Switch from travis-ci to GitHub workflow.

[davidlehn]

No description provided.

[anatoly-scherbakov]

I thought requrements.txt already contained test dependencies, so could we just add the new packages to that file perhaps?

[BigBlueHat]

flake8 isn't in requirements.txt, so I think this is till needed for the additional test code below.

[davidlehn]

The intent of requiremenst-test.txt was to avoid those deps being installed other than for testing. No need to install a linter for a regular install. What is best practice for that concept? I saw that somewhere and it seemed a reasonable approach.

[anatoly-scherbakov]

I don't believe requirements.txt somehow influences a regular installation of pyld project, like from PyPI. What matters is content of setup.py, which specifies:

    install_requires=[
        'cachetools',
        'frozendict',
        'lxml',
    ],
    extras_require={
        'requests': ['requests'],
        'aiohttp': ['aiohttp'],
        'cachetools': ['cachetools'],
        'frozendict': ['frozendict'],
    }

These are the dependencies. This opens the possibility to use requirements.txt as development dependencies.

As an ideal solution, I would use something like poetry (python-poetry.org) which manages both use cases, but that's a separate question I'd rather not propose doing in this PR.

[anatoly-scherbakov]

I 👏 applaud the decision to deprecate currently unsupported Python versions, which are < 3.8.

[anatoly-scherbakov]

These three aren't going to be necessary if #182 is merged.

[BigBlueHat]

@davidlehn I would like to do this on top of #182 if possible.

[davidlehn]

• I believe what I was thinking when this PR was written was to start with testing the code similar to how travis-ci was doing it but with github actions.
• This will give immediate familiar feedback for testing further changes.
• Implement git submodules for specs #182 is not addressing the test runner and code changes needed for rdf-canon tests. Every rdf-canon test will be skipped leaving a large hole in coverage if we are to release any versions before the issue is addressed. I don't think changing that test suite before rdf-canon is supported can be justified.
• It's arguably better to merge this first to get tests flowing. Implement git submodules for specs #182 can be rebased off of it and will then have tests running. The changes to make it use submodules are probably straightforward and would be best done with workflow tests to confirm. (Assuming we can get outside contributor PR tests running properly.)

[anatoly-scherbakov]

is not addressing the test runner and code changes needed for rdf-canon tests. Every rdf-canon test will be skipped

Should we fix that? I thought I did add that as a dependency and we should be running against that there...

[anatoly-scherbakov]

Suggested change

	python tests/runtests.py ./_json-ld-api/tests -l ${{ matrix.loader }}
	python tests/runtests.py ./_json-ld-framing/tests -l ${{ matrix.loader }}
	python tests/runtests.py ./_normalization/tests -l ${{ matrix.loader }}
	python tests/runtests.py -l ${{ matrix.loader }}

is going to be sufficient.

[BigBlueHat]

@anatoly-scherbakov the submodule command will have to be added.

[anatoly-scherbakov]

@BigBlueHat it seems that actions/checkout has an option to fetch submodules so we can likely avoid running make even.

[anatoly-scherbakov]

Suggested change

	- uses: actions/checkout@v4
	- uses: actions/checkout@v4
	with:
	submodules: true

As per https://github.com/actions/checkout

Applies to the below again.

[BigBlueHat]

Good to know we can do this, but I agree with @davidlehn that we should get this merged before #182 and can revisit this if/when #182 makes it in.

[anatoly-scherbakov]

@BigBlueHat it seems that actions/checkout has an option to fetch submodules so we can likely avoid running make even.

[BigBlueHat]

I'm in favor of getting this merged sooner than later--and it can happen now (vs. as part of a release) since it only affects the master branch and nothing about the running code (or even the tests themselves).

@anatoly-scherbakov we should separate the rdf-canon introduction from #182, so both can move forward at different speeds.