on pull request, checkout@v2 built some ref other than the one that was claimed

[jepler]

Recently, I had a PR adafruit/circuitpython#2831 which checked out green, so a fellow developer merged it. It broke the build.

In subsequent investigation, I noticed that while the linked Actions run https://github.com/adafruit/circuitpython/runs/633447824?check_suite_focus=true shows that it was building the same ref as the one that was merged, down in the logs you find something different.

Under the step "Run actions/checkout@v2", "Checking out the ref",

HEAD is now at e7eef6f0c Merge 6dfe9ab606352b36f8d295b250ac7118c459618b into 2d7cf4b792afa1b6755b9575b84cae4b65ae7ab9

6dfe9ab represents another revision of the code that was apparently replaced by force-push with the later value 75be728:

@jepler force-pushed the jepler:rgbmatrix-stm branch from 6dfe9ab to e5be728 7 days ago

[ericsciple]

Thanks! I'll loop in some folks.

Indeed something looks wrong.

From the mpy-cross-mac check run, the step "Dump GitHub context" reveals some key info:

{
  [...]
  "ref": "refs/pull/2831/merge",
  "sha": "e7eef6f0ca41e82a3082df79b7abbd6a369433a6",
  "repository": "adafruit/circuitpython",
  "event_name": "pull_request",
  [...]
  "event": {
    "action": "synchronize",
    "after": "e5be728c1c8a3a455b7095c9dab9e758c5016191",
    "before": "6dfe9ab606352b36f8d295b250ac7118c459618b",
    "number": 2831,
    [...]

Yet appears to contradict the git log -1 output at the end of the step "Run actions/checkout@v2":

/usr/local/bin/git log -1
commit e7eef6f0ca41e82a3082df79b7abbd6a369433a6
Merge: 2d7cf4b79 6dfe9ab60
Author: Jeff Epler <jepler@gmail.com>
Date:   Wed Apr 29 19:23:50 2020 +0000

    Merge 6dfe9ab606352b36f8d295b250ac7118c459618b into 2d7cf4b792afa1b6755b9575b84cae4b65ae7ab9

It looks like ${{ github.sha }} is the old merge commit.

[edit]: The commit message should read: Merge e5be728c1c8a3a455b7095c9dab9e758c5016191 into 2d7cf4b792afa1b6755b9575b84cae4b65ae7ab9 (since e5be7 is the "after" sha)

[ericsciple]

@andymckay have you heard of any similar occurrences?

The workflow payload contained the wrong merge commit.

[iheanyi]

👋🏾 Will be taking a look into this today and tomorrow.

[ericsciple]

Thanks @iheanyi for digging into this! We dug into this a bit and are unable to reproduce. This may be due to replication lag (received previous merge commit). Going forward we'll add more telemetry to the checkout step. So if this happens again we will be able to detect from telemetry.

[Bankst]

Hi, I just ran in to this same issue here:
PhotonVision/photonvision@f2305bb
The SHA for the commit is f2305bb501ec200d022a1bac4bac8ee316f68e54 but the workflow ran with commit 2e69b3e24b82faefc3702fc946834c61b31d8c4

https://github.com/PhotonVision/photonvision/pull/18/checks?check_run_id=844516246
Hopefully this can somehow help, having another known occurence