Look for unequal bytecode pairs in EQ2 from the same compiler major version

[wtwhite]

This is part of #33.

[wtwhite]

Ignoring the ecj compiler, there are no different-bytecode pairs from run 18 on wtwhite-vuw-vm with the same major compiler version:

wtwhite@wtwhite-vuw-vm:~/code/jcompile/oracle-construction$ perl -lane 'foreach ($F[0], $F[1]) { s|/[^/]+$||; s|^.*/|| } print join "\t", $F[0], $F[1]' < SameArtifactDifferentCompilerClass.txt | sort > just_compiler_versions.txt
wtwhite@wtwhite-vuw-vm:~/code/jcompile/oracle-construction$ grep -v 'ecj-' just_compiler_versions.txt | uniq -c
    307 openjdk-10.0.2	openjdk-11.0.12
  10651 openjdk-11.0.19	openjdk-15.0.1
     49 openjdk-16.0.1	openjdk-17.0.1
    158 openjdk-17.0.7	openjdk-18.0.2
    127 openjdk-18.0.2	openjdk-19.0.2
   5655 openjdk-8.0.372	openjdk-9.0.1
    945 openjdk-9.0.1	openjdk-10.0.2

Including the ecj compiler gives unclear results since (a) it uses a different versioning scheme and (b) I just realised that the EQ2 approach of only comparing "adjacent" compiler versions doesn't work sensibly when there is more than one compiler "lineage". In any case:

wtwhite@wtwhite-vuw-vm:~/code/jcompile/oracle-construction$ uniq -c just_compiler_versions.txt
    168 ecj-3.11.1.v20150902-1521_openjdk-11.0.19	ecj-3.12.3_openjdk-11.0.19
   1000 ecj-3.12.3_openjdk-11.0.19	ecj-3.15.1_openjdk-11.0.19
    328 ecj-3.12.3_openjdk-11.0.19	ecj-3.24.0_openjdk-11.0.19
    896 ecj-3.15.1_openjdk-11.0.19	ecj-3.24.0_openjdk-11.0.19
     36 ecj-3.24.0_openjdk-11.0.19	ecj-3.28.0_openjdk-11.0.19
    787 ecj-3.24.0_openjdk-11.0.19	openjdk-11.0.19
  12816 ecj-3.32.0_openjdk-11.0.19	openjdk-11.0.19
    307 openjdk-10.0.2	openjdk-11.0.12
   7596 openjdk-11.0.16	ecj-3.11.1.v20150902-1521_openjdk-11.0.19
    534 openjdk-11.0.16	ecj-3.24.0_openjdk-11.0.19
    303 openjdk-11.0.16	ecj-3.28.0_openjdk-11.0.19
   2570 openjdk-11.0.16	ecj-3.29.0_openjdk-11.0.19
   2600 openjdk-11.0.16	ecj-3.30.0_openjdk-11.0.19
  10651 openjdk-11.0.19	openjdk-15.0.1
     49 openjdk-16.0.1	openjdk-17.0.1
    158 openjdk-17.0.7	openjdk-18.0.2
    127 openjdk-18.0.2	openjdk-19.0.2
   5655 openjdk-8.0.372	openjdk-9.0.1
    945 openjdk-9.0.1	openjdk-10.0.2

[wtwhite]

Another thing to try: The "Towards Build Verifiability for Java-based Systems" (Xiong et al.) paper mentions on p. 9 as P14 that JDK versions newer than jdk8-b44 fix a nondeterminism issue with lambda expression names "in a multithreaded setting". So we could add an older version of JDK8 to the compiler set and see whether that finds bytecode differences across many runs (perhaps starting with an artificial lambda-heavy project).