RFR: 8253495: CDS generates non-deterministic output [v2]

[Thomas Stuefe]

On Thu, 10 Mar 2022 19:34:29 GMT, Ioi Lam <iklam at openjdk.org> wrote:

>> src/hotspot/share/prims/jvm.cpp line 2887:
>> 
>>> 2885:     return;
>>> 2886:   }
>>> 2887: #endif
>> 
>> Should we do this for jni_AttachCurrentThread too?
>
> This hasn't been necessary for me because jni_AttachCurrentThread is not called during "java -Xshare:dump", which executes under a very strict condition and doesn't normally allow arbitrary JNI libraries to be loaded.

Is reproducibility also a topic for users calling -Xdump with custom JNI coding? Or maybe having the VM instrumented somehow? Since it seems such an easy fix, I would prevent attaching too. At least the user would get a clear error message.

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

PR: https://git.openjdk.java.net/jdk/pull/7748