RFR(L): JDK-8222525 : [lworld] Runtime support for LW2 Arrays

[Tobias Hartmann]

Hi,

On 24.04.19 20:54, Frederic Parain wrote:
> ciReplay.cpp:916:
>   Could the resolution fail? If yes, what would be the consequences?

Yes, we print an error message and stop replay compilation if ReplayIgnoreInitErrors is not set:
http://hg.openjdk.java.net/jdk/jdk/file/fe9f33555993/src/hotspot/share/ci/ciReplay.cpp#l427

This can already happen when resolving an 'L' array or object:
http://hg.openjdk.java.net/jdk/jdk/file/fe9f33555993/src/hotspot/share/ci/ciReplay.cpp#l849
http://hg.openjdk.java.net/jdk/jdk/file/fe9f33555993/src/hotspot/share/ci/ciReplay.cpp#l891

And it actually does happen frequently when trying to replay compilation without all the dependent
classes on the classpath. If ReplayIgnoreInitErrors is set, we might crash but that's okay because
replay compilation is just a debugging tool that is very powerful if it works but does not work in
many cases for various reasons (for example, dynamically generated bytecodes).

Best regards,
Tobias