I am looking into changing this a little bit in order to fix #8014 (comment) The problem with current approach (collecting all test results and generating coverage report) for large codebases is that it requires bunch of mem to store results (in superform case more than 40G, which is OOMed in CI). Going to make a PR to stream result/process coverage report as it goes so hopefully large tests can be better accommodated (and also exit early if any failure), hope this makes sense.