Re: [PATCH] contrib/subtree: verify HEAD is valid before adding a subtree

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 10, 2025, 16:28 UTC
Message-ID: <xmqq8qqddarl.fsf@gitster.g>
In-Reply-To: <20250210021128.31083-2-xtex@envs.net>

Bingwu Zhang <xtex@envs.net> writes:

Show 12 quoted lines

> From: Bingwu Zhang <xtex@aosc.io>
>
> After initializing a new repository or switching to a orphan branch,
> HEAD is a symbolic reference to refs/heads/xxx while the pointed branch
> head does not exist until a initial commit.
>
> "git subtree add" will try to ensure that working tree and index are
> clean, but as HEAD is invalid, diff-index always fails:
>   fatal: ambiguous argument 'HEAD': unknown revision or path not in the working tree.
>   Use '--' to separate paths from revisions, like this:
>   'git <command> [<revision>...] -- [<file>...]'
>   fatal: working tree has modifications.  Cannot add.

[Disclaimer. I do not use "git subtree" at all myself, and I may be missing the usual expectation of end-users of that command in the following description.]

Good finding. I can see that an unborn HEAD would cause the command fail.

> It says "working tree has modifications" but it is not the case.

I am not sure if "it is not the case" is true, though. If the index is empty (i.e. nothing has been added yet) and the HEAD is unborn, shouldn't that state be considered that working tree has no modifications? IOW, wouldn't this part of the code that uses "diff-index HEAD" want to consider that an unborn HEAD equivalent to an empty tree for the purpose of the comparison?

In short, "is not" -> "may not be", perhaps?

> Add a check using "git show-ref --verify" to ensure that HEAD is a valid
> reference and give a clearer error message.

And if we want to treat an unborn HEAD equivalent to an empty tree, then dying upon seeing "show-ref" fail would not be a good solution to the problem, no? Shouldn't the updated logic to deal with an unborn HEAD be more like "if we see that the HEAD is unborn, then we are happy iff the index is empty; if HEAD already points at a commit, then we are happy iff the working tree has no changes relative to it"?

Show 23 quoted lines

> Signed-off-by: Bingwu Zhang <xtex@aosc.io>
> ---
>  contrib/subtree/git-subtree.sh | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/contrib/subtree/git-subtree.sh b/contrib/subtree/git-subtree.sh
> index 15ae86db1b27..41eb816e454a 100755
> --- a/contrib/subtree/git-subtree.sh
> +++ b/contrib/subtree/git-subtree.sh
> @@ -770,6 +770,11 @@ copy_or_skip () {
>  # Usage: ensure_clean
>  ensure_clean () {
>  	assert test $# = 0
> +	# verify HEAD, or else "git diff-index HEAD" will fail
> +	if ! git show-ref --verify --quiet HEAD 2>&1
> +	then
> +		die "fatal: HEAD is not a valid reference.   Subtree cannot be committed as the first commit of a branch."
> +	fi
>  	if ! git diff-index HEAD --exit-code --quiet 2>&1
>  	then
>  		die "fatal: working tree has modifications.  Cannot add."
>
> base-commit: f93ff170b93a1782659637824b25923245ac9dd1