Current CI/CD pipeline for Python 3.7 fails because of issue with importlib_metadata and flake8

[fpgmaas]

Example pipeline that fails: https://github.com/rstudio/py-shiny/actions/runs/3183453613

See also relevant issues here and here.

[fpgmaas]

Tried to fix it with #349, but the issue persists because flake8>5.0.0 is still installed by py-htmltools:
https://github.com/rstudio/py-shiny/actions/runs/3183453613/jobs/5192917088#step:5:21

I opened a Pull Request in py-htmltools as well; posit-dev/py-htmltools#44

[fpgmaas]

I figured we don't need to install the full requirements-dev.txt from py-htmltools, but only the wheel package. Therefore I propose the following change: link to diff

However, the issue still persists, because the following command seems to install importlib-metadata 5.0.0 anyway:

python3 -m pip install --force-reinstall dist/shiny*.whl --find-links ../py-htmltools/dist/

see this line in GitHub workflow run

I think the current workflow can use some improvement, because some packages, such as flake8 and importlib-metadata, seem to be installed by a pip install command thrice now.

[fpgmaas]

Weirdly enough, after pinning importlib_metadata in setup.cfg, the tests for Python 3.7 succeed, but they fail for 3.8.

workflow run

playwright._impl._api_types.Error: Protocol error (Page.navigate): error in channel "content::page": exception while running method "navigate" in namespace "page": Component returned failure code: 0x804b0013 (NS_ERROR_PORT_ACCESS_NOT_ALLOWED) [nsIWebNavigation.loadURI] _navigate@chrome://juggler/content/content/PageAgent.js:385:14

EDIT: This seems to be an unrelated issue. Workflow run triggered again and now succeeds.

[jcheng5]

@fpgmaas Yeah, I think the random port number generator just happened to pick a port number that browsers refuse to connect to. I'll fix that.

[fpgmaas]

@jcheng5 I will close this issue since the original issue seems to be resolved.