If the "special environment" is one that can be locally controlled, this can be done in the test. For something like adding a user, this is too intrusive to do on developers' machines. This has been handled in the past by writing tests that should pass in both the normal and special environment, then modifying the environment of one or more of the CICD runners. See, e.g., this additional setup for pinky (if these steps get too complicated, it's better to put them in their own script, like the script to set up an android emulator). That way, even if the test passes in the dev's normal environment, it will still be caught when the PR is filed.

If the "special environment" is one that can be locally controlled, this can be done in the test. For something like adding a user, this is too intrusive to do on developers' machines. This has been handled in the past by writing tests that should pass in both the normal and special environment, then modifying the environment of one or more of the CICD runners. See, e.g., this additional setup for pinky (if these steps get too complicated, it's better to put them in their own script, like the script to set up an android emulator). That way, even if the test passes in the dev's normal environment, it will still be caught when the PR is filed.

Thanks, the pinky example is very useful for testing the PR #7788, however the test in PR #7789 need to ensure that a specified userid is not in the system, what should we do if the userid is on the CICD runners?

I suspect the runner environments are pretty stable, so if one of them is missing userid 2, just adding a script to fail loudly if that changes would be sufficient; but I suspect that's unlikely, since we presumably would've caught the problem in that case. If you're confident you can safely modify the system to change userid 2 to something else, you could do so in the freebsd VM.

Ah, never mind, just read the test in question, I don't think you need a special environment at all (continuing discussion there--if there's nothing actionable here, I think we can close this issue).