entrypoint: fail on skipped package#66
Conversation
|
@GeorgeSapkin What do you think? |
When packages are not enabled in the configuration (e.g. due to unsupported architecture), they were previously skipped with a warning. This allowed the build to pass without compiling/testing the packages. Change the warning to an error and exit with status 1. Fixes: 94d922b ("Warn if using incompatible architecutre") Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
76ca5b8 to
0285ddb
Compare
|
I don't quite understand the intention. Is it because there's no warning, i.e. some intermediary run status in GHA? |
|
You are essentially correct. Having a 'warning' status in GitHub Actions would be ideal. However, we are currently hitting an issue where a submitted PR will show the following deep in the run details: Consequently, the build appears to have passed everywhere, despite warnings showing in the run Summary (could be hidden for someone). It might be worth writing a script to post a comment about the skipped build, but a reviewer might still overlook it if all the visible checks are green. It's an awkward setup. While extracting the warning to a comment is a potential solution, in my humble opinion, the pipeline should just fail in this scenario. Unsupported should equal a failed run. |
|
It is worth noting that in this case no tests execute - they are completely bypassed. |
When packages are not enabled in the configuration (e.g. due to unsupported architecture), they were previously skipped with a warning. This allowed the build to pass without compiling/testing the packages. Change the warning to an error and exit with status 1.
This was noticed in openwrt/packages#29803, where all checks passed ✅, but the package wasn't actually compiled for any platform.