Repo provides a mechanism to hook specific stages of the runtime with custom python modules. All the hooks live in one git project which is checked out by the manifest (specified during
repo init), and the manifest itself defines which hooks are registered.
These are useful to run linters, check formatting, and run quick unittests before allowing a step to proceed (e.g. before uploading a commit to Gerrit).
A complete example can be found in the Android project. It can be easily re-used by any repo based project and is not specific to Android.
When a hook is processed the first time, the user is prompted for approval. We don't want to execute arbitrary code without explicit consent. For manifests fetched via secure protocols (e.g. https://), the user is prompted once. For insecure protocols (e.g. http://), the user is prompted whenever the registered repohooks project is updated and a hook is triggered.
For the full syntax, see the repo manifest format.
Here's a short example from Android. The
<project> line checks out the repohooks git repo to the local
tools/repohooks/ path. The
<repo-hooks> line says to look in the project with the name
platform/tools/repohooks for hooks to run during the
<project path="tools/repohooks" name="platform/tools/repohooks" /> <repo-hooks in-project="platform/tools/repohooks" enabled-list="pre-upload" />
The repohooks git repo should have a python file with the same name as the hook. So if you want to support the
pre-upload hook, you'll need to create a file named
pre-upload.py. Repo will dynamically load that module when processing the hook and then call the
main function in it.
Hooks should have their
**kwargs for future compatibility.
Hook return values are ignored.
Any uncaught exceptions from the hook will cause the step to fail. This is intended as a fallback safety check though rather than the normal flow. If you want your hook to trigger a failure, it should call
sys.exit() (after displaying relevant diagnostics).
Output (stdout & stderr) are not filtered in any way. Hooks should generally not be too verbose. A short summary is nice, and some status information when long running operations occur, but long/verbose output should be used only if the hook ultimately fails.
The hook runs from the top level of the git repo where the operation is started. e.g. If you're in the git repo
src/foo/, that is where the hook runs, even if the
repo command was started from a subdir like
sys.path is modified so that the top of repohooks directory comes first. This should help simplify the hook logic to easily allow importing of local modules.
Repo does not modify the state of the git checkout. This means that the hooks might be running in a dirty git repo with many commits and checked out to the latest one. If the hook wants to operate on specific git commits, it needs to manually discover the list of pending commits, extract the diff/commit, and then check it directly. Hooks should not normally modify the active git repo (such as checking out a specific commit to run checks) without first prompting the user. Although user interaction is discouraged in the common case, it can be useful when deploying automatic fixes.
Here are all the points available for hooking.
This hook runs when people run
pre-upload.py file should be defined like:
def main(project_list, worktree_list=None, **kwargs): """Main function invoked directly by repo. We must use the name "main" as that is what repo requires. Args: project_list: List of projects to run on. worktree_list: A list of directories. It should be the same length as project_list, so that each entry in project_list matches with a directory in worktree_list. If None, we will attempt to calculate the directories automatically. kwargs: Leave this here for forward-compatibility. """