Re: [PATCH] ls-remote: add --show-object-format

[Patrick Steinhardt <ps@pks.im>]

On Thu, Nov 05, 2026 at 12:45:41PM +0200, Adrian Ratiu wrote:
> While working on JJ sha256 remote support (it uses Git for networking),
> I noticed there is no way to get the object format of a repo without
> cloning it. [1]
> 
> Git clone learns the remote format during the protocol handshake, but
> it is not shown anywhere else. For example, when creating an empty repo
> and running "git fetch":
> 
> 1. If the remote has references, fetch can fail with a mismatch error
> and "git ls-remote" shows the format indirectly through the length
> of the object IDs. Not ideal, but the it can be figured out eventually.

s/the it/then it/

Also, this is a somewhat imperfect way to detect the hash algorithm as
it assumes that we don't ever add a third hash that has the same length
as either of the existing ones. So it works now, but it may not work in
the future anymore.

> 2. If the remote is empty, "git ls-remote" prints nothing and fetch
> succeeds, so the new local repo can silently get the wrong format,
> which only shows up later, on the first push.
> 
> To fill this gap, add a ls-remote --show-object-format option which
> prints the object format learned from the remote on a line of its own,
> before the references.

Seems sensible.

> diff --git a/Documentation/git-ls-remote.adoc b/Documentation/git-ls-remote.adoc
> index d71c4ab3e2..1abf7b60b7 100644
> --- a/Documentation/git-ls-remote.adoc
> +++ b/Documentation/git-ls-remote.adoc
> @@ -64,6 +65,13 @@ OPTIONS
>  	upload-pack only shows the symref HEAD, so it will be the only
>  	one shown by ls-remote.
>  
> +--show-object-format::
> +	Show the object format (hash algorithm, e.g. `sha1` or `sha256`) used
> +	by the remote repository before printing the references, on a line of
> +	its own. This also works for a remote repo without any refs, which is
> +	useful to find out which object format to use for a local repository,

A bit curious that you first say "repo" and then "repository". I'd just
drop "repo".

> diff --git a/builtin/ls-remote.c b/builtin/ls-remote.c
> index fe77829557..6d961f9961 100644
> --- a/builtin/ls-remote.c
> +++ b/builtin/ls-remote.c
> @@ -150,6 +154,8 @@ int cmd_ls_remote(int argc,
>  
>  	if (!dest && !quiet)
>  		fprintf(stderr, "From %s\n", remote->url.v[0]);
> +	if (show_object_format)
> +		printf("%s\n", transport_get_hash_algo(transport)->name);

Should we maybe make this a tiny bit more extensible by not printing the
object format without any kind of prefix? We could for example instead
print "object-format: %s", which would allow us to add more fields in
the future without the need to make parsers be order-dependent.

Also, one more question I had is whether we know that the hash algo will
always be populated here. I would assume so, as we have this logic a
couple lines further up:

	ref = transport_get_remote_refs(transport, &transport_options);
	if (ref) {
		int hash_algo = hash_algo_by_ptr(transport_get_hash_algo(transport));
		repo_set_hash_algo(the_repository, hash_algo);
	}

But in theory, this logic would also work alright if the hash was NULL.

> diff --git a/t/t5512-ls-remote.sh b/t/t5512-ls-remote.sh
> index 2e1d4f7586..7e82154fe4 100755
> --- a/t/t5512-ls-remote.sh
> +++ b/t/t5512-ls-remote.sh
> @@ -332,6 +332,52 @@ test_expect_success 'indicate no refs in v0 standards-compliant empty remote' '
>  	test_expect_code 2 git ls-remote --exit-code --upload-pack=./cat-input .
>  '
>  
> +# Run "ls-remote --show-object-format" with the given arguments over
> +# protocol v0 and v2, and compare its output with "expect".
> +test_show_object_format () {
> +	for version in 0 2
> +	do
> +		git -c protocol.version=$version \
> +			ls-remote --show-object-format "$@" >actual &&
> +		test_cmp expect actual || return 1

It's a bit weird that we expect the caller to already have written
`expect`. A better interface would be to have them call the expected
object format as another parameter.

> +	done
> +}
> +
> +test_expect_success 'ls-remote --show-object-format with empty sha1 remote' '
> +	git init --bare --object-format=sha1 empty-sha1.git &&
> +	echo sha1 >expect &&
> +	test_show_object_format empty-sha1.git
> +'
> +
> +test_expect_success 'ls-remote --show-object-format with empty sha256 remote' '
> +	git init --bare --object-format=sha256 empty-sha256.git &&
> +	echo sha256 >expect &&
> +	test_show_object_format empty-sha256.git
> +'

We could easily deduplicate these tests by looping around `for hash in
sha1 sha256`.

Thanks!

Patrick