayush Jha wrote:

Show 8 quoted lines

> This patch series is v3 of an earlier RFC that aimed to remove the
> dependency on the global `the_repository` from read_attr().
> 
> Based on feedback from Lucas Seiki Oshiro and Junio C Hamano, this
> iteration introduces a repository-scoped, lazily initialized "bare"
> state in struct repo_settings, exposed via repo_settings_get_is_bare().
> The read_attr() function is then updated to use this helper instead of
> is_bare_repository().

Doesn't the lazy initialization here suffer from the same problems that Junio explained in https://lore.kernel.org/xmqqpl63b2tm.fsf@gitster.g ? Changing config settings that are currently parsed eagerly to being parsed lazily opens up a whole can of worms and is best avoided. As Junio said Bello Olamide is exploring a different approach that avoids the lazy parsing.

Thanks

Phillip

Show 15 quoted lines

> The series is split as follows:
>    1/2 repo-settings: add repo_settings_get_is_bare
>    2/2 attr: use local repository state in read_attr
> 
> Comments and suggestions are welcome.
> 
> Ayush Jha (2):
>    repo-settings: add repo_settings_get_is_bare
>    attr: use local repository state in read_attr
> 
>   attr.c          |  3 ++-
>   repo-settings.c | 12 ++++++++++++
>   repo-settings.h |  5 +++++
>   3 files changed, 19 insertions(+), 1 deletion(-)
>