Issue1389

/home/waf/jython/bin/jython -c "import subprocess,
time;t=time.time();[subprocess.Popen(['true'], shell=False).wait() for x
in xrange(1000)];print time.time()-t"

cpython: 3 seconds
jython: 25 seconds

Is this really our fault? How does Java's plain ProcessBuilder fair 
against CPython?

Thomas Nagy emailed me and says it is our fault as Java's plain 
ProcessBuilder is faster. If someone could post a code snippet of the 
Java ProcessBuilder running the same task, that would help for someone 
looking deeper into this, wanting to compare the two

I think we're obviously going to be slower than CPython and 
ProcessBuilder, the question is by how much. Seems like there could be 
room for improvement

Some of this is Java's fault; some of this is ours.  Profiling it, all I 
could see was a lot of time attributed to calling through 
PyReflectedFunction.

In a quick test, subprocess.Popen in Jython took 28 seconds to spawn 1000 
processes, using ProcessBuilder from Jython took 22 and using 
ProcessBuilder from Java took 12.  subprocess from CPython took 3.

What I used to test with Java:

class sample {
    public static void main(String[] args) {
        try {
        long start = System.currentTimeMillis();
        for (int i = 0 ; i < 1000 ; i++) {
            new ProcessBuilder("true").start().waitFor();
        }
        System.out.println(System.currentTimeMillis() - start);
        } catch (Exception e) {
            e.printStackTrace();
        }

    }
}