gistfile1.pyw

# This emulates what happens in Titanium's android support "run.py"
# script
import subprocess

p = subprocess.Popen(args=['adb', '-e', 'devices'], 
        stderr=subprocess.PIPE,
        stdout=subprocess.PIPE)

# If the adb daemon is NOT already running, 
# (i.e., if you run 'adb kill-server' without
# also running 'adb start-server')
# the following line will hang under Windows XP with
# Python 2.6.4 (possibly also with other Windows/Python
# combinations -- this is the only combo i have.)
# It does NOT hang under Linux/OSX.
(so, se) = p.communicate()
print 'Stdout: %s' % so
print 'Stderr: %s' % se

# Discussion: when 'adb -e devices' is run, it will
# start the adb daemon if it's not already running.
# If you do it from the command prompt you see this output:
#
# ----------------------------------------------------
#    C:\temp\test_popen>adb -e devices
#    * daemon not running. starting it now *
#    * daemon started successfully *
#    List of devices attached
#    
#    
#    C:\temp\test_popen>
# ----------------------------------------------------

# The fork that occurs to start the adb daemon seems to
# mess around with STDOUT and cause a deadlock if you then
# attempt to read STDOUT on that Popen instance, which 
# is exactly what communicate() does.  The same deadlock
# occurs if you use .wait() and .stdout.read() instead of
# communicate(), like this:
# >>> p.wait()
# >>> print p.stdout.read() # deadlock
# 
#
# Because the adb daemon is _NOT_ running by default on my
# XP machine (I didn't do anything like include 'adb start-server' in a 
# startup script), this deadlock _always_ happens the first time I 
# launch an app from Titanium Developer in any given XP session.
# Each subsequent launch works, because the adb daemon is then 
# successfully started already, and the call to 'adb -e devices'
# does not fork any processes to start the daemon, so STDOUT 
# operates normally and no deadlock occurs.
#
# I would recommend checking if platform.system() == 'windows' and, if so, running a Popen
# of 'adb start-server' (without bothering to check stdout -- just fire-and-forget) before
# calling 'adb -e devices'. There doesn't appear be any harm done
# by calling 'adb start-server' even when daemon already running.