Running lintr can take quite some time for large packages, so I think it is a good idea to exclude it from being run during a code coverage check (where it is useless?), e.g. when using covr. When lintr is invoked in testthat using expect_lint_free, the test is skipped if NOT_CRAN is not "true". Therefore, when using package_coverage() to get coverage data locally, lintr is excluded.

Travis seems to set NOT_CRAN to "true", however, so I assume lintr is run twice there and is leading to longer Travis build times.

I suggest also checking for the "R_COVR" environment variable in expect_lint_free, and skipping linting it is set to "true".