related to #58

Do you have examples to reproduce the issue?

i don't have example, but i have this issue in production system

I recently fixed a bug related to this issue:
"kill the subprocess if the creation failed"
http://bugs.python.org/issue23173
change: 
https://code.google.com/p/tulip/source/detail?r=31b0c15d8a4d7213a514f25d848d5f00
336083b4

The bug is similar because on error during subprocess creation, the caller 
doesn't have access to the subprocess and so it cannot kill it. It is now done 
automatically.

> i don't have example, but i have this issue in production system

Ok, I found a way to find such bug: I wrote a patch to "emit ResourceWarning 
warnings if transports/event loops are not explicitly closed"
http://bugs.python.org/issue23243

Using this patch, I already found one obvious bug in create_connection():
https://code.google.com/p/tulip/source/detail?r=3c53308c720a4043eb165d67afe8d182
b1f1e1ec
"Fix BaseEventLoop._create_connection_transport(). Close the transport if the 
creation of the transport (if the waiter) gets an exception."

I also opened another issue related to this one:
http://bugs.python.org/issue23242
"Process must close the transport when the process exit"

I made a change in StreamWriter to clear its reference to the transport when 
the transport is closed. But I had to revert the change because it changed the 
behaviour of Process.stdin.write(): it raised RuntimeError instead of 
BrokenPipeError or ConnectionResetError when stdin is closed.

I don't know if it's possible to clear the reference without changing the 
behaviour. Otherwise, the issue must be documented in a comment in 
StreamWriter.close().

The following Tulip change (not merged in Python yet) should also fix even more 
resource leaks on cancellation:

changeset:   1470:10b8730112c6
user:        Victor Stinner <victor.stinner@gmail.com>
date:        Fri Jan 23 00:52:06 2015 +0100
files:       asyncio/base_events.py
description:
Close transports on error

Fix create_datagram_endpoint(), connect_read_pipe() and connect_write_pipe():
close the transport if the task is cancelled or on error.

@fafhrd91, @nikolay: Hi, I fixed many issues related to this bug report. Could 
you please try the development version of asyncio to check if your resource 
leaks are gone?

Without any script to reproduce the issue, this bug report is not very useful.

The ResourceWarning patch may help you to identify bugs in your code or bugs in 
asyncio.
http://bugs.python.org/issue23243

I just closed http://bugs.python.org/issue23243

There is a remaining issue related to subprocesses: 
http://bugs.python.org/issue23347

I close this issue because there is no scenario to reproduce the bug ("i have 
this issue in production system" is not a valid scenario).

Please open a thread on the Tulip mailing list if you want help on the topic. 
Please also test the latest development version of asyncio: I fixed a lot of 
resource leaks and added destructors to emit warnings when event loops or 
transports are not explicitly closed. Destructors are currently only enabled on 
Python 3.4 and newer to not create objects which cannot be destroyed by the 
garbage collector (see the PEP 442).

Thanks for the feedback! It helped me to identify and fix a lot of bugs.