Inaccurate code coverage metrics

[amclin]

Test coverage numbers are incorrectly significantly lower than they should be because Jest cannot track coverage for spawned processes. The integration test runs the full installer as a spawned process, so we are effectively testing a lot more lines than code coverage would lead us to believe.

This is a limitation in Jest:
jestjs/jest#5274

Once workaround is to use NYC to capture code coverage (since it can track spawned processes) and then wrap that with Jest:
jestjs/jest#3190 (comment)

[jamesliu-usb]

Getting a 403 when I attempt to push a fix/codeCoverage branch 😢

[amclin]

NYC maintainers discourage using spawn-wrap:
jestjs/jest#5274 (comment)

[jamesliu-usb]

Wait for nyc 15 then? Or go with the custom spawning approach that node-tmp took w/ their testing strategy?

The way I read their warning made me think most of the concern was around edge cases run on Windows runtimes, which made the spawn-wrap approach seem like an okay-ish interim solution til nyc 15 is released.

Though I suppose if CI is run on Windows-based servers that could monkey-wrench things... but don't think it is? Worth confirming?

[amclin]

Something is better than nothing. I'm happy caveating the edge case limitation with a //TODO to replace in the future when NYC 15 is available. NYC 15 is already in beta. Or we can look at using the beta version. It would only be needed in the generator, not the generated projects, so I think using a beta version could be acceptable as long as it doesn't have show-stopping bugs in our otherwise simple use case.

[amclin]

This new(ish) small runner might be a good alternative to NYC:
https://www.npmjs.com/package/coverage-node