Do not trust nullable arrayclass type in VP

[a7ehuo]

Update isUnreliableSignatureType to check if the class is an array class. If the class is an array class and its component class is value type and the array class is not null-restricted array, the type is not reliable because it can either be a nullable array class type or a null-restricted array type.

Fixes: #19914

[a7ehuo]

@hzongaro May I ask you to review this change? Thank you!

The detailed analysis related to this change is in #19914 (comment)

[a7ehuo]

@hzongaro All comments are addressed in 49082a. Ready for another review. Thank you!

[hzongaro]

Looks good. Thanks!

[hzongaro]

Jenkins test sanity.functional,extended.functional xlinuxval,alinuxval,zlinuxval,plinuxval jdknext

[hzongaro]

Jenkins test sanity.functional,sanity.openjdk xlinux jdk8,jdk21

[a7ehuo]

Test_openjdk21_j9_sanity.openjdk_x86-64_linux_Personal failed with the following 4 vector tests (#55).

They also fail in other PR tests: #52 #54 #53

The nightly build hasn't completed yet but I can see it shows jdk_vector_long128_j9_0 fails in nightly build, which seems similar to these failures.

[2024-10-29T21:21:30.760Z] FAILED test targets:
[2024-10-29T21:21:30.760Z] 	jdk_vector_double128_j9_0 - Test results: error: 1 
[2024-10-29T21:21:30.760Z] 		Failed test cases: 
[2024-10-29T21:21:30.760Z] 			TEST: jdk/incubator/vector/Double128VectorTests.java

[2024-10-29T23:26:16.492Z] FAILED test targets:
[2024-10-29T23:26:16.492Z] 	jdk_vector_float128_j9_0 - Test results: error: 1 
[2024-10-29T23:26:16.492Z] 		Failed test cases: 
[2024-10-29T23:26:16.492Z] 			TEST: jdk/incubator/vector/Float128VectorTests.java
[2024-10-29T23:26:16.492Z]         
[2024-10-29T23:26:16.492Z] 	jdk_vector_short128_j9_0 - Test results: error: 1 
[2024-10-29T23:26:16.492Z] 		Failed test cases: 
[2024-10-29T23:26:16.492Z] 			TEST: jdk/incubator/vector/Short128VectorTests.java
[2024-10-29T23:26:16.492Z]         
[2024-10-29T23:26:16.492Z] 	jdk_vector_byte64_j9_0 - Test results: error: 1 
[2024-10-29T23:26:16.492Z] 		Failed test cases: 
[2024-10-29T23:26:16.492Z] 			TEST: jdk/incubator/vector/Byte64VectorTests.java

[hzongaro]

Value types testing was successful. Failures in non-value types testing were due to known problems. Merging.

[jdmpapin]

I don't think arrays of value types really need to be treated as unreliable like interfaces are

If the class is an array class and its component class is value type and the array class is not null-restricted array, the type is not reliable because it can either be a nullable array class type or a null-restricted array type.

This should be fine because the null-restricted array type is a subtype of the regular/unrestricted array type (#20305), so in particular null-restricted arrays are also instances of the unrestricted array type (just not exactly that type), which is the one named in the signature

AFAICT the real problem in #19914 (comment) is the "fixed class" here:

n6n aload gets new constraint: value 6 is fixed class 0000000001A18A00 [L...SomeValueClass;

VP sees that the component type is final, and so it creates a fixed-type constraint for the array. This is correct when the component type is not a value type (so until recently it was always correct). But now with value types having these two different array types, when the component type is a value type and the innermost array type is unrestricted, that array type now has a (strict) subtype even though its component type is final. So I'm pretty sure that VP can still use the type bound, but it just needs to detect that situation and avoid concluding that the type is fixed

[a7ehuo]

now with value types having these two different array types, when the component type is a value type and the innermost array type is unrestricted, that array type now has a subtype even though its component type is final. ... needs to detect that situation and avoid concluding that the type is fixed

@jdmpapin Thank you very much for finding this issue! I'll take a look at an appropriate fix