RFR: 8014514: Fix jvm args for sjavac

Erik Joelsson erik.joelsson at oracle.com
Wed May 15 11:31:55 UTC 2013



On 2013-05-15 09:01, Erik Joelsson wrote:
>
>
> On 2013-05-15 06:21, David Holmes wrote:
>> On 15/05/2013 12:34 AM, Erik Joelsson wrote:
>>> This patch tries to be smarter when figuring out optimal jvm args for
>>> sjavac, especially if the jvm used is 32bit.
>>>
>>> http://cr.openjdk.java.net/~erikj/8014514/webrev.root.01/
>>>
>>> For this to work properly, two patches to the sjavac src langtools are
>>> also needed, but they will be handled in separate bugs and code 
>>> reviews.
>>
>> can't really comment but does sjavac really need to mess with these:
>>
>> ! 
>> ADD_JVM_ARG_IF_OK([-XX:PermSize=32m],SJAVAC_SERVER_JAVA,[$SJAVAC_SERVER_JAVA])
>> ! 
>> ADD_JVM_ARG_IF_OK([-XX:MaxPermSize=160m],SJAVAC_SERVER_JAVA,[$SJAVAC_SERVER_JAVA])
>> ! 
>> ADD_JVM_ARG_IF_OK([-XX:ThreadStackSize=$STACK_SIZE],SJAVAC_SERVER_JAVA,[$SJAVAC_SERVER_JAVA])
>>
> I honestly don't know. Fredrik added those way back in the beginning 
> when Hotspot had trouble running the javac server. I know he got much 
> better performance with JRockit as the server compiler runtime at the 
> time. I will remove them and run another jprt job and see if something 
> nasty happens. Getting rid of it would be nice.
>
It worked just as well so I will declare those args as no longer needed 
legacy and remove them. New webrev:

http://cr.openjdk.java.net/~erikj/8014514/webrev.root.02/

/Erik



More information about the build-dev mailing list