[23][Preview][Primitive type patterns] VerifyError with instanceof record patterns with conversion from double to Long

[jarthana]

Testcase:

record Record(Long i) {}
public class X  {
    public static double convert(Record r) {
    	return r instanceof Record(double d) ? d : 0;
    }
    public static void main(String[] args) {}
}

Exception thrown when run:

Caused by: java.lang.VerifyError: Inconsistent stackmap frames at branch target 35
Exception Details:
  Location:
    X.convert(LRecord;)D @15: ifnull
  Reason:
    Current frame's stack size doesn't match stackmap.
  Current Frame:
    bci: @15
    flags: { }
    locals: { 'Record' }
    stack: { 'java/lang/Long', 'java/lang/Long' }
  Stackmap Frame:
    bci: @35
    flags: { }
    locals: { 'Record' }
    stack: { }

Note: It's also cool to see the compiler error about "Internal inconsistency: Inappropriate operand stack size encountered during translation". After some more testing, may be we can make that an error instead of warning - @srikanth-sankaran ?

[srikanth-sankaran]

Byte code generated:

  public static double convert(Record);
    descriptor: (LRecord;)D
    flags: (0x0009) ACC_PUBLIC, ACC_STATIC
    Code:
      stack=4, locals=3, args_size=1
         0: aload_0                                
         1: instanceof    #16                 // class Record
         4: ifeq          35                        // <---------------  reach branch target with stackDepth == 0 
         7: aload_0
         8: checkcast     #16                 // class Record
        11: invokevirtual #18                 // Method Record.i:()Ljava/lang/Long;
        14: dup
        15: ifnull        35                     //  <---------------  reach branch target with stackDepth == 1
        18: iconst_1
        19: ifne          26
        22: pop
        23: goto          35
        26: invokevirtual #22                 // Method java/lang/Long.longValue:()J
        29: l2d
        30: dstore_1
        31: dload_1
        32: goto          36
        35: dconst_0
        36: dreturn
        37: new           #28                 // class java/lang/MatchException
        40: dup_x1
        41: swap
        42: dup
        43: invokevirtual #30                 // Method java/lang/Throwable.toString:()Ljava/lang/String;
        46: swap
        47: invokespecial #36                 // Method java/lang/MatchException."<init>":(Ljava/lang/String;Ljava/lang/Throwable;)V
        50: athrow

It seems we can reach bci with stack depth being 0 through one arc and stack depth being 1 through another arc.

[srikanth-sankaran]

See @bci 4 and @bci 15 and @bci 35 in the assembly dump above

[srikanth-sankaran]

@stephan-herrmann - thanks for taking a look

[srikanth-sankaran]

Note: It's also cool to see the compiler error about "Internal inconsistency: Inappropriate operand stack size encountered during translation". After some more testing, may be we can make that an error instead of warning - @srikanth-sankaran ?

It was the plan originally that we would as we generate code call out verification failures - but that is a project whose time has to come :)

We can consider the warning -> error change at some point. yes.

[srikanth-sankaran]

This is same as issue #3181 except for the indirection through the type variable. Which is worth capturing as a separate test.

The codeStream.instance_of(scope.getJavaLangObject()); fix outlined in #3301 also solves this