Why the ci gate in fork repo has been disabled

Core Workflow
1

Hi, all.
Anybody know why the ci gate in fork repo has been disabled?

image
image852×344 26 KB

2

Which PR is this from?
It should done for documentation PRs: if the code isn’t modified, it doesn’t make sense to run tests.

3

Hi, petr. I created a PR in https://github.com/shihai1991/cpython/pull/26.
In this PR, I depend the ci gate to do some regrtests, but it’s can not be triggered in fact :frowning:
The ci gate of this PR in official repo can be triggered: https://github.com/python/cpython/runs/949786331

So I am not sure what have changed~

4

The reason is in your action logs.

I don’t think this is specifically done to prevent forks from running CI (and I don’t even know why it failed). I’m afraid you will have to dig into it to find out why it didn’t work in your fork.

5

Got it. I will try to take a look.

6

Hi, some pointers.

https://github.com/python/cpython/blob/master/.github/workflows/build.yml has an initial step check_source / “Check for source changes”, which does a git diff to check the changes are NOT only .rst, Doc or Misc (this is running in your PR).

        run: |
          if [ -z "GITHUB_BASE_REF" ]; then
            echo '::set-output name=run_tests::true'
          else
            git fetch origin $GITHUB_BASE_REF --depth=1
            git diff --name-only origin/$GITHUB_BASE_REF... | grep -qvE '(\.rst$|^Doc|^Misc)' && echo '::set-output name=run_tests::true' || true
          fi

If that passes, it sets a variable called run_tests to true, and the other build_[OS] steps only run when it’s true.

    if: needs.check_source.outputs.run_tests == 'true'

Your PR doesn’t set this for some reason, so the build steps are skipped.

Looking at your logs from https://github.com/shihai1991/cpython/pull/26/checks:

Run if [ -z "GITHUB_BASE_REF" ]; then
  if [ -z "GITHUB_BASE_REF" ]; then
    echo '::set-output name=run_tests::true'
  else
    git fetch origin $GITHUB_BASE_REF --depth=1
    git diff --name-only origin/$GITHUB_BASE_REF... | grep -qvE '(\.rst$|^Doc|^Misc)' && echo '::set-output name=run_tests::true' || true
  fi
  shell: /bin/bash -e {0}
From https://github.com/shihai1991/cpython
 * branch            master     -> FETCH_HEAD
 * [new branch]      master     -> origin/master
fatal: origin/master...HEAD: no merge base

So it’s failing to even do the diff. I’m not sure why.

I did a couple of test PRs in my own fork and they triggered builds as expected (#1, #2).

The second one is basically the same and has no error:

Run if [ -z "GITHUB_BASE_REF" ]; then
  if [ -z "GITHUB_BASE_REF" ]; then
    echo '::set-output name=run_tests::true'
  else
    git fetch origin $GITHUB_BASE_REF --depth=1
    git diff --name-only origin/$GITHUB_BASE_REF... | grep -qvE '(\.rst$|^Doc|^Misc)' && echo '::set-output name=run_tests::true' || true
  fi
  shell: /bin/bash -e {0}
From https://github.com/hugovk/cpython
 * branch            master     -> FETCH_HEAD
 * [new branch]      master     -> origin/master

Perhaps git fetch origin $GITHUB_BASE_REF --depth=1 needs a greater depth or something?

(These checks were added in https://github.com/python/cpython/pull/19983 / https://bugs.python.org/issue40548).

If you want to see the output of the git diff, add a new line without the -q:

git diff --name-only origin/$GITHUB_BASE_REF... | grep -vE '(\.rst$|^Doc|^Misc)'

But that’s not going to show anything when it has a fatal error.

As a temporary workaround, you could force a build every time:

diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index 5649a6670e..d0cefcc433 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -28,12 +28,7 @@ jobs:
       - name: Check for source changes
         id: check
         run: |
-          if [ -z "GITHUB_BASE_REF" ]; then
-            echo '::set-output name=run_tests::true'
-          else
-            git fetch origin $GITHUB_BASE_REF --depth=1
-            git diff --name-only origin/$GITHUB_BASE_REF... | grep -qvE '(\.rst$|^Doc|^Misc)' && echo '::set-output name=run_tests::true' || true
-          fi
+          echo '::set-output name=run_tests::true'
   build_win32:
     name: 'Windows (x86)'
     runs-on: windows-latest
1 Like
7

Wow, thanks hugo for your details :wink:

8

Hi, hugo. You are right. I have test it in my vm. Maybe you can create a PR to fix it :wink:

9

victor’s fix PR: https://github.com/python/cpython/pull/21806 (it’s not merged now, but it can fix this problem), thanks.

1 Like
10

It’s now merged.

2 Likes