[15] RFR (S): 8247502: PhaseStringOpts crashes while optimising effectively dead code

[Tobias Hartmann]

+1

Thanks for taking care of this while I'm on vacation!

Best regards,
Tobias


On 11.07.20 01:19, Vladimir Kozlov wrote:
> I agree with this small fix.
> 
> Thanks,
> Vladimir
> 
> On 7/10/20 9:26 AM, Vladimir Ivanov wrote:
>> https://bugs.openjdk.java.net/browse/JDK-8247502
>> http://cr.openjdk.java.net/~vlivanov/8247502/webrev.00/
>>
>> As Tobias discovered, PhaseStringOpts crashes when it encounters String::append() argument being
>> TOP: TOP is a constant, but the code expects to see a String constant instead.
>>
>> It happens while processing a call in unreachable infinite loop. The code is effectively dead, but
>> IGVN and PhaseRemoveUseless don't see that. It is discovered later when loop opts kick in which
>> clean it up.
>>
>> Proposed fix tries to make the code more robust and just bails out the optimization when TOP is
>> encountered.
>>
>> Alternative way to fix the problem would be to clean up the graph before PhaseStringOpts (e.g., by
>> running PhaseIdealLoop(LoopOptsNone) since PhaseRemoveUseless is not enough), but PhaseIdealLoop
>> pass can be expensive. So, I'm in favor of the local fix in PhaseStringOpts.
>>
>> Testing: crash reproducer, hs-precheckin-comp, hs-tier1, hs-tier2, tier1
>>
>> Thanks!
>>
>> PS: no regression test since I wasn't able to extract a simple reproducer from the crash log.
>>
>> Best regards,
>> Vladimir Ivanov