RFR 8043138: Attach API should not require jvmstat rmi protocol

[Jaroslav Bachorik]

On 25.11.2015 10:02, Alan Bateman wrote:
>
> On 24/11/2015 16:03, Jaroslav Bachorik wrote:
>> Please, review the following change
>>
>> Issue : https://bugs.openjdk.java.net/browse/JDK-8043138
>> Webrevs:
>> * top level: http://cr.openjdk.java.net/~jbachorik/8043138/webrev.00/top
>> * jdk: http://cr.openjdk.java.net/~jbachorik/8043138/webrev.00/jdk
>>
>> This patch splits up the jdk.jvmstat module to jdk.jvmstat and
>> jdk.jvmstat.rmi to make the basic jvmstat functionality available
>> without requiring dependencies on RMI.
>>
>> The split is pretty straightforward - all the RMI dependent
>> implementation is moved to the new module as well as 'jstatd'
>> implementation.
>>
>> The change requires changes in the makefile (for merging
>> META-INF/services resources from jdk.jvmstat and jdk.jvmstat.rmi) and
>> therefore I am posting this request also to the build-dev mailing list.
>
> The main question here is whether the module should be named
> jdk.jvmstat.remote or jdk.jvmstat.rmi. I could imagine other protocols
> for remote access to jdk.jvmstat.rmi seems right, I just wonder if the
> RemoteHost and RemoteVM should be in package sun.jvmstat.monitor.rmi
> instead of .remote.

I don't think we can just repackage these interfaces - they are remote 
API and we would break compatibility (eg. JDK 8 jvmstat would not be 
able to query JVM exposed via JDK 9 jstatd).

>
> In the top-level modules.xml then I assume that jdk.jvmstat should no
> longer depend on java.rmi.

Right. I missed this one. Will fix.

>
> Should jdk/make/launcher/Launcher-jdk.jvmstat.gmk be renamed/changed
> into Launcher-jdk.jvmstat.rmi.gmk?

Possible. What is the rule for naming these make files? Are they named 
after the modules or after the launchers they actually build (in this 
case 'jstatd').

-JB-

>
> -Alan
>
>