On Tue, Aug 31, 2010 at 01:51, Alan Bateman <[email protected]> wrote:
> > On the test, it might be best to leave out wakeupJeff for now. Even if we > come up with a solution for the asynchronous close issues then I assume this > test will need to change anyway because it runs commands that are highly > platform dependent. > I prefer to leave in wakeupJeff, and not only because of the toddler-parent-in-joke. It demonstrates something I have been wondering about for years, namely occasional non-reproducible Bad file descriptor errors (because EINTR can happen at any time, and it is unspecified which thread in a process is chosen to handle the signal). You're right that more test work is needed if this difficult problem is ever addressed. Martin
