RFR: 8025856 - Fix typos in the GC code
Kirk Pepperdine
kirk at kodewerk.com
Mon Oct 14 15:48:47 UTC 2013
Hi Bengt,
I think I should have added that we're facing a regression here so there is a loss of information that can only be picked up by setting another flag. I've been spending literally years educating people about GC logging and GC flags and all I can say is that it's not easy getting flags set in prod systems. I can accept the regression for a better implementation of cause but it would be nice to stick to traditional labelling unless there is a good reason to deviate. In this case I don't see a good reason to deviate and I'm not sure how to take the "last change in wins" suggestion you've made. Avoiding this change now is exactly what we shouldn't do and no, it's doesn't make parsing across versions easier, it makes it more difficult.
Kind regards,
Kirk Pepperdine
On 2013-10-13, at 10:12 PM, Bengt Rutisson <bengt.rutisson at oracle.com> wrote:
>
> Hi Kirk,
>
> On 10/13/13 4:11 PM, Kirk Pepperdine wrote:
>> Hi,
>>
>> In gcCause.cpp I have this...
>>
>> case _java_lang_system_gc:
>> return "System.gc()";
>>
>> This use to simply be System. Again, the arbitrary change does create added parsing difficulties when dealing with different versions of GC logs. Can I request that this string be reverted back to "System" ?
>
> This code has been the same since revision 0:
>
> http://hg.openjdk.java.net/hsx/hotspot-gc/hotspot/annotate/f2110083203d/src/share/vm/gc_interface/gcCause.cpp
>
> So, to avoid the kind of log parsing issues that you mention I suggestion that we don't change this now.
>
> I think you are referring to the fact that until a change that was made 1.5 years ago the collectors were not consistent in how they logged the GC cause when the GC was started by a call to System.gc(). The ParallelGC logged just System while the others logged System.gc(). Now that all collectors use the GC cause string they are all consistent, which I think should make parsing the logs simpler not more difficult.
>
> Regards,
> Bengt
>
>>
>> Regards,
>> Kirk Pepperdine
>
More information about the hotspot-gc-dev
mailing list