bin/176713: [patch] nc(1) closes network socket too soon

[Adrian Chadd]

Hi,

Well, I've done this before. More than once. I'm glad that you've
stuck through helping me understand what nc is doing; I'm
unfortunately busy doing other things

What you end up doing is:

* tracking the state of the two sockets, both for read EOF and write EOF;
* whenever you get an EOF from one of the above four conditions, you
see whether you can still make progress. If so, you don't close the
socket - you just stop registering for that event. (Eg, if you see
read EOF on a socket, stop registering for read)
* if you see that both sides are read EOF'ed, then you can't possibly
make any more progress, right?
* .. no, you also have to THEN ensure that all the data you have
queued is written, OR that you hit write EOF (and thus can't make any
more progress anyway)

_then_ you shut things down.

Anyway. Leave it with me. Bug me if I don't commit either your fix, or
rearchitect it to do what I said above (and then test it, obviously :)


-adrian