JNI Hadoop file wrappers truncate paths containing literal #

[zhtttylz]

Describe the bug
Auron's JNI Hadoop file wrappers may truncate paths containing a literal #.
This happens because JniBridge rebuilds paths via new Path(new URI(path)), where Java URI treats everything after # as a fragment, so Hadoop receives a truncated path.

For example, this intended path:

hdfs://mycluster/auron-it-hdfs-rbf-repro/raw#mini.txt

is opened as:

/auron-it-hdfs-rbf-repro/raw

To Reproduce

from pyspark.sql import SparkSession


spark = SparkSession.builder.appName("auron-hdfs-hash-repro").getOrCreate()
jvm = spark._jvm
conf = spark._jsc.hadoopConfiguration()

path = "hdfs://mycluster/auron-it-hdfs-rbf-repro/raw#mini.txt"
hadoop_path = jvm.org.apache.hadoop.fs.Path(path)
fs = hadoop_path.getFileSystem(conf)

print("path=", path)

if fs.exists(hadoop_path):
   fs.delete(hadoop_path, False)

out = fs.create(hadoop_path, True)
out.write(bytearray(b"data"))
out.close()
print("plain_hadoop_create_ok")

plain_in = fs.open(hadoop_path)
plain_in.close()
print("plain_hadoop_open_ok")

jvm.org.apache.auron.jni.JniBridge.openFileAsDataInputWrapper(fs, path).close()
print("jni_open_ok")

spark.stop()

The relevant output is:

Expected behavior
Auron should preserve Hadoop Path(String) semantics when opening or creating files through JNI Hadoop wrappers.
JniBridge.openFileAsDataInputWrapper and JniBridge.createFileAsDataOutputWrapper should open the exact path string passed by the caller, including literal # characters in the Hadoop path.

[cxzl25]

Thank you for reporting this issue. Is it possible to submit a PR to fix this issue?