Repository navigation
GitHub Actions issues not triggering #373
Description
Activity
Did we shove all of these into a single file with this trick due to requiring status checks that don't get run blocks a PR?
Something has happened to the source checker task, e.g. python/cpython#21748
I confirmed that it worked well since May ( https://bugs.python.org/issue40548 ) but it started to fail with
fatal: origin/master...HEAD: no merge base. I'm not sure why it stopped working.I'm not sure if
git fetchfails or ifgit difffails.cc @FFY00
Did we shove all of these into a single file with this trick due to requiring status checks that don't get run blocks a PR?
I understood that @FFY00 chose to use a separated job to check if a PR is a "doc only" to easy maintenance of the check: the job uses Linux and reuses services like
uses: actions/checkout@v2(rather than having to implement the check for Windows, macOS, Linux, ...)The advantage is to not even start other jobs if it's a doc-only PR. It reduces resource consumption.
Oh! I can reproduce the issue with Git 2.28 (on Fedora Rawhide), but not with Git 2.26 (Fedora 32). Script to reproduce the issue:
set -e -x DIR=clone_python export GITHUB_BASE_REF=master rm -rf $DIR git init $DIR cd $DIR git remote add origin https://github.com/python/cpython git fetch --no-tags --prune --depth=1 origin +fe42ecf3c875b31a281ecb5b4e257655cf820134:refs/remotes/pull/21748/merge git fetch origin master --depth=1 git checkout --force refs/remotes/pull/21748/merge git diff --name-only origin/master... | catgit diff --name-only origin/master...fails withfatal: origin/master...HEAD: no merge base.But
git diff --name-only origin/master..refs/remotes/pull/21748/mergeworks as expected.Difference between 2 dots (..) and 3 dots (...):
https://matthew-brett.github.io/pydagogue/git_diff_dots.htmlIt seems like using 2 dots should be enough for the workflow job, since GitHub starts by merging master into the PR anyway.
I didn't find any obvious change on the "merge-base" command in the Git 2.28 changelog, but this version contains many changes, so I'm not surprised that the behavior changed.
I proposed python/cpython#21806 to replace "..." (3 dots) with ".." (2 dots).
Sorry for the delay! I wasn't able to look at this earlier.
See my comment in python/cpython#21806 (comment).
My python/cpython#21806 was merged into master and then backported to 3.8 and 3.9 branches. I close the issue.
Something has happened to the source checker task, e.g. python/cpython#21748
I'm not sure what this error means, but because the failure is treated as a success, the checks decide they don't need to run and so we get all green with no CI!