Merge pull request #181 from metarank/fix/double-release-guard

add guards for native resources

src/main/scala/io/github/metarank/ltrlib/booster/Booster.scala

@@ -5,11 +5,14 @@ import io.github.metarank.ltrlib.model.{Dataset, Model}
 import org.apache.commons.math3.linear.{Array2DRowRealMatrix, ArrayRealVector, RealMatrix, RealVector}
 
 trait Booster[D] extends Model {
+  protected var isClosed = false
   def save(): Array[Byte]
   def predictMat(values: Array[Double], rows: Int, cols: Int): Array[Double]
   def weights(): Array[Double]
   def close(): Unit
 
+  def whenNotClosed[T](f: => T): T = if (!isClosed) f else throw new Exception("booster is already closed")
+
   override def predict(values: RealMatrix): ArrayRealVector = {
     val rows = values.getRowDimension
     val cols = values.getColumnDimension

src/main/scala/io/github/metarank/ltrlib/booster/CatboostBooster.scala

@@ -11,11 +11,11 @@ import java.io.ByteArrayInputStream
 case class CatboostBooster(booster: CatBoostModel, bytes: Array[Byte]) extends Booster[String] {
   override def save(): Array[Byte] = bytes
 
-  override def close(): Unit = booster.close()
+  override def close(): Unit = whenNotClosed { booster.close() }
 
   override def weights(): Array[Double] = Array.emptyDoubleArray
 
-  override def predictMat(values: Array[Double], rows: Int, cols: Int): Array[Double] = {
+  override def predictMat(values: Array[Double], rows: Int, cols: Int): Array[Double] = whenNotClosed {
     val split = new Array[Array[Float]](rows)
     var i     = 0
     while (i < rows) {

src/main/scala/io/github/metarank/ltrlib/booster/LightGBMBooster.scala

@@ -11,19 +11,24 @@ import scala.collection.mutable
 
 case class LightGBMBooster(model: LGBMBooster) extends Booster[LGBMDataset] with Logging {
 
-  override def predictMat(values: Array[Double], rows: Int, cols: Int): Array[Double] = {
+  override def predictMat(values: Array[Double], rows: Int, cols: Int): Array[Double] = whenNotClosed {
     model.predictForMat(values, rows, cols, true, PredictionType.C_API_PREDICT_NORMAL)
   }
 
-  override def close(): Unit = model.close()
+  override def close(): Unit = whenNotClosed {
+    isClosed = true
+    model.close()
+  }
 
-  override def save(): Array[Byte] =
+  override def save(): Array[Byte] = whenNotClosed {
     model.saveModelToString(0, 0, FeatureImportanceType.SPLIT).getBytes(StandardCharsets.UTF_8)
+  }
 
-  override def weights(): Array[Double] =
+  override def weights(): Array[Double] = whenNotClosed {
     // numIteration=0 means "use all of them"
     // we use split there to match xgboost, which can only do split
     model.featureImportance(0, FeatureImportanceType.SPLIT)
+  }
 }
 
 object LightGBMBooster extends BoosterFactory[LGBMDataset, LightGBMBooster, LightGBMOptions] with Logging {

src/main/scala/io/github/metarank/ltrlib/booster/XGBoostBooster.scala

@@ -16,7 +16,7 @@ case class XGBoostBooster(
 ) extends Booster[DMatrix]
     with Logging {
 
-  override def predictMat(values: Array[Double], rows: Int, cols: Int): Array[Double] = {
+  override def predictMat(values: Array[Double], rows: Int, cols: Int): Array[Double] = whenNotClosed {
     val mat = new DMatrix(values.map(_.toFloat), rows, cols, Float.NaN)
     mat.setGroup(Array(rows))
     mat.setFeatureTypes(featureTypes)
@@ -30,9 +30,9 @@ case class XGBoostBooster(
     out
   }
 
-  override def close(): Unit = model.dispose()
+  override def close(): Unit = whenNotClosed { model.dispose() }
 
-  override def save(): Array[Byte] = {
+  override def save(): Array[Byte] = whenNotClosed {
     val bytes = new ByteArrayOutputStream()
     val data  = new DataOutputStream(bytes)
     data.writeByte(BITSTREAM_VERSION)
@@ -47,7 +47,7 @@ case class XGBoostBooster(
     bytes.toByteArray
   }
 
-  override def weights(): Array[Double] = {
+  override def weights(): Array[Double] = whenNotClosed {
     val names   = (0 until model.getNumFeature.toInt).map(i => s"feature$i").toArray
     val weights = model.getFeatureScore(names).asScala
     val result = for {