[ADAM-1402] Fix INDEL realigner bad binary search.

Resolves bigdatagenomics#1402.

adam-core/src/main/scala/org/bdgenomics/adam/rdd/read/realignment/IndelRealignmentTarget.scala

@@ -65,7 +65,7 @@ private[realignment] object TargetOrdering extends Ordering[IndelRealignmentTarg
   }
 
   /**
-   * Compares a read to an indel realignment target to see if it starts before the start of the indel realignment target.
+   * Compares a read to an indel realignment target to see if the target is before the read.
    *
    * @param target Realignment target to compare.
    * @param read Read to compare.

adam-core/src/main/scala/org/bdgenomics/adam/rdd/read/realignment/RealignIndels.scala

@@ -74,24 +74,27 @@ private[read] object RealignIndels extends Serializable with Logging {
     read: RichAlignmentRecord,
     targets: TreeSet[(IndelRealignmentTarget, Int)]): Int = {
     // Perform tail call recursive binary search
-    if (targets.size == 1) {
-      if (TargetOrdering.contains(targets.head._1, read)) {
-        // if there is overlap, return the overlapping target
-        targets.head._2
-      } else {
-        // else, return an empty target (negative index)
-        // to prevent key skew, split up by max indel alignment length
-        (-1 - (read.record.getStart / 3000L)).toInt
-      }
+    if (TargetOrdering.contains(targets.head._1, read)) {
+      // if there is overlap, return the overlapping target
+      targets.head._2
+    } else if (targets.size <= 1) {
+      // else, return an empty target (negative index)
+      // to prevent key skew, split up by max indel alignment length
+      (-1 - (read.record.getStart / 3000L)).toInt
     } else {
       // split the set and recurse
       val (head, tail) = targets.splitAt(targets.size / 2)
-      val reducedSet = if (TargetOrdering.lt(tail.head._1, read)) {
-        head
+
+      if (TargetOrdering.contains(tail.head._1, read)) {
+        tail.head._2
       } else {
-        tail
+        val reducedSet = if (TargetOrdering.lt(tail.head._1, read)) {
+          tail
+        } else {
+          head
+        }
+        mapToTarget(read, reducedSet, test)
       }
-      mapToTarget(read, reducedSet)
     }
   }
 

adam-core/src/test/scala/org/bdgenomics/adam/rdd/read/realignment/IndelRealignmentTargetSuite.scala

@@ -79,6 +79,12 @@ class IndelRealignmentTargetSuite extends ADAMFunSuite {
     assert(targets.head.variation.get.end === 8)
     assert(targets.head.readRange.start === 3)
     assert(targets.head.readRange.end === 10)
+    assert(TargetOrdering.contains(targets.head, read))
+    assert(!TargetOrdering.lt(targets.head, read))
+    val read2 = make_read(2L, "2M3D2M", "2^AAA2", 4, 7)
+    val read4 = make_read(4L, "2M3D2M", "2^AAA2", 4, 7)
+    assert(!TargetOrdering.lt(targets.head, read))
+    assert(TargetOrdering.lt(targets.head, read4))
   }
 
   sparkTest("creating simple target from read with insertion") {

adam-core/src/test/scala/org/bdgenomics/adam/rdd/read/realignment/RealignIndelsSuite.scala

@@ -22,13 +22,14 @@ import org.bdgenomics.adam.algorithms.consensus.{
   Consensus,
   ConsensusGenerator
 }
-import org.bdgenomics.adam.models.ReferencePosition
+import org.bdgenomics.adam.models.{ ReferencePosition, ReferenceRegion }
 import org.bdgenomics.adam.rdd.ADAMContext._
 import org.bdgenomics.adam.rdd.read.AlignmentRecordRDD
 import org.bdgenomics.adam.rdd.variant.VariantRDD
 import org.bdgenomics.adam.rich.RichAlignmentRecord
 import org.bdgenomics.adam.util.ADAMFunSuite
 import org.bdgenomics.formats.avro.{ AlignmentRecord, Contig, Variant }
+import scala.collection.immutable.TreeSet
 
 class RealignIndelsSuite extends ADAMFunSuite {
 
@@ -53,6 +54,40 @@ class RealignIndelsSuite extends ADAMFunSuite {
     sc.loadAlignments(path).rdd
   }
 
+  def makeRead(start: Long, end: Long): RichAlignmentRecord = {
+    RichAlignmentRecord(AlignmentRecord.newBuilder()
+      .setContigName("ctg")
+      .setStart(start)
+      .setEnd(end)
+      .setReadMapped(true)
+      .build())
+  }
+
+  test("map reads to targets") {
+    val ts = Array(
+      new IndelRealignmentTarget(None, ReferenceRegion("ctg", 1L, 4L)),
+      new IndelRealignmentTarget(None, ReferenceRegion("ctg", 10L, 44L)),
+      new IndelRealignmentTarget(None, ReferenceRegion("ctg", 100L, 400L)))
+      .zipWithIndex
+    val targets = TreeSet(ts: _*)(ZippedTargetOrdering)
+
+    assert(RealignIndels.mapToTarget(makeRead(0L, 1L), targets) < 0)
+    assert(RealignIndels.mapToTarget(makeRead(0L, 2L), targets) === 0)
+    assert(RealignIndels.mapToTarget(makeRead(1L, 2L), targets) === 0)
+    assert(RealignIndels.mapToTarget(makeRead(3L, 6L), targets) === 0)
+    assert(RealignIndels.mapToTarget(makeRead(6L, 8L), targets) < 0)
+    assert(RealignIndels.mapToTarget(makeRead(8L, 12L), targets) === 1)
+    assert(RealignIndels.mapToTarget(makeRead(10L, 12L), targets) === 1)
+    assert(RealignIndels.mapToTarget(makeRead(14L, 36L), targets) === 1)
+    assert(RealignIndels.mapToTarget(makeRead(35L, 50L), targets) === 1)
+    assert(RealignIndels.mapToTarget(makeRead(45L, 50L), targets) < 0)
+    assert(RealignIndels.mapToTarget(makeRead(90L, 100L), targets) < 0)
+    assert(RealignIndels.mapToTarget(makeRead(90L, 101L), targets) === 2)
+    assert(RealignIndels.mapToTarget(makeRead(200L, 300L), targets) === 2)
+    assert(RealignIndels.mapToTarget(makeRead(200L, 600L), targets) === 2)
+    assert(RealignIndels.mapToTarget(makeRead(700L, 1000L), targets) < 0)
+  }
+
   sparkTest("checking mapping to targets for artificial reads") {
     val targets = RealignmentTargetFinder(artificialReads.map(RichAlignmentRecord(_)))
     assert(targets.size === 1)