git-repo/docs/repo-hooks.md
Mike Frysinger f7c51606f0 hooks: support external hooks running different Python version
As we convert repo to support Python 3, the version of Python that we
use might not be the version that repo hooks users have written for.
Since repo upgrades are not immediate, and not easily under direct
control of end users (relative to the projects maintaining the hook
code), allow hook authors to declare the version of Python that they
want to use.

Now repo will read the shebang from the hook script and compare it
against the version of Python repo itself is running under.  If they
differ, we'll try to execute a separate instance of Python and have
it load & execute the hook.  If things are compatible, then we still
use the inprocess execution logic that we have today.

This allows repo hook users to upgrade on their own schedule (they
could even upgrade to Python 3 ahead of us) without having to worry
about their supported version being exactly in sync with repo's.

Bug: https://crbug.com/gerrit/10418
Change-Id: I97c7c96b64fb2ee465c39b90e9bdcc76394a146a
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/228432
Reviewed-by: David Pursehouse <dpursehouse@collab.net>
Tested-by: Mike Frysinger <vapier@google.com>
2019-07-27 01:10:40 +00:00

5.5 KiB

repo hooks

[TOC]

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.
https://android.googlesource.com/platform/tools/repohooks

Approvals

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.

Manifest Settings

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 pre-upload phase.

<project path="tools/repohooks" name="platform/tools/repohooks" />
<repo-hooks in-project="platform/tools/repohooks" enabled-list="pre-upload" />

Source Layout

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 main accept **kwargs for future compatibility.

Runtime

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 repo client where the operation is started. For example, if the repo client is under ~/tree/, then that is where the hook runs, even if you ran repo in a git repository at ~/tree/src/foo/, or in a subdirectory of that git repository in ~/tree/src/foo/bar/. Hooks frequently start off by doing a os.chdir to the specific project they're called on (see below) and then changing back to the original dir when they're finished.

Python's 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.

Shebang Handling

*** note This is intended as a transitional feature. Hooks are expected to eventually migrate to Python 3 only as Python 2 is EOL & deprecated.


If the hook is written against a specific version of Python (either 2 or 3), the script can declare that explicitly. Repo will then attempt to execute it under the right version of Python regardless of the version repo itself might be executing under.

Here are the shebangs that are recognized.

  • #!/usr/bin/env python & #!/usr/bin/python: The hook is compatible with Python 2 & Python 3. For maximum compatibility, these are recommended.
  • #!/usr/bin/env python2 & #!/usr/bin/python2: The hook requires Python 2. Version specific names like python2.7 are also recognized.
  • #!/usr/bin/env python3 & #!/usr/bin/python3: The hook requires Python 3. Version specific names like python3.6 are also recognized.

If no shebang is detected, or does not match the forms above, we assume that the hook is compatible with both Python 2 & Python 3 as if #!/usr/bin/python was used.

Hooks

Here are all the points available for hooking.

pre-upload

This hook runs when people run repo upload.

The 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.
    """