Thread (3 messages) flat view 3 messages, 2 authors, 2018-08-03

Re: Squash & Merge Implementation

From: Jeff King <hidden>
Date: 2018-08-03 13:49:05

On Fri, Aug 03, 2018 at 03:01:15PM +0300, Vadim Belov wrote:
1. This merges successfully without squash:
git checkout origin/master
git merge ${PR-Branch}
git push origin HEAD:master
git push origin --delete ${PR-Branch}
Right, this is a normal merge.
2. This closes the PR, but there is no update seen on master:
git checkout origin/master
git merge --squash --commit ${PR-Branch}
git push origin HEAD:master
git push origin --delete ${PR-Branch}
Doing "merge --squash --commit" doesn't do what you expect; namely
"--commit" does not override the non-committing nature of "--squash". It
only override a "--no-commit" found elsewhere.

IMHO this is something that could be improved in Git (i.e., telling the
difference between "the user did not say --no-commit" and "the user said
--commit" and respecting it for --squash).

But that explains what you see. The push to master is a noop, since you
didn't make a new commit. And then deleting the PR branch on GitHub
auto-closes the PR.
3. This fails to push to master with the error "GH006: Protected branch
update failed"  (despite that the PR is set to SUCCESS):
git checkout origin/master
git merge --squash ${PR-Branch}
git commit -am"comment"
git push origin HEAD:${m_mainBranch}
git push origin --delete ${m_prBranch}
So here you _do_ make an actual commit. But that commit doesn't look
like a merge to the receiver; it just looks like a single commit that
has all the changes there were on PR-Branch.

The tree of that commit should be the same tree that would result from a
real merge. So in theory things like protected-branch status checks
could handle that, but I suspect they use the actual commit id (the tree
id is fine if you're just doing CI, but if you wanted to have a status
check for commit messages, say, you'd obviously want that to be tied to
the actual commit object).

I don't offhand recall how that is implemented (and you could also be
falling afoul of other checks, like required reviews). But this is a
GitHub-specific question, and you should probably ask GitHub support to
go further.

-Peff
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help