[GR-47922] Enforce -H:±UnlockExperimentalVMOptions in CI.

[graalvmbot]

This PR sets NATIVE_IMAGE_EXPERIMENTAL_OPTIONS_ARE_FATAL=true in CI to ensure that gates are properly unlocking experimental options. The PR also adjusts various mx/CI infrastructure bits and fixes a bug in the unlocking mechanism itself.

[zakkak]

Hi @fniephaus, although this makes sense as the next step we still see 3rd party libraries using experimental VM options, e.g. netty/netty#13595 and neo4j/neo4j-java-driver#1498, so I expect significant disruption if this goes in now.

I am afraid that would possibly force frameworks to just pass -H:+UnlockExperimentalVMOptions in the beginning of the native-image arguments, which is not desirable.

[zakkak]

Furthermore I don't see the PR resolving the "Bug fixes and improvements" and "API changes" parts of #7370 is that intentional are they resolved in separate PRs?

[fniephaus]

Hey @zakkak, are you suggesting that we enforce unlocking in a later release of GraalVM? The JDK 22 release is half a year away from the JDK 21 release (that's where we introduced -H:+UnlockExperimentalVMOptions) , and the JDK 23 release is a year away.

[fniephaus]

Furthermore I don't see the PR resolving the "Bug fixes and improvements" and "API changes" parts of #7370 is that intentional are they resolved in separate PRs?

They will be resolved in separate PRs.

[zakkak]

Hey @zakkak, are you suggesting that we enforce unlocking in a later release of GraalVM? The JDK 22 release is half a year away from the JDK 21 release (that's where we introduced -H:+UnlockExperimentalVMOptions) , and the JDK 23 release is a year away.

Not necessarily, I would suggest delaying the merge of the PR for the time being to see how the ecosystem adapts. We still have enough time for the code freeze, right?

They will be resolved in separate PRs.

In that case it would make more sense to merge those PRs first, no?

[fniephaus]

Not necessarily, I would suggest delaying the merge of the PR for the time being to see how the ecosystem adapts. We still have enough time for the code freeze, right?

Sure, code freeze is in December, and the freeze is much longer (3 months as opposed to one month). Keep in mind that doing this now will only affect end-users that use a dev build.

In that case it would make more sense to merge those PRs first, no?

It depends. On #7370, we said that enabling this has the highest priority.

[fniephaus]

I have adjusted this PR so that it doesn't change the behavior just yet. Instead, we use NATIVE_IMAGE_EXPERIMENTAL_OPTIONS_ARE_FATAL=true in CI.