I strongly suspect that a bunch of ProcessBuilder tests are a bit faulty - most test multiple launchMechanism modes, which is okay, but some of them, during testing, span process trees of three layers or more, and those child JVMs must also get passed the launchMechanism, otherwise we repeatedly just test the POSIX_SPAWN mode.

The typical reason for using deeper trees is to pass an environment variable to the middle process (not possible, easily, in-process in the outermost test JVM), or to test stderr redirection.

I think at least affected are:
- Basic.java (the redirectIO test part)
- JspawnHelperProtocol (see parentCode())