[Gradle] Use processIsolation instead of classLoaderIsolation

[ru5j4r0]

Hi.
First of all, I appreciate your great work.
This is my first PR for the Play projects, so sorry in advance if I'm doing something wrong.

I was trying to migrate a few Play Framework projects to Gradle, but I couldn't migrate some of them because compileTwirl task gave me an out of memory error of JVM's Metaspace.
I was able to migrate a project that has dozens of views, but I couldn't which has around 1,000 views.
My PC has 32GB RAM, but increasing Xmx nor MaxMetaspaceSize didn't fix the issue.

After searching for a bit, I found this Gradle's issue.
I forked the Twirl repo, replaced classLoaderIsolation with processIsolation, published it to mavenLocal() and using that SNAPSHOT version of the Twirl plugin from my local Maven repo actually fixed the problem.

I just filed a PR right away because it's a small change.
What are your thoughts on this change?

Possibly related

• Gradle Plugin: Documentation of a few gotchas #745
  • Specifically about OOM part

Other projects doing the same fix

• sigstore-java
• ktfmt-gradle

[ru5j4r0]

I suppose the drawback of using processIsolation is a increased overhead due to the need to start a new process?
Perhaps I can add an option to choose the isolation method, in case replacing classLoaderIsolation with processIsolation by default is not acceptable.

[ihostage]

Hi, @ru5j4r0!

I suppose the drawback of using processIsolation is a increased overhead due to the need to start a new process?

It's okay, don't worry. I will investigate this issue in more detail next week.

[ihostage]

Unfortunately, I couldn't find a way to write a unit test for this case 😞
But I've made a lot of experiments locally and I agree with the proposed changes.
Thank you very much, @ru5j4r0, and apologies for the delay!

[ru5j4r0]

@ihostage
I’m the one who owes you an apology for opening a PR without tests or reproducible examples. Thank you for all your work!