[Bug]Hadoop-bam not closing seekable stream

build.sbt

@@ -4,7 +4,7 @@ import scala.util.Properties
 
 name := """bdg-sequila"""
 
-version := "0.5.4-spark-2.4.1"
+version := "0.5.5-spark-2.4.1-SNAPSHOT"
 
 organization := "org.biodatageeks"
 
@@ -103,7 +103,7 @@ resolvers ++= Seq(
 //fix for hdtsdjk patch in hadoop-bam and disq
 assemblyShadeRules in assembly := Seq(
   ShadeRule.rename("htsjdk.samtools.SAMRecordHelper" -> "htsjdk.samtools.SAMRecordHelperDisq").inLibrary("org.disq-bio" % "disq" % "0.3.0"),
-  ShadeRule.rename("htsjdk.samtools.SAMRecordHelper" -> "htsjdk.samtools.SAMRecordHelperHadoopBAM").inLibrary("org.seqdoop" % "hadoop-bam" % "7.10.0")
+  ShadeRule.rename("htsjdk.samtools.SAMRecordHelper" -> "htsjdk.samtools.SAMRecordHelperHadoopBAM").inLibrary("org.seqdoop" % "hadoop-bam" % "7.10.0").inProject
 
 )
 

src/main/scala/org/biodatageeks/datasources/BAM/BDGAlignmentRelation.scala

@@ -125,6 +125,7 @@ trait BDGAlignFileReaderWriter [T <: BDGAlignInputFormat]{
     logger.info(s"######## Reading ${resolvedPath} or ${path}")
     val alignReadMethod = spark.sqlContext.getConf(BDGInternalParams.IOReadAlignmentMethod,"hadoopBAM").toLowerCase
     logger.info(s"######## Using ${alignReadMethod} for reading alignment files.")
+    logger.info(s"######## Using inputformat: ${c.toString()}")
 
     alignReadMethod match {
       case "hadoopbam" => {

src/main/scala/org/biodatageeks/inputformats/BAMBDGInputFormat.java

@@ -233,7 +233,7 @@ private int addProbabilisticSplits(
         final SeekableStream sin =
                 WrapSeekable.openPath(path.getFileSystem(cfg), path);
 
-        final BAMSplitGuesser guesser = new BAMSplitGuesser(sin, cfg);
+        final org.seqdoop.hadoop_bam.BAMBDGSplitGuesser guesser = new org.seqdoop.hadoop_bam.BAMBDGSplitGuesser(sin, cfg);
 
         FileVirtualSplit previousSplit = null;
 
@@ -247,6 +247,7 @@ private int addProbabilisticSplits(
 
             long alignedBeg = guesser.guessNextBAMRecordStart(beg, end);
 
+
             // As the guesser goes to the next BGZF block before looking for BAM
             // records, the ending BGZF blocks have to always be traversed fully.
             // Hence force the length to be 0xffff, the maximum possible.

src/main/scala/org/biodatageeks/inputformats/BAMBDGRecord.java

@@ -0,0 +1,35 @@
+package htsjdk.samtools;
+
+
+public class BAMBDGRecord extends BAMRecord {
+
+    private byte[] mRestOfBinaryData = null;
+    protected BAMBDGRecord(final SAMFileHeader header,
+                        final int referenceID,
+                        final int coordinate,
+                        final short readNameLength,
+                        final short mappingQuality,
+                        final int indexingBin,
+                        final int cigarLen,
+                        final int flags,
+                        final int readLen,
+                        final int mateReferenceID,
+                        final int mateCoordinate,
+                        final int insertSize,
+                        final byte[] restOfData) {
+        super(header,referenceID ,coordinate, readNameLength, mappingQuality, indexingBin, cigarLen, flags, readLen, mateReferenceID,mateCoordinate,insertSize,restOfData );
+
+    }
+
+    @Override
+    protected void eagerDecode() {
+        getReadName();
+        getCigar();
+        getReadBases();
+        getBaseQualities();
+        //getBinaryAttributes();
+        super.eagerDecode();
+        mRestOfBinaryData = null;
+    }
+
+}

src/main/scala/org/biodatageeks/inputformats/BAMBDGRecordReader.java

@@ -133,8 +133,10 @@ public static long getKey0(int refIdx, int alignmentStart0) {
         // initialize() and Hadoop-BAM's own code that relies on
         // {@link BAMInputFormat} to call initialize() when the reader is
         // created. Therefore we add this check for the time being.
-        if(isInitialized)
+        if(isInitialized) {
+            if(in != null) in.close();
             close();
+        }
         isInitialized = true;
         reachedEnd = false;