RFR 8214689 [lworld][c1] aaload cannot handle unloaded class
tobias.hartmann at oracle.com
Tue Dec 11 10:17:33 UTC 2018
On 11.12.18 06:34, Ioi Lam wrote:
Looks good to me!
I would set a default false value for the "is_value_type" argument of get_unloaded_klass() so you
don't have to explicitly pass "false" at all these places.
> My next step is to fix https://bugs.openjdk.java.net/browse/JDK-8215201, so we can actually
> recompile on deoptimization.
I think we can just go with Action_make_not_entrant for now. It will invalidate the code and
recompile if the method is still hot. Long term, we need a better solution anyway (probably a
runtime call) because C1 code is likely to see unloaded classes and frequent deopts are just too
expensive. Also, most likely array covariance will come back and we then need a way to handle the
Object array access case (flattened value type arrays may "hide" as Object arrays).
More information about the valhalla-dev