Column order is incorrect and nondeterministic when writing to CSV with capital letters

When serializing to a CSV the column order output is incorrect but only when one or more attributes has a capital letter. The output order is also nondeterministic. Two back-to-back runs with the exact same values can produce different output orders.

Code to reproduce bug (Kotlin):

package com.draper.surveillance.common

import com.fasterxml.jackson.annotation.JsonPropertyOrder
import com.fasterxml.jackson.databind.MapperFeature
import com.fasterxml.jackson.databind.ObjectWriter
import com.fasterxml.jackson.dataformat.csv.CsvMapper
import java.io.File


class MyClass(
   val zProp1: Int,
   val Prop2: Int,
   val Prop3: Int)

val csvMapper = CsvMapper().apply {
   disable(MapperFeature.SORT_PROPERTIES_ALPHABETICALLY)
}
val bootstrapSchema = csvMapper.schemaFor(MyClass::class.java)
   .withUseHeader(true)
val writer: ObjectWriter = csvMapper
   .writerFor(MyClass::class.java)
   .with(bootstrapSchema)

val sequenceWriter = writer.writeValues(File("test.csv"))
object Test {
   val l = listOf(
      MyClass(1,2,3),
      MyClass(4,5,6)
   )

   @JvmStatic
   fun main(args: Array<String>) {
      sequenceWriter.writeAll(l)
   }

}

Running the main method multiple times will result outputs with different orders. If the property names in MyClass are changed to all lowercase then the issue disappears. This issue has been confirmed to happen in Java as well, version 2.13.0. See this.

[cowtowncoder]

Right: earlier JDKs tended to return Fields and Methods in order but even that is not guaranteed and apparently has lately been changing. In addition, combination ordering across Fields/Methods is not defined and use of inheritance can basically make it impossible to find out full declaration order anyway. In hindsight it would probably have made for Jackson to default to alphabetic ordering from version 1.0. But at the time it was not as obvious as later on; and wasn’t as much an issue with JSON anyway as ordering should not have semantic meaning. This is definitely different with positional formats like CSV.

[JacobGH111]

Thank you, this is tremendously helpful and now I understand why this is not a bug but just a limitation of the JVM.