RFR: 8273165: GraphKit::combine_exception_states fails with "matching stack sizes" assert [v2]

Tobias Hartmann thartmann at openjdk.java.net
Wed Sep 1 14:43:00 UTC 2021


On Wed, 1 Sep 2021 14:23:10 GMT, Vladimir Ivanov <vlivanov at openjdk.org> wrote:

>> The fix for JDK-8271276 uncovered another problem with incremental inlining through 
>> virtual call sites: receiver null check breaks exception state combining in `GraphKit::replace_call()`
>> because the associtated exception path has the JVM state representing the point
>> right before the call (arguments are on stack).
>> 
>> I propose a conservative fix which bails out the inlining attempt when receiver is not provably non-null.
>> 
>> IMO the proper fix is to always add explicit receiver null check and teach
>> `Block::implicit_null_check()` about CallDynamicJava nodes. But that's for a
>> separate change.
>> 
>> Testing: failing tests, hs-tier1 - hs-tier4
>
> Vladimir Ivanov has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Migrate to Type::maybe_null()

Looks good.

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

Marked as reviewed by thartmann (Reviewer).

PR: https://git.openjdk.java.net/jdk/pull/5330


More information about the hotspot-compiler-dev mailing list