Sorry friends, it was STILL not completely correct! :-(
On Mon, Oct 20, 2008 at 07:09:06AM +0200, Alexander Burger wrote:
> Fixed! I uploaded a new "testing" version.
> Both effects were a result of the fact that 'pipe' did not write to the
> process' standard output as it should. Instead, it used the parent
> process' current output channel, causing the conflicts. Fixing that
> involved only inserting some output channel initialization.
It was still wrong when the child process nested another output channel.
But now I believe it is right. It is fixed in the latest "testing".