OpenJDK 8292(and212)SIGSEGVinInstanceKlass::method_with_orig_idnum

[Aleksey Shipilev]

On 7/5/21 12:37 PM, Gunter Ohrner wrote:
> Am 05.07.21 um 12:13 schrieb Aleksey Shipilev:
>> On 7/5/21 12:04 PM, Gunter Ohrner wrote:
>>> #  Internal Error
>>> (/home/buildbot/worker/build-jdk8u-dev-linux/build/hotspot/src/share/vm/utilities/array.hpp:383),
>>>
>>> pid=38746, tid=0x00007f5b3d3c3700
>>> #  assert(i >= 0 && i< _length) failed: oob: 0 <= -32023 < 33514
>>> #
>>
>> Yeah, that looks like https://bugs.openjdk.java.net/browse/JDK-8194246
>> -- on a hunch, I did the 8u backport due diligence this morning, and
>> now the fix waits for 8u maintainers approval
> 
> So my options would be to
> 
> a) wait for a fixed u release

Yes, assuming that was the issue. I would like to ask you to run with the new nightly binary to 
confirm. Anyhow, unfortunately, it is too late to fix for July release, and so it would have to wait 
for October.

> b) avoid an Exception to happen in this situation
> Does Gradle introduce classes with overly many methods into the call
> stack? All other classes on the stack should already have been there
> before...

Hard to guess. I would suspect that the generated code shape depends on the environment. If you were 
able to build the OpenJDK by yourself, you could probably print out the problematic class name on 
the failing path.

There is also another option:

  c) Migrate to 11u that has this fix already; not sure how viable is that for you.

-- 
Thanks,
-Aleksey