Strict static field support fixes #23061
Conversation
theresa-m
added
comp:vm
project:valhalla
theresa-m
force-pushed
the
strictfields_7
branch
from
ad18c5a to
a9f63c3
Compare
| buildGenericSpecialStackFrame(REGISTER_ARGS, 0); | ||
| updateVMStruct(REGISTER_ARGS); | ||
| void *resolveResult = resolveStaticFieldRef(_currentThread, NULL, ramConstantPool, index, J9_RESOLVE_FLAG_RUNTIME_RESOLVE | J9_RESOLVE_FLAG_CHECK_CLINIT, NULL); | ||
| void *resolveResult = resolveStaticFieldRef(_currentThread, NULL, ramConstantPool, index, J9_RESOLVE_FLAG_RUNTIME_RESOLVE | J9_RESOLVE_FLAG_CHECK_CLINIT, NULL |
There was a problem hiding this comment.
When resolveResult is -1 (the <clinit> case where the class is in larval state ), ramStaticFieldRef is set to _currentThread->floatTemp1, the modification to classAndFlags at line 7706 won't be saved to the class/class constant pool.
I think putstatic has the same issue.
There was a problem hiding this comment.
putstatic on a strict field will not go through the clinit case, the check at line 918 should always be true. For getstatic when resolveResult is -1 the code will attempt to resolve the field again.
There was a problem hiding this comment.
putstatic on a strict field will not go through the clinit case
Is it correct to change the behaviour of <clinit> for putstatic ?
For getstatic when resolveResult is -1 the code will attempt to resolve the field again.
The code in getstatic is:
if ((void*)-1 != resolveResult) {
goto retry;
}
So when resolveResult is not -1, it will attempt to resolve again.
It can be resolved by another call of getstatic or putstatic though. But J9StaticFieldRefStrictInitRead is not recorded for this call.
|
|
||
| #if defined(J9VM_OPT_VALHALLA_STRICT_FIELDS) | ||
| if (J9ClassInitNotInitialized == (ramConstantPool->ramClass->initializeStatus & J9ClassInitStatusMask)) { | ||
| if (J9_STATIC_FIELD_STRICT_INIT_IS_UNSET(classAndFlags)) { |
There was a problem hiding this comment.
In the the <clinit> case, the classAndFlags here and in the else if below is also from ramStaticFieldRef (which is _currentThread->floatTemp1), not from the class/class constant pool.
- support fields set through the ConstantValue attribute - exclude 8th bit representing null restricted static field from type calculations - verbose messages for IllegalStateException - save unset strict field flags for final fields Signed-off-by: Theresa Mammarella <Theresa.T.Mammarella@ibm.com>
theresa-m
force-pushed
the
strictfields_7
branch
from
a9f63c3 to
d48ea80
Compare
2 participants
This fixes more tests in
test/hotspot/jtreg/runtime/valhalla/inlinetypes/verifier/StrictStaticFieldsTest.java. The final tests require the implementation of notifyStrictStaticAccess.Related: #21884