Re: write_pipe_chunks patch messes up early error message output

Andrew Dunstan <andrew@dunslane.net> writes:
> Anyway, I think the attached patch will do what we need.

I think you could leave Redirect_stderr out of the elog.c tests
entirely, since redirection_done can never become set without it.

Also, you introduced a bug: pgwin32_is_service is a function no?
Needs invocation parens.

Otherwise looks reasonable.
        regards, tom lane