RFR: Workaround CacheDataStore use with source launcher [v2]

[Aleksey Shipilev]

On Fri, 6 Sep 2024 08:59:54 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

>> I was trying to test the simplest workload that involves javac, and that is source launcher. When I attempt to use source launcher with `-XX:CacheDataStore`, it fails with:
>> 
>> 
>> $ rm -f app.cds*; build/linux-x86_64-server-release/images/jdk/bin/java -XX:CacheDataStore=app.cds -Xmx256m -Xms256m -XX:+UnlockExperimentalVMOptions -XX:+UseEpsilonGC HelloStream.java
>> [0.001s][warning][cds] optimized module handling/full module graph: disabled due to incompatible property: jdk.module.addmods=ALL-DEFAULT
>> Error occurred during initialization of VM
>> CacheDataStore cannot be created because AOTClassLinking is enabled but full module graph is disabled
>> 
>> 
>> Source launcher adds `--add-modules=ALL-DEFAULT`:
>> 
>> 
>> $ man java
>> ...
>>        In source-file mode, execution proceeds as follows:
>> 
>> ...
>>        • The compiled classes are executed in  the  context  of  an  unnamed  module,  as  though  --add-mod‐
>>          ules=ALL-DEFAULT  is  in  effect.  This is in addition to any other --add-module options that may be
>>          have been specified on the command line.
>> 
>> 
>> I think we can work that around specifically for Leyden, and also leave a TODO breadcrumb in the code that we need to figure this out on CDS side. This PR does that workaround. With it, source launcher works with simple examples:
>> 
>> 
>> $ build/linux-x86_64-server-release/images/jdk/bin/java -XX:CacheDataStore=app.cds -Xmx256m -Xms256m -XX:+UnlockExperimentalVMOptions -XX:+UseEpsilonGC  HelloStream.java
>> Warning: CacheDataStore cannot be used with default source launcher environment, disabling --add-modules=ALL-DEFAULT
>> hello, world
>
> Aleksey Shipilev has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Indenting

I think what Leyden _targets_ is a different question.

After a day of playing with this, I think enabling source launcher does look very convenient for current performance testing, as it includes javac automatically, without any additional setup. I am not insisting on doing this workaround, it is just a handy development aid. If you all feel strongly against it, I can go the longer and less convenient route to measure CDS perf.

-------------

PR Comment: https://git.openjdk.org/leyden/pull/15#issuecomment-2334640218