Just a quick update; I have windows running again after creating a NUL (/dev/null) association for the parent processes' stdout handle. It solves a regression introduced in r483967 on Windows.
I'm somewhat puzzled, on traditional unix, NO_PIPE values for stdin/out/err would pass no fd at all, AIUI. It seems (?) that we presume they clone the parent's respective handles, if spelled out explicitly? We have a ton of leaking pipes and I'm trying to hunt them all down. Found the big one, we never set our own stderr handle as non-inherited before spawning the replacement stderr handler process, so it was *still* holding the old piped error logger open (just, not as stderr 'proper', but a plain old file handle). Still problems to hunt down, this is 40 hours into it. But I should have the simplest possible patches ready sometime tomorrow for folks to comment on and review. Bill
