Expressions returning primitives can't return null, so changing how

constant tiles are created to keep catalyst optimizer from pruning them.

Author: metasim

core/src/main/scala/org/locationtech/rasterframes/RasterFunctions.scala

@@ -231,9 +231,8 @@ trait RasterFunctions {
 
   /** Constructor for tile column with a single cell value. */
   def rf_make_constant_tile(value: Number, cols: Int, rows: Int, cellTypeName: String): TypedColumn[Any, Tile] = {
-    import org.apache.spark.sql.rf.TileUDT.tileSerializer
-    val constTile = encoders.serialized_literal(F.makeConstantTile(value, cols, rows, cellTypeName))
-    withTypedAlias(s"rf_make_constant_tile($value, $cols, $rows, $cellTypeName)")(constTile)
+    val constTile = udf(() => F.makeConstantTile(value, cols, rows, cellTypeName))
+    withTypedAlias(s"rf_make_constant_tile($value, $cols, $rows, $cellTypeName)")(constTile.apply())
   }
 
   /** Create a column constant tiles of zero */

core/src/main/scala/org/locationtech/rasterframes/expressions/tilestats/DataCells.scala

@@ -21,12 +21,12 @@
 
 package org.locationtech.rasterframes.expressions.tilestats
 
+import org.locationtech.rasterframes.expressions.{NullToValue, UnaryRasterOp}
 import geotrellis.raster._
-import org.apache.spark.sql.catalyst.expressions.codegen.CodegenFallback
+import org.apache.spark.sql.{Column, TypedColumn}
 import org.apache.spark.sql.catalyst.expressions.{Expression, ExpressionDescription}
+import org.apache.spark.sql.catalyst.expressions.codegen.CodegenFallback
 import org.apache.spark.sql.types.{DataType, LongType}
-import org.apache.spark.sql.{Column, TypedColumn}
-import org.locationtech.rasterframes.expressions.UnaryRasterOp
 import org.locationtech.rasterframes.model.TileContext
 
 @ExpressionDescription(
@@ -40,10 +40,11 @@ import org.locationtech.rasterframes.model.TileContext
        357"""
 )
 case class DataCells(child: Expression) extends UnaryRasterOp
-  with CodegenFallback {
+  with CodegenFallback with NullToValue {
   override def nodeName: String = "rf_data_cells"
   override def dataType: DataType = LongType
   override protected def eval(tile: Tile, ctx: Option[TileContext]): Any = DataCells.op(tile)
+  override def na: Any = 0L
 }
 object DataCells {
   import org.locationtech.rasterframes.encoders.StandardEncoders.PrimitiveEncoders.longEnc

core/src/main/scala/org/locationtech/rasterframes/expressions/tilestats/NoDataCells.scala

@@ -21,12 +21,12 @@
 
 package org.locationtech.rasterframes.expressions.tilestats
 
+import org.locationtech.rasterframes.expressions.{NullToValue, UnaryRasterOp}
 import geotrellis.raster._
-import org.apache.spark.sql.catalyst.expressions.codegen.CodegenFallback
+import org.apache.spark.sql.{Column, TypedColumn}
 import org.apache.spark.sql.catalyst.expressions.{Expression, ExpressionDescription}
+import org.apache.spark.sql.catalyst.expressions.codegen.CodegenFallback
 import org.apache.spark.sql.types.{DataType, LongType}
-import org.apache.spark.sql.{Column, TypedColumn}
-import org.locationtech.rasterframes.expressions.UnaryRasterOp
 import org.locationtech.rasterframes.model.TileContext
 
 @ExpressionDescription(
@@ -40,10 +40,11 @@ import org.locationtech.rasterframes.model.TileContext
        12"""
 )
 case class NoDataCells(child: Expression) extends UnaryRasterOp
-  with CodegenFallback {
+  with CodegenFallback with NullToValue {
   override def nodeName: String = "rf_no_data_cells"
   override def dataType: DataType = LongType
   override protected def eval(tile: Tile, ctx: Option[TileContext]): Any = NoDataCells.op(tile)
+  override def na: Any = 0L
 }
 object NoDataCells {
   import org.locationtech.rasterframes.encoders.StandardEncoders.PrimitiveEncoders.longEnc

core/src/test/scala/org/locationtech/rasterframes/RasterFunctionsSpec.scala

@@ -374,15 +374,14 @@ class RasterFunctionsSpec extends TestEnvironment with RasterMatchers {
 
       checkDocs("rf_no_data_cells")
     }
+
     it("should properly count data and nodata cells on constant tiles") {
       val rf = Seq(randPRT).toDF("tile")
 
       val df = rf
         .withColumn("make", rf_make_constant_tile(99, 3, 4, ByteConstantNoDataCellType))
         .withColumn("make2", rf_with_no_data($"make", 99))
 
-      df.show(false)
-
       val counts = df.select(
         rf_no_data_cells($"make").alias("nodata1"),
         rf_data_cells($"make").alias("data1"),