Trying to understand a bit more about CI approaches (https://www.pantsbuild.org/stable/docs/using-pants/using-pants-in-ci) First question: The "Approach #1" is to use

--changed-since

and

--changed-dependents=transitive

to run only on changes. It does mention as an example that this wouldn't run on all files in a situation such as adding a plugin to

flake8

. Is it safe to assume that

pants.toml

changes are the only way to "bypass" dependency inference? Second question: If I use "Approach #1" and only run tests against (transitively) changed files, how does this affect the coverage report? Since my CI runners are ARC runners (ephemeral), it seems I would always have low coverage because I would get a global coverage report only listing the coverage from the tests that were run in this instance...

We run tests in prs only for transitive deps without coverage, and we have a separate nightly test run that includes coverage

Is it safe to assume that

pants.toml

changes are the only way to "bypass" dependency inference?

changing config files outside of

pants.toml

(e.g

.pylintrc

) could also introduce new errors when running linters on any python file, even though they wouldn't be considered

changed

so "Approach #1" wouldn't run them.