Agent tests provide useful high entropy coverage of actor code but their runtime is too long to fit within testing runs (local and CI) without major increases in test time and hence decreases in useability.
We should figure out a way to schedule these tests to run periodically so that they don't disrupt development but catch errors. An obvious first candidate is to run all simulation tests in a separate nightly CI job. The nature of these tests also means that it could be worth running them continuously with large ranges of seeds so that many different operation sequences are hit.
Agent tests provide useful high entropy coverage of actor code but their runtime is too long to fit within testing runs (local and CI) without major increases in test time and hence decreases in useability.
We should figure out a way to schedule these tests to run periodically so that they don't disrupt development but catch errors. An obvious first candidate is to run all simulation tests in a separate nightly CI job. The nature of these tests also means that it could be worth running them continuously with large ranges of seeds so that many different operation sequences are hit.