RFR: 8339488: Extended NPE message doesn't handle CONSTANT_Dynamic

[Chen Liang]

On Fri, 6 Sep 2024 11:22:50 GMT, Oli Gillespie <ogillespie at openjdk.org> wrote:

> Extended NPE messages require walking the bytecode in the method where it was created, and simulating the effect on the stack. `ldc` for dynamic constants is not handled, which can lead to assertion failures in debug builds, or incorrect NPE messages in release builds (possibly worse with the right conditions?).
> 
> Simply handle dynamic constants by figuring out their return type so we know how they will affect the stack.
> 
> The new test fails without the fix in both release and debug builds. In release, it fails because the NPE message is wrong: `Exception in thread "main" java.lang.NullPointerException: Cannot invoke "Object.notify()" because "`, and in debug builds it hits this assert:
> 
> 
> #
> # A fatal error has been detected by the Java Runtime Environment:
> #
> # Internal Error (<redacted>/src/hotspot/share/interpreter/bytecodeUtils.cpp:660), pid=12677, tid=12678
> # assert(false) failed: Unexpected tag

A few questions:
1. Can the reproduction case be as simple as:

ldc Dynamic(BootstrapMethod(ConstantBootstraps::nullConstant) NameAndType(ConstantDescs.DEFAULT_NAME, "Ljava/lang/Object;"))
invokevirtual Methodref(Class("java/lang/Object") NameAndType("notify" "()V"))

2. Is there a preference of ASM/Java generator vs hardcoded jasm? Think it's better to use one file for such a small test.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/20889#issuecomment-2333925354