Failed to cargo package --allow-dirty for repo with no commit

[weihanglo]

Problem

#13960 unconditionally checks VCS status when doing cargo package.
However, a VCS (Git specifically) repository could have no commit history, and then it fails cargo package with such an error:

cargo +beta package --allow-dirty
warning: manifest has no description, license, license-file, documentation, homepage or repository.
See https://doc.rust-lang.org/cargo/reference/manifest.html#package-metadata for more info.
error: revspec 'HEAD' not found; class=Reference (4); code=NotFound (-3)

Steps

1. cargo new foo
2. cd foo
3. cargo package --allow-dirty

Possible Solution(s)

The current .cargo_vcs_info.json format looks like:

{
 "git": {
   "sha1": "aac20b6e7e543e6dd4118b246c77225e3a3a1302",
   "dirty": true
 },
 "path_in_vcs": ""
}

If the HEAD hash cannot be determined, should we instead emit a JSON like

{
 "git": {
   "dirty": true
 },
 "path_in_vcs": ""
}

or just don't generate .cargo_vcs_info.json at all?

Notes

No response

Version

cargo 1.81.0-beta.2 (a2b58c3da 2024-07-16)
release: 1.81.0-beta.2
commit-hash: a2b58c3dad4d554ba01ed6c45c41ff85390560f2
commit-date: 2024-07-16

[linyihai]

@rustbot claim

[weihanglo]

Here are some possible solutions:

1. Don't include .cargo_vcs_info.json at all
   • This maintains the old behavior without blocking cargo package.
2. Don't include the git.sha key.
   • The git.dirty is still here to tell it is a dirty Git repo without a commit.
3. Don't include the entire git object.
   • A Git repository without any commit is kinda useless so stripping it off.
4. Treat it as a non-Git package.
   • Non-git package is assumed clean and will not include .cargo_vcs_info.json.

Any other solutions?

Mind that this is a bit time-bound as it breaks the stable usage of Cargo.

[epage]

Don't include .cargo_vcs_info.json at all

I lean towards this seems the least controversial starting point, especially if this is time bound.

Mind that this is a bit time-bound as it breaks the stable usage of Cargo.

While this is a stable-to-beta regression, I'm assuming the end-user impact is pretty minimal. I can't see it being terribly common (and most likely a mistake) to publish from an empty repo.

[weihanglo]

Agree on least controversial and minimal impact parts. I'll go ahead and merge #14359.