I need connect the hdfs-server. How do I set username and groupname? Thank you 

The following will crash the jvm, as long the as the path exists:

    fs = pyhdfs.connect(host, port)
    l = pyhdfs.listdir(fs, '/doesnotexist')

This really does not seem to be a bug in libpyhdfs, but I figured I'll file 
this ticket so anyone else looking for this problem sees this.

I have no idea how hadoop can trigger a jvm segfault.

This is the output:

#
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x00007ff0fadb0ee8, pid=29972, tid=140673005336304
#
# JRE version: 6.0_16-b01
# Java VM: Java HotSpot(TM) 64-Bit Server VM (14.2-b01 mixed mode linux-
amd64 )
# Problematic frame:
# V  [libjvm.so+0x3d4ee8]
#
# An error report file with more information is saved as:
# /home/tv/src/libpyhdfs.git/test/hs_err_pid29972.log
#
# If you would like to submit a bug report, please visit:
#   http://java.sun.com/webapps/bugreport/crash.jsp
#

A sample logfile is attached.

Including jars, hadoop config and a prebuilt shared lib is extremely 
unhelpful:

1. it actually prevents using this package with an actual Hadoop setup, 
e.g. looking up the real config file

2. putting the files at a hardcoded /usr/local location prevents installing 
the project with a --prefix= (distutils aborts with a sandbox error).

3. using the same .so isn't going to work across architectures

4. using a .so or jars you just download over the net is a security risk

Attaching a suggested patch dont-bundle.diff that just recommends using the 
Cloudera debs.

Setting CLASSPATH inside the C source is also a bad idea, as it embeds the 
version numbers of the jars in it, and they just won't ever match what the 
person running the thing will have. Attaching patch no-classpath.diff that 
removes that. Sadly, as running Java just never really got nicely 
integrated with the operating system, and Hadoop provides no easy mechanism 
for finding the right paths, the caller will have to set the CLASSPATH. But 
at least that can happen outside of C, in Python or shell, and it can be 
done so that it matches the runtime environment.

If given hostname "default" and port 0, libhdfs will read the right values 
from hadoop config. Attached patch makes the script to "just work" for most 
of the users.

Hi. Here's a patch that adds listdir support. Enjoy!

What steps will reproduce the problem?

1. hadoop fs -mkdir /empty
2. print pyhdfs.listdir(fs, '/empty')

What is the expected output? What do you see instead?

Expected [], got None.

This would let me differentiate between errors and empty directories.
Attached patch to fix this.

see https://issues.apache.org/jira/browse/HDFS-596