Thanks for that link; the answer by Eryksun <https://stackoverflow.com/questions/23688492/oserror-errno-22-invalid-argument-in-subprocess#28020096> is particularly enlightening. Apparently EINVAL actually represents an underlying broken pipe condition in Windows. Anyway, as far as the test is concerned, it doesn’t matter what particular exception is raised.

This still doesn’t explain Serhiy’s observation that sometimes no exception is raised at all, <https://bugs.python.org/issue21619#msg236949>. Does Windows sometimes not report broken pipe errors, or is my signalling logic not working and there is a race condition? The whole point of the test is to trigger an exception when stdin is closed, so it would be nice to be able to reliably do that on Windows.