Hadoop classpath discovery broken in 0.12 when HADOOP_HOME is a symlink

[keyaaron]

CLASSPATH discovery was recently changed in d911850 to resolve ARROW-2113 and ARROW-3768.

Specifically, the logic used to find all jars under HADOOP_HOME uses the find command directly

arrow/python/pyarrow/hdfs.py

Line 144 in d911850

find_args = ('find', os.environ['HADOOP_HOME'], '-name', '*.jar')

This will not work when HADOOP_HOME is a symlink, in which case '-L' needs to be passed to the find command.

CLASSPATH can still be set explicitly, but this is a change in behavior as HADOOP_HOME symlinks worked without issue before.

[emkornfield]

Hi @keyaaron,
I opened a JIRA-4802 to track this. Would you be willing to submit a patch?

[chenfj068]

This issue has been resolved @emkornfield we can close it.