We have started seeing failures in the cmp-baseline builds that look like this:

Differing files in ./lib/modules
/java.base/java/lang/invoke/MemberName.class

This looks like a new case of non deterministic behavior when generating classes. The first appearance in CI is build jdk-17+15-1221. The previous build that got to tier5 was 1209, but since this is an intermittent problem, we can't know for sure if that build was actually affected or not.

From what I can tell, this file is just compiled with javac and not subject to any further modification by any build tool.

Looking at javap output for two differing builds of this class, the difference looks to be benign from a functional perspective. (The value in the "slot" column for 2 and 3 have switched places)

Variant1:
    LocalVariableTable:
      Start Length Slot Name Signature
         38 12 1 type Ljava/lang/Object;
         84 6 4 res Ljava/lang/invoke/MethodType;
         74 19 3 sig Ljava/lang/String;
        121 24 4 ptypes [Ljava/lang/Class;
        129 16 5 rtype Ljava/lang/Class;
        139 6 6 res Ljava/lang/invoke/MethodType;
        113 32 2 typeInfo [Ljava/lang/Object;
          0 211 0 this Ljava/lang/invoke/MemberName;

Variant2:
    LocalVariableTable:
      Start Length Slot Name Signature
         38 12 1 type Ljava/lang/Object;
         84 6 4 res Ljava/lang/invoke/MethodType;
         74 19 2 sig Ljava/lang/String;
        121 24 4 ptypes [Ljava/lang/Class;
        129 16 5 rtype Ljava/lang/Class;
        139 6 6 res Ljava/lang/invoke/MethodType;
        113 32 3 typeInfo [Ljava/lang/Object;
          0 211 0 this Ljava/lang/invoke/MemberName;


While this will not have any functional effect on compiled java classes, we would like to continue having predictable and repeatable output from javac.