Enforcing code standards in git before commit is accepted

Question

Alright, here's the scenario: A team of developers wants to ensure all new code matches the defined coding standards and all the unit tests are passing before a commit is accepted. Here's the trick, all of the tests need to run on a dedicated testing machine and we do not have access to modify the git server so this must be done using a local commit hook on each dev machine.

While the specs are pretty strict (we're not switching to windows or subversion, for example) this is a real world problem so there is some flexibility if you have a solution that almost fits.

• We're using Git and *nix.
• The updated code needs to be sent to another server to run the test suite.
• A list of modified files needs to be provided to ensure they match the coding standard.
• Its a rather large codebase, so we should send the smallest amount of information necessary to ensure identical copies of the codebase.
• If the tests fail a message needs to be displayed with the error and the commit should be blocked.
• Assume we trust our dev team and its okay to allow the tests to be bypassed with --no-verify option.

The question: What is the best way to get the test server to sync up with the local environment to run the tests? Some sort of hash-to-hash matching with a git patch for the new commit? Skip Git altogether and just do an rsync? Something else altogether?

Update 8/7/13: I shot myself in the foot by even mentioning the remote repo. The point isn't to block the code from being pushed to the shared / remote repo, its to prevent the local commit from even happening. Whether or not this would be considered a best practice is not really the point in this case, as this is specific to a small team of developers who all want this exact functionality. The question is about the best way to achieve the goal.

Answer 1

Add a custom git command which:

1. Temporarily does the commit
2. Pushes it to the remote server
3. Runs the tests (using a CI server, post-receive hook, or ssh)
4. Reverts the commit if the tests fail

Create an executable called git-test-n-commit and place it in your path:

#!/bin/bash
echo "Committing results..."
git commit "$@"
echo "Pushing to remote testing server..."
git push remote-server-git-url remote-branch -f
echo "Running tests"
ssh remote-server 'cd my_repo; run-my-tests' || 
   (echo "Tests failed, undoing commit" && git reset HEAD^)

Then instead of calling git commit ARGS, the developers can call git test-n-commit ARGS. If the tests pass the code will remain committed. If the tests fail it will be as if it was never committed.

Answer 2

Short answer:

DON'T

---

Long answer:

Not being able to do a commit because of some strange local hooks seems odd to me: you have to separate making a commit (i.e. saving your changes) from publishing this commit.

While it's perfecly reasonable to have a pre-receive hook on the git server you mention that enforces your rules, it would be dramatic for the repos of your devs: every time they want to save their work, try out something new or whatever, they would first have to polish their code before they could do the commit.

This would be counterproductive: people would feel like back in the bad old days when they had to commit anything to the repo and everybody could see their errors, bad coding style and so on. You would lose the freedom that a DVCS gives you: cheap branching, local history while maintaining a central repo for the production code.

Do not enforce anything when doing a local commit.

Answer 3

Local commit hooks are definitely not what you want here.

Your requirement that 'we do not have access to modify the git server so this must be done using a local commit hook on each dev machine' is completely bogus. You can always set up another repository that is your 'test remote' which you have full control over (which will then sync up with the git server you have no control over).

Once you set up this test remote, you can add hooks to run your tests on any push. The effort to type git push test-remote my-branch to get test results is pretty minimal.

Continuous integration with Git

Also check out Jenkins, gitlab, etc...

---

Update after 8/7/13:

So you really want to do some 'tests' on a remote server to prevent commits. If you want to prevent based on the content of the commit itself, use the pre-commit hook. See this question for how to get a list of changed files. Once you have those changed files, you can get them to a remote server using scp or rsync and run a test command with ssh.

If you need to check the commit message use the commit-msg hook.

Here is a good tutorial on hooks: http://git-scm.com/book/en/Customizing-Git-Git-Hooks

It also mentions a few reasons why it might be a bad idea.

They’re often used to enforce certain policies, although it’s important to note that these scripts aren’t transferred during a clone. You can enforce policy on the server side to reject pushes of commits that don’t conform to some policy, but it’s entirely up to the developer to use these scripts on the client side. So, these are scripts to help developers, and they must be set up and maintained by them, although they can be overridden or modified by them at any time.

Answer 4

I think the best approach I've seen is git+gerrit+jenkins. The gerrit introduces a notion of a changeset. The jenkins gerrit plugin could build each published changeset (running any kind of tests you would like) and mark them as "verified", then verified changeset could be merged into a master branch. If changeset fails build, committer receives a notification email, then he could amend the commit and update the changeset.

Answer 5

After setting up an intermediate server as others have suggested, set your pre-receive hook up to run a script over the incoming commit and normalise it to your coding standards. Often coding standards are defined by computer-enforcable rules. Don't make your developers go around tweaking whitespace hither and tither when a simple bash script or whatever could do it for them. And if you have a script to do that, why make the developer run it manually when it can be ran automatically on each push to your intermediate repo.

Answer 6

Write a Makefile that:

1. Copies current files to test server.

2. Finds the output of the unit tests.

2.1. If there are no anomalies, run 'git commit'

2.2 If there are anomalies, echo an error.

Using Makefiles to do my compiling AND commits has been a godsend. I can automate complex formatting and file cleanup before committing.

Edit:

Of course, Makefiles are not the only things you can do. Ant/Bash/other scripting languages can do this for you.