COMPRESSED_FILTERED_TSI produces different output on different JDKs

[reckart]

Describe the bug COMPRESSED_FILTERED_TSI produces different output on different JDKs or at least on different systems.

To Reproduce Steps to reproduce the behavior: Run CasSerializationDeserialization_COMPRESSED_FILTERED_TSI_Test.serializeAndCompareToReferenceTest in particular test 12 (casWithLists) and 13 (casWithArrays) on different JDKs:

• ❌ MacOS openjdk version "1.8.0_252" OpenJDK Runtime Environment (AdoptOpenJDK)(build 1.8.0_252-b09) OpenJDK 64-Bit Server VM (AdoptOpenJDK)(build 25.252-b09, mixed mode)
• ✅ Mac OS openjdk version "1.8.0_332" OpenJDK Runtime Environment (Zulu 8.62.0.19-CA-macosx) (build 1.8.0_332-b09) OpenJDK 64-Bit Server VM (Zulu 8.62.0.19-CA-macosx) (build 25.332-b09, mixed mode)
• ❌ Linux Azul_JDK_8/zulu8.62.0.19-ca-jdk8.0.332-linux_x64

Expected behavior The format should be consistent across deployments.

Screenshots

• casWithLists-reference-cas.bin.zip (MacOS Zulu 8.62.0.19-CA-macosx) (build 25.332-b09, mixed mode)
• casWithLists-actual-cas.bin.zip (MacOS AdoptOpenJDK)(build 1.8.0_252-b09)

Please complete the following information:

• Version: 3.4.0-SNAPSHOT

[reckart]

Interestingly, the deserialized CASes seem to be equivalent though.

Maybe the difference is related to differences or non-deterministic behaviour in the compression or in the iteration order through the feature structures.

class Investigate {
  @Test
  void test() throws Exception {
    CAS ref = CasCreationUtils.createCas();
    CAS act = CasCreationUtils.createCas();

    CasIOUtils.load(new File("actual-cas.bin").toURL(), act);
    CasIOUtils.load(new File("reference-cas.bin").toURL(), ref);

    assertThat(toComparableString(act)).isEqualTo(toComparableString(ref));
  }
}