Builds are failing
javagl opened this issue · comments
As mentioned in #24 (comment) (and as it can be seen in the PR list), the builds are currently failing when a new model is added. I zoomed into that, a little bit, but ... am not a GitHub actions expert. So with the usual disclaimers (I have no idea what I'm doing here. LOL), a few comments:
Starting with an aside:
The workflows are run twice in the PR that I linked to above. This is probably related to the combination of
- the two events that trigger them, namely
push
andpull_request
- @echadwick-artist working directly on the
KhronosGroup
repo, and not on a fork
For example, in #26 , I'm creating a PR from a fork, and there, the workflows are only run once. Maybe there's a tricky way to avoid running them twice when the PR comes from KhronosGroup
, but ... it shouldn't be so much an issue. (In fact, there sometimes seem to be difficulties when trying to run such actions based on PRs from a fork, but ... maybe we can make it work both ways).
The actual failing build:
- It does work when it is triggered by the
push
event - It does not work when it is triggered by the
pull_request
event
Looking at the logs:
-
For the
push
case, it ends up saying
On branch SpecularSilkPuf
at https://github.com/KhronosGroup/glTF-Sample-Assets/actions/runs/5968588148/job/16192718277?pr=24#step:5:16 -
For the
pull_request
case, it says
HEAD detached at pull/24/merge
at https://github.com/KhronosGroup/glTF-Sample-Assets/actions/runs/5968588374/job/16192719075?pr=24#step:5:16
And it cannot commit with a deached head.
Now... why is the head detached here? I don't know. I haven't looked at the implementation of the actions/checkout@v2
action. From quickly skimming the documentation, it talks about a ref
property that may determine what exactly is checked out, depending on what triggered the action. This might be related (or similar) to actions/checkout#455.
I could reproduce the issue in my "Sandbox" repo (so apparently, ~"something in GitHub" had changed and caused this issue, because it worked smoothly a while ago!). And I played a bit with some some comments from the linked issue, and this change seems to resolve the problem:
Maybe we want to try this out here as well.
I opened a PR with an attempt to fix this at #28 , maybe someone wants to have a look.
I'll just claim that it is not possible to use GitHub actions for the tasks that are required for this repository.
(It's not that I think that it is not possible. It was possible until a few weeks ago, and then someone screwed things up. I just claim that it is not possible)
We should remove the failing GitHub action, and someone should manually update the model index periodically.
The action has been changed to just perform the necessary consistency checks and report the results. The build of the repo indices is still a manual operation. Leaving this issue open until the build of the repo can be addressed.
Replaced by #54