🧪 Fix uploading coverage from GHA to Codecov

[Jamim]

Hello @webknjaz,

What do these changes do?

The newest version of the codecov-action fails to submit coverage, so it isn't ready for use yet.

These changes downgrade codecov-action to v3.

Are there changes in behavior for the user?

No.

Related issues

• [v4] Traceback with ValueError when Git repo is not checked out codecov/codecov-action#1252
• [v4 upgrade] Config file codecov.yml not found, or is not a file. Ignoring config codecov/codecov-action#1242
• CLI does not detect codecov.yml files from allowed locations codecov/codecov-cli#365

Checklist

• I think the code is well written
• Documentation reflects the changes

Best regards!

[Jamim]

One more remarkable issue:

• Codecov fails, action passes? codecov/feedback#266

[Jamim]

🤦🏼 Well, while this PR fixes coverage uploading, some Codecov-related logic marks this PR with ❌

[Jamim]

For the sake of transparency, one more related issue:

• [v4] Traceback with ValueError when Git repo is not checked out codecov/codecov-action#1252

[Jamim]

@webknjaz wait a minute, if there's no checkout for the Tests status job, that's why the action can't find the config.

[webknjaz]

@webknjaz wait a minute, if there's no checkout at for the Tests status job, that's why the action can't find the config.

I think this is because Codecov suddenly swapped the defaults without telling anybody:

• https://github.com/codecov/uploader/blob/1e505ba8a27a1e3311a7a8db7f314e3705a2ee08/src/helpers/token.ts#L71-L76
• https://github.com/codecov/codecov-cli/blame/019a55a5e3c4c54fe587e318658fe093c304db99/codecov_cli/main.py#L38

[Jamim]

I'm not sure what the hell is going on this time, but coverage data is looking weird.
Just compare these two runs:

• https://app.codecov.io/github/aio-libs/multidict/commit/57de3b698e5a7c3b1c16f44481f0b91f1ce8a65c/tree/tests
• https://app.codecov.io/github/aio-libs/multidict/commit/7229252e71107a4202d8be277767a8c2f7ea1a39/tree/tests

It counts lines that shouldn't be counted.
I guess it's related to mypy.

[Jamim]

@webknjaz,
After a quick research I came to conclusion that it's mypy who spoils coverage reports.
I don't think it's a wise idea to mix test coverage with type annotation coverage data.
How can we distinct them on Codecov?

[Jamim]

How can we distinct them on Codecov?

Well, I just didn't know about the Flags.

[webknjaz]

Yes, the metrics for flags are combined. The details are somewhat granular, which is enough for now. I'm planning to experiment with combining the coverage setup with https://hynek.me/articles/ditch-codecov-python/ to expose the HTML files and stuff additionally. And maybe, add more stable checks for some code areas.

[webknjaz]

Interestingly, the MyPy-produced coverage doesn't include anything from the multidict/ folder. I wonder why.

[webknjaz]

https://github.com/aio-libs/multidict/blob/19d0511/.mypy.ini#L2-L4
well, that explains it