Now, note that this is on Windows 7; the "start /wait" does basically
exactly what you think. *And indeed it operates correctly in a single
process*, but when nose tries to run multi-process, it fails to block,
and the test executes far sooner than it should.
Indeed, Nose claims that it ran 0 tests, when it actually at least ran
1! So, I think the myprog.exe is sending some signal to kill it.
It so happens that start has a "/B" parameter that may also help
here, but actually it doesn't. The result is that I don't get any
further output from the test (that is, the xunit xml isn't generated,
and even the console doesn't report anything as having happened).