Re: [PATCH] Use correct default remote for fetching in submodule.

From: Eric Sunshine <sunshine@sunshineco.com>
Date: Feb 13, 2023, 13:45 UTC
Message-ID: <CAPig+cRotsJZHnRMUqWaYz=nw1sBRsW5ms7WWuApviG2be2GTQ@mail.gmail.com>
In-Reply-To: <623b0b8a-a5b3-408c-b924-9f88d9763b0e@app.fastmail.com>

On Mon, Feb 13, 2023 at 8:31 AM Mara <mara@marabos.nl> wrote:

Show 9 quoted lines

> "git submodule update" first tries a regular "git fetch"
> to fetch the commit, but when that doesn't retrieve the commit
> it wants, it tries "git fetch <remote> <commit>".
> For <remote>, it used the wrong default remote: the default
> remote of the outer repository, rather than the default remote
> of the submodule.
>
> Signed-off-by: Mara Bos <mara@marabos.nl>
> ---

I'm not a submodule user and I don't have any particular familiarity with this code, so I may be wrong, but...

Show 12 quoted lines

> diff --git a/builtin/submodule--helper.c b/builtin/submodule--helper.c
> index 4c173d8b37..50b96e0b9d 100644
> --- a/builtin/submodule--helper.c
> +++ b/builtin/submodule--helper.c
> @@ -2225,7 +2225,10 @@ static int fetch_in_submodule(const char *module_path, int depth, int quiet,
> -               char *remote = get_default_remote();
> +               char *remote;
> +               int code = get_default_remote_submodule(module_path, &remote);
> +               if (code)
> +                       return code;
>                 strvec_pushl(&cp.args, remote, hex, NULL);
>                 free(remote);

... it looks like this change may introduce a memory leak. Digging down through get_default_remote_submodule() and the functions it calls, it appears that repo_get_default_remote() can return a non-zero code _after_ it has allocated memory for `remote`. If I'm reading this correctly, then the above should probably be:

    char *remote = NULL;
    int code = get_default_remote_submodule(module_path, &remote);
    if (code) {
        free(remote);
        return code;
    }

Also, if possible, add a new test, perhaps to t/t7406-submodule-update.sh, demonstrating that this change fixes the problem and to ensure that it doesn't get broken again. If you have a minimal-reproduction recipe which exhibits the problem, then you may be able to turn it into an actual test.