Initializing a project with a submodule fails if directory exists

[abizjak]

Describe the bug

Suppose the template to generate from is in a repository b that has a git submodule a as a dependency.

So the repository b looks like the following (assuming ahas a filefoo`).

├── b
│   └── a
│       └── foo

Now suppose we wish to initialize a project using cargo generate in /root where

/root
    └── a
        └── somefile

that is, root has a non-empty subdirectory called a.

This fails with the error.

Caused by:
    'a' exists and is not an empty directory; class=Invalid (3); code=Exists (-4)

To Reproduce
Steps to reproduce the behavior:

1. cargo generate <WHAT DID YOU RUN>
2. output it generated..
3. See error

Expected behavior

cargo generate succeeds and offers the usual options.

Desktop (please complete the following information):

• OS: Linux
• Version: v0.5.0-453-g51deb01
• Build tools: gcc
• cargo generate --version: 0.17.4

** Root cause **

The root cause of this is the following issue in libgit2 libgit2/libgit2#5830 .
Essentially during the cloning of the submodule there is an incorrect check which checks in the current working directory whether the directory a exists even though that is not relevant.

It is possible to work around the issue in cargo-generate by modifying the clone_with_submodules function to change the working directory

    pub fn clone_with_submodules(self, dest_path: &Path) -> Result<Repository> {                                                                                                                 
        self.clone(dest_path).and_then(|repo| {
            // set the working directory to the root of the newly created repository
            // this works around the issue https://github.com/libgit2/libgit2/issues/5830
            // in libgit2                                                                                                                                                  
            std::env::set_current_dir(dest_path)?;                                                                                                                                               
            for mut sub in repo.submodules()? {                                                                                                                                                  
                sub.update(true, None)?;                                                                                                                                                         
            }                                                                                                                                                                                    
                                                                                                                                                                                                 
            Ok(repo)                                                                                                                                                                             
        })                                                                                                                                                                                       
    }

Seeing that the libgit2 has not resolved the issue in more than a year, would it be acceptable to this project to work around the issue here? I'm more than happy to contribute a PR.

[sassman]

I guess the proposed workaround is fine, but IMO it is important to restore the CWD to it's previous value in order to minimize side effects as much as possible.

Update: feel free and very welcome to open a PR for it.