Re: [PATCH v9] setup: improve error diagnosis for invalid .git files

From: Tian Yuchen <a3205153416@gmail.com>
Date: Feb 22, 2026, 10:28 UTC
Message-ID: <9d79bff0-eb97-4332-8260-236761c9c3da@gmail.com>
In-Reply-To: <xmqqseatqqpr.fsf@gitster.g>

Hi Junio,

On 2/22/26 13:42, Junio C Hamano wrote:

> The changes to these two functions require us to audit callers of
> them that are outside the call graph of the main focus of this
> patch.  For example, we see the following code in submodule.c:

Oops, I was so focused on setup.c that I completely overlooked the external callers of the API. Thank you for pointing out!

I ran git grep for READ_GITFILE_ERR_STAT_FAILED and READ_GITFILE_ERR_NOT_A_FILE across the tree and audited the hits. Besides the expected changes in setup.c, here is what I found and fixed:

  - In submodule.c, exactly as you pointed out.
  - In worktree.c there were two places relying on NOT_A_FILE to print 
the specific ".git is not a file" error. I added 
READ_GITFILE_ERR_IS_A_DIR to those conditions to prevent the error from 
degrading to the generic ".git file broken".

I have also added the NEEDSWORK comment in the switch statedment, as you mentioned earlier.

Thanks,

Yuchen