RFR: Verifier "After Mark" should assert matrix consistency
Aleksey Shipilev
shade at redhat.com
Fri Aug 4 12:03:21 UTC 2017
On 08/04/2017 01:59 PM, Roman Kennke wrote:
> Am 04.08.2017 um 13:04 schrieb Aleksey Shipilev:
>> I think we are missing a crucial bit of matrix verification after concurrent mark:
>> http://cr.openjdk.java.net/~shade/shenandoah/verifier-aftermark-matrix/webrev.01/
>>
>> It crashes our current tests today, so I would first pinpoint the bug itself.
>>
>> Thanks,
>> -Aleksey
>>
> As discussed on IRC, we don't bother keeping matrix up-to-date outside
> of UR. We currently need UR before partial cycle, and thus rebuild the
> matrix during UR.
>
> Any crazy combination of non-partial heuristics (i.e. adaptive) and
> manually setting +UseShenandoahMatrix is tricky to verify because it may
> run into piggy-backed UR, where we don't keep matrix updated.
Agreed. My verifier change seems incorrect here then.
-Aleksey
More information about the shenandoah-dev
mailing list