Split CI into multiple workflows and use more granular steps

[mmuetzel]

Pull request purpose

GitHub allows temporarily disabling workflows (e.g., in forks of the repository). It doesn't have built-in granularity to disable jobs inside the same workflow though.
To do that currently, a user could, e.g., remove jobs manually and then rebase their actual changes on a branch where the jobs are still intact.

Additionally, GitHub truncates the output for long steps in its web interface.

Detailed changes proposed in this pull request

Simplify disabling parts of the CI temporarily by splitting the workflow files into three (one per platform).

Additionally, use more granular steps to make it easier to catch where the runs potentially fail. This also reduces the risk of the output being truncated by GitHub's web interface.

There is no difference in coverage by the CI with these changes. The exact same configurations as before still exists after this change. They are only slightly re-organized to help with future changes.

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[mmuetzel]

Hm...
One of the runners failed for a workflow that I didn't even touch:
https://github.com/opencollab/arpack-ng/actions/runs/30544381285/job/90876712775?pr=489#step:3:140

Install packages
    ▷ Installing bash build-base ccache cmake gfortran eigen-dev lapack-dev 
    ERROR: Unable to read database: file format is invalid or inconsistent
    ERROR: Failed to open apk database: file format is invalid or inconsistent
    
    Error occurred at line 318:
      315 | 		echo '▷ Installing $pkgs'
      316 | 		apk add --update-cache $pkgs
      317 | 	SHELL
    > 318 | 	abin/"$INPUT_SHELL_NAME" --root /.setup.sh
      319 | fi
      320 | 
      321 | 
    Error: Error occurred at line 318: 	abin/"$INPUT_SHELL_NAME" --root /.setup.sh (see the job log for more information)
    Error: Process completed with exit code 1.

Maybe, a network error on the runner which caused it to download a corrupt file?

[fghoussen]

Nope not a network error... I re-triggered the job: repeated fail.
#483 was OK but CI fails ont the same job now on master after merge...

@mmuetzel: can you have a look at it? I guess the problem lies around the github action

[mmuetzel]

Nope not a network error... I re-triggered the job: repeated fail. #483 was OK but CI fails ont the same job now on master after merge...

@mmuetzel: can you have a look at it? I guess the problem lies around the github action

I'm not sure why that suddenly started failing. I'd guess that it is an issue with the packages database of Alpine Linux for that platform.

Thinking about that again, I'd probably do that entire workflow a bit differently today: Instead of using emulation to run a "native" compiler for each respective target platform, it would probably make more sense to use a cross-compiler to build without emulation for the respective target platforms. And then use qemu only to run the tests.
Also, GitHub started providing runners on 64-bit ARM hardware. So, that part of the workflow could run natively entirely.

I'll try to overhaul that part of the CI soonish, and then come back to this PR.

Marking this PR as a draft for the time being.

[fghoussen]

it would probably make more sense to use a cross-compiler to build without emulation for the respective target platforms. And then use qemu only to run the tests.
Also, GitHub started providing runners on 64-bit ARM hardware. So, that part of the workflow could run natively entirely.

@sylvestre: what do you think?

[mmuetzel]

Thanks for merging #491.
I rebased this PR on top of a current head of the master branch.

Marking as ready for review again.

[sylvestre]

Sounds good

[fghoussen]

@mmuetzel: thanks!