@bradrydzewski Thanks for taking the time to outline this. I’m wary of “rolling my own” approach for something so critical. Just wondering if you’d consider this to be a feature that’s still far-out in terms of landing in a release? Or something we might see in the coming weeks/months?
@geekdave I had issues with git clone (taking forever) and git merge (failing, asking for git username and email) in the approach I described above. Our current solution that seems to be working so far is:
# We clone submodules ourselves in ci-clone
# We add a custom clone step to workaround a bug with GitHub merge ref
And ci-clone defined as:
# if build is PR rebase on top of target branch
if [ "$DRONE_BUILD_EVENT" = "pull_request" ]; then
git config user.email "email@example.com"
git config user.name "Some Name"
git pull "$DRONE_REMOTE_URL" "$DRONE_BRANCH"
git submodule update --init --recursive
uses the head ref for pull requests and manually performs the merge
guarantees the correct pull request sha and will fail when encountering merge conflicts
preserves the branch and sha and avoids a detached HEAD state (supporting tools that use git rev-parse HEAD)
There are some limitations with the implementation
pull requests tested with github only; ref logic in place for gitlab, gitea, gogs and bitbucket
does not yet work with github deployment events
does not yet support skip verify
does not yet support non-empty directories (important if caching .git directory)
does not support submodules (this is not planned, cloning submodules should instead be handled as a step in your pipeline)
There is solid unit testing in place and I think this provides a good base that we can build upon. If this is something you would like to help accelerate, please consider donating a few hours of time to help test and provide feedback and ways to improve.
You can use this plugin today for non-critical projects with the following syntax:
Glad to be a guinea pig, @bradrydzewski! I fired up this new image in my clone phase, and got this error right out-of-the gate on the /pr job (/push job ran normally):
chmod: 600: No such file or directory
Initialized empty Git repository in /drone/src/github.com/myorg/dummy-repo/.git/
+ git fetch origin +refs/heads/master:
* branch master -> FETCH_HEAD
* [new branch] master -> origin/master
+ git checkout master
Branch master set up to track remote branch master from origin.
Already on 'master'
+ git fetch origin pull/28/head:
* branch refs/pull/28/head -> FETCH_HEAD
+ git rebase 38482f303c72888a96d50338a5d8b466db9c097d
First, rewinding head to replay your work on top of it...
*** Please tell me who you are.
git config --global user.email "firstname.lastname@example.org"
git config --global user.name "Your Name"
to set your account's default identity.
Omit --global to set the identity only in this repository.
fatal: unable to auto-detect email address (got 'root@f8500a73a847.(none)')
Here’s my drone.yml:
- echo "This is a test to make sure Drone is running jobs."
I added a default email and name (via environment variable) which should satisfy that github prompt if it appears again. A new image is now available. Note that I didn’t have a test PR to reproduce the prompt, so this is untested, but according to git docs it should work.
First of all, big thanks for this. Outdated merge heads from Github started to pop up more and more in our deployment.
We now started using this in a few repo. First executions look good, and we are happy that this is hash based now.
There was one case though that may be of interest: in one of our long running branches the rebase failed with a conflict, while a manual merge of master to this branch succeeded without conflict. Retrying the build after this merge, the rebase succeeded as nothing left to rebase really.
I’m thinking that git rebase <hash> is perhaps more fragile than git reset --hard <hash> followed by a git merge master. But I’m sure you have reasons why it’s a rebase and not a merge.
not really, it was just the first thing that passed unit tests
The primary goal was was to preserve the hash and not create a new merge commit or create a detached head state. If the unit tests pass without modification with reset --hard I am open to adjusting. Thanks for testing, and for the feedback!
@gtaylor I recommend either testing the shell script locally, or maybe even testing with one of your repositories if possible. If we identify issues over the next few weeks, it will give us time to fix before this becomes the default in 0.9