See https://github.com/apache/incubator-tvm/commits/master From today (JPN time, starting from @merrymercy’s commit) the commit author is not who sends the PR, but who merges it.
This affects the contribution history. My big PR https://github.com/apache/incubator-tvm/pull/4977 was merged by @janimesh a few hours ago, but it is on @janimesh’s commit history https://github.com/apache/incubator-tvm/commits?author=anijain2305
Seems to be project setting problem? IPMC members should have permission to review the setting.
sorry what is IPMC? Is this apache problem?
UPDATE: ok, Incubator Project Management Committee https://incubator.apache.org/guides/roles_and_responsibilities.html#incubator_project_management_committee_ipmc
Is it possible to escalate this? This is not ok.
If we can fix this quickly, maybe we should not merge PRs till then.
cc @tqchen we do need to fix this, otherwise only people with commiters’ right can build op contribution history. I don’t think anybody would like that.
It is also happening on mxnet, see the PR 17747 for example
The same issue reported in other repo
If it is indeed a github bug, it would be interesting
we can create test repo to check if it is a github issue
Sent an email to github support. Perhaps all related project should do so to escalate the issue. I also reported to the ASF infra to see if they have ways of escalation to github. Will update the status
This is a pretty big problem because as a project we will not have clear attribution of authorship based on the git commit history. That will be problematic for the project as we would have to do the messing about in the pull requests to find out authorship and no one would be able to trust the git authorship records. I think we should we stop merges till this is solved .
Ramana
see thread https://github.com/scikit-learn/scikit-learn/pull/16640 hopefully it will be resolved soon
Yes, seems pretty fundamentally broken. Good it’s being fixed. Thanks for picking this up with the relevant parties.
Update, github has reverted to the old behavior and it will take about 24 hours for it to take effect, we will delay new code merges until then.
The problem has been resolved
Can you fix the wrong authors that were already committed?
In order to amend the authors in these commits, we will need to generate new commits that are in parallel of the current chain. Essentially it means we will need to regenerate all these commits which leads to a different(parallel) chain, and force push to master and rewrite the commit history.
Force push to the master could cause quite a lot of problems (we will first need to ask ASF to remove the branch protection in the master, and run the rewrite, then finally ask everyone in the community to resync to the master, normal sync back from master to forks will break because of the diverged chain).
I want to say that the community and committers should are made aware of the github problem. More importantly, we evaluate contributions in a comprehensive manner rather than directly counting the loc, so hopefully the few wrong commit stats won’t affect the overall recognition of everyone’s controbutions
It is unfortunate that we cannot amend the info easily. Given the few commits involved, and potential new problems that can be caused by force pushing to the master, perhaps we should continue with the current commit history. Would also love to hear from everyone’s thoughts
I fully support your proposal. Personally I’m not happy the way it turned out, but I don’t see any other choice.
Thanks for the discussion.
First of all, I think it is not great that GitHub makes such breaking change without making customers well noticed - This is not the fault of our community.
For our community, the fact that these are only a few commits doesn’t mean those commits are not meaningful, and they are in fact extremely valuable. I would say it is important for the community to give credits to everyone’s hard work - community is all about people.
I fully understand that force pushing could cause quite a lot of problems, but given the fact that it was only a few days ago, not like a one-year-ago history, I think we should do a voting thread, making every community member aware of this breaking behavior, and see how others think. If the majority think it is just okay, we should do this.
Also, I think the problem is not alone - all projects on GitHub are affected. It is good if Apache or GitHub could have a good way to deal with it.