[Bug] Codecov patch bugginess

[genedan]

From our observations, the Codecov patch workflow has been failing in some weird ways:

• Failing when a PR, even if it does not modify library code
• Failing when merging, even after passing during a PR

[Abhayindia]

happy to take this. codecov.yml sets only project, so patch falls back to target: auto (must beat 92.79%) and there's no ignore, which is why non-library files fail it - 6b51f64 changed only .github/scripts/check_spec0.py + a workflow and patch failed at 85.71%.
the merge-time flakiness looks separate: 6 uploads share the unittests flag with no after_n_builds, so codecov scores partial reports - 8a42e83 was a markdown-only merge that still reported project -1.04%. PR with the config fix if you want it?

[Abhayindia]

correction to my earlier comment - the ignore/patch config isn't the main cause. codecov is scoring merge commits against the second parent (the PR branch tip) rather than the first: on 8a42e83 its status says "compared to 445a2ef", which is 8a42e835^2, so the diff it scored included chainladder/__init__.py, _config/deprecation.py, options.py and others - none of which that merge touched. The real merge diff there is one markdown file, .github/pull_request_template.md, which is why a docs-only merge could report project -1.04%.
same shape on 6b51f64: the real merge diff is 727 lines of .github/ only, but codecov scored 7 coverable lines against 3c56ace (^2), which is a 30-file/2222-line diff of work that was already on main. that one wrong base explains both of your symptoms together. i don't think this is fixable purely in codecov.yml - the options look like either pinning the base or not uploading on pushes to main at all, since the PR run already scored it. which direction would you rather take?