Fix MemberName fields after Fast HCR and FSD

[EricYangIBM]

fixMemberNames called after HCR to do a heap walk to fix all MemberName objects' vmtarget. MemberName->clazz and MemberName->vmindex have already been fixed by existing mechanisms.

• If the object is a fieldID, set vmtarget to the field offset + modifiers. If it is a method/constructor ID, set vmtarget to the ID's method
• Move needed MemberName macros to j9nonbuilder.h
• Define struct consisting of J9VMThread * and void * to be passed into the callback function as it needs the current thread and the class hash table

Issue: #11528
Signed-off-by: Eric Yang eric.yang@ibm.com

[EricYangIBM]

Currently, vmtarget updates with the old IDs, even if they are outdated. Perhaps in the future we could add some handling to improve UX.

[EricYangIBM]

Changes build successfully.

[fengxue-IS]

minor formating suggestions, rest lgtm

[gacholio]

I'm not sure this is doing what you expect - is the purpose of this to fix references from instance of MemberName to fields/methods of classes which have been redefined?

[EricYangIBM]

I'm not sure this is doing what you expect - is the purpose of this to fix references from instance of MemberName to fields/methods of classes which have been redefined?

It fixes the vmtarget field of every MemberName object using its already up to date vmindex field (updated by fixJNIRefs) and clazz field.

[gacholio]

Also, please resolve any conversations that have been completed.

[EricYangIBM]

@gacholio Done.
Also, may I ask why the intermediate UDATA cast when casting between pointers and ints? I followed existing code and it didn't have that cast. E.g.

openj9/runtime/jcl/common/java_lang_invoke_MethodHandleNatives.cpp

Line 203 in 5530af2

J9OBJECT_U64_STORE(currentThread, membernameObject, vm->vmindexOffset, (U_64)vmindex);

[gacholio]

It's not legal to cast between pointers and integers whose size does not match the size of a pointer. For the U64 case, this would fail to compile on some 32-bit platforms without the intermediate cast.

The line you cited is not casting to a pointer.

[gacholio]

jenkins test sanity zlinux jdk8

[gacholio]

jenkins test sanity winojdk292 jdk16 depends ibmruntimes/openj9-openjdk-jdk16#29

[gacholio]

@babsingh Please triage the failures.

[babsingh]

https://openj9-jenkins.osuosl.org/job/Test_openjdk16_j9_sanity.functional_x86-64_windows_ojdk292_Personal/3/

@gacholio In the above test build, SharedCPEntryInvokerTests test targets successfully pass: #11528 (comment). All failures are known issues. This PR is good to merge.

[EricYangIBM]

@gacholio I had not squashed the commits when you merged. The commits/commit messages I made are not the most meaningful. Do you want this to be fixed?

[gacholio]

I think we may as well let it go this time (wish I had seen that before the merge). The only way to "correct" this would be to revert this and use a new PR with a single commit, which would just be even more churn.