Weekly cron job checking for submodule updates

[brenthuisman]

• Create a script that generates diff.log if git submodules have tags more recent that the commit they're currently at.
• A Github Action runs this script monthly.
• If updates found, Github Action creates an Issue like so: [AUTOMATED] Git submodule updates found brenthuisman/arbor#9
• Actually updating the git submodule and Spack file are still manual

First step in addressing #1731

[schmitts]

Consider adding set -euxo pipefail (https://vaneyckt.io/posts/safer_bash_scripts_with_set_euxo_pipefail)

[brenthuisman]

I think for this use-case it isn't what we want, because it would exit the script and Github Action early and we would not get output for any submodules that were not checked yet. We want to exit late, not early, right?

[schmitts]

I would still put it. We want to exit whenever a command fails. Consider 4 fails (silently). Then we would get an empty diff.

[schmitts]

add a small summary of what the script does.

[brenthuisman]

diff returns nonzero upon any difference. So no, here we don't want to exit upon any 'fail', we'd ever only get the difference reported for a single module.

I'll add a summary!

[schmitts]

You're right (Cf. http://mywiki.wooledge.org/BashFAQ/105). But what about error handling, e.g., if git submodule foreach 'git fetch' fails?

[brenthuisman]

When I disable network and run the script, the fetch loop terminates early (after the first remote being inaccessible) and the diff is generated based on unupdated module remotes. That is not terribly useful, but also not terribly bad, assuming that by running this weekly, such a failure every now and then will do no more harm than postpone us being notified a week late of updates.

Is that acceptable?

[schmitts]

Some minor comments inline

[schmitts]

LGTM

[noraabiakar]

If there are differences between local and upstream submodules, and the automatically generated issue is not addressed, does that mean that every week an identical issue will be opened?

[noraabiakar]

Why do we need both | tee and > ?

[brenthuisman]

To both display the output on stdout and have it in the file.

[noraabiakar]

I just tried it out locally and it didn't display the output on stdout. But git submodule foreach 'git describe HEAD --tags' | tee current_state_of_git_submodules_in_arbor_repo.log does what you describe.

[brenthuisman]

Whoops :)

[brenthuisman]

If there are differences between local and upstream submodules, and the automatically generated issue is not addressed, does that mean that every week an identical issue will be opened?

Correct.

[noraabiakar]

I think a new issue every week might be too much, what do you think?
Edit: I guess it's fine. We can just close the old ones.

[brenthuisman]

Weekly is perhaps too frequent, I've changed it to a monthly job.