# $Id: TODO,v 2.10 2012/10/07 16:00:53 ksb Exp $ Fix quick-mode. New tactic. Every request for just stdout starts in quick-mode with a buffer of some fixed length (e.g. 16k or 64k). The client asks for the size of the buffer (remaining space) and sends what they can. A request for remaining space returns any non-zero space immediately, but blocks (the client in read) when there is no space left. When the client gets control we send them back a non-a-number which tells them to read the access rights for a direct output. When requesting more than just stdout we don't do this, we just block them until they get the token. Not much, just a detail to two. Maybe support the "M?" command to pass the process-id of the utility back on the wrapper port. We do that in ptbw and wrapw. Under -m we could take (at about function Notify) -a exit success only when the master and all clients exit 0 -o exit failure unless any client exits 0 (ignore master) -ao exit failure unless any client exits 0 (and the master does) xapply could't pass these, but one could use an explicit wrapper, or $XCLATE_1. When stdout (the managed stream) is /dev/null we could allow multiple output-only (non-IED) clients, after all mixing streams to the null device is not going to confuse anyone :-). That actually happens more than you'd think. I'm not sure about the -i option with a process. The process gets the original stdin for it's stdin, but -I still retains the same fd (which means we can steal data). I guess the Implementor knows what she's asking for with -I and -i. I don't think we need a way to get util's stdin, because it could be an xclate -m if you wanted that (for some strange reason). If you don't want the original stdin you can use "