If you have multiple git repositories that you would like to bundle, for archiving or cataloguing, this program may be for you. Requires Python 3.
./bundle_repos_exec.py INDEXFILE
...where INDEXFILE is a text file containing one repository URL per line, like
https://github.com/octocat/Hello-World.git
https://bitbucket.org/atlassian_tutorial/helloworld.git
https://github.com/Microsoft/api-guidelines.git
Run the program with --help to see execution options. In normal operation,
each repository is cloned with all branches and a corresponding bundle is
created beneath the bundles directory (./repositories by default). The
bundles are organized by host and path.
For each repository URL, if a local bundle already exists, the latest commit
in the fresh clone is compared to the latest commit in the local bundle, and
the bundle is updated only if they differ. This comparison can be skipped with
the --ignore-rev option. Note that a non-identical bundle may be created even
if the latest commit is the same if the local system's git configuration is
different. (This is not intuitive, and I suspect the cause is the record of
the username of the cloning user in the cloned repository's logs.)
There's currently a prohibition against URLs without HTTPS scheme, meaning no
git@github.com (scheme-less) or HTTP (insecure) URLs. The URL is passed as
an argument to git clone, so in theory a git@github.com:username/repo.git
URL could work, but an explicit guard clause currently prohibits it. URLs with
scheme HTTP are disallowed for security reasons, but also could work in
theory. (For testing purposes, file URLs pointing to bundle file locations
on the local filesystem are also allowed, but I can't imagine a use case for
outside of testing.)
The program ignores repositories where cloning fails, unless it fails for all of them, in which case it exits with a nonzero exit code. A common cause of failure is a URL that specifies a repository that doesn't exist. Evidence of this error is a log message about a failure with exit code 128 and a message that a username could not be read. (I think this is because git assumes that if gets a 404 response for an HTTP URL, the reason is that you lack authorization, not that the resource doesn't actually exist, so it asks for credentials, but we disable the terminal prompt, so it can't get the credentials and aborts.)
Windows is not supported as an execution platform, but POSIX-like platforms should all be supported, though testing is only performed on Linux.
From the cloned repository directory, execute:
$ ./run_tests.py
Note that this will run all unit tests, including those that connect to
external git repositories like GitHub. To run only the unit tests that
stay local, add the --no-external flag. See ./run_tests.py --help
for other options.