Restore stdio in test rig

[som-snytt]

Save and restore stdio in the test rig child process.

Also add a test from the duplicate ticket for 23245.

[hamzaremmal]

Maybe there's no more skipping...

There is no more skipping.

[som-snytt]

Added a bit of output for more stack trace. The test will always fail, to make re-running easier.

My first guess was NPE in clinit. My second guess was a bad "nullable field" getting nulled after lazy val init; perhaps the outer pointer itself, which was used only for the lazy init. But the outer is not a class param in the example; its rhs is just a static module. Maybe the trait init? but that is empty. Maybe a class isn't initialized? Maybe a missing inner class, except the module is top level.

I could not replicate locally (WSL) but interacting with the test rig made me want to contribute improvements, as I intended.

I see that the flaky test is not included in the group with -Ylegacy-lazy-vals, which was my fourth guess.

[som-snytt]

Woo-hoo!

Output from 'tests/run/i23245a' did not match check file. Actual output:
java.lang.NullPointerException: Cannot invoke "java.io.PrintStream.println(String)" because "java.lang.System.err" is null
	at logadapter.LogAdapter.info(api.scala:23)
	at Test$.main(test_2.scala:3)
	at Test.main(test_2.scala)
	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:77)
	at java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
	at java.base/java.lang.reflect.Method.invoke(Method.java:569)
	at dotty.tools.vulpix.ChildJVMMain.runMain(ChildJVMMain.java:40)
	at dotty.tools.vulpix.ChildJVMMain.main(ChildJVMMain.java:47)

Oh wait, does that make any sense? I'll look at it later.

[som-snytt]

The test for #2781 was aware that nulling stdio streams is unfriendly, but did it anyway, in the interests of science.

Any test thereafter in the same runner JVM that used System.err, including the default exception handler in doing e.printStackTrace(), would throw NPE.

This doesn't seem to be a rampant problem, but for safety, the ChildJVMMain that communicates with the test rig now saves and restores stdio.

Equivalent in partest:

https://github.com/scala/scala/blob/v2.13.18/src/partest/scala/tools/partest/nest/Runner.scala#L242

[hamzaremmal]

why branch here? we can just call System.setErr regardless of if it changed or not.

[som-snytt]

Just to be finicky. I needed a block for a println while verifying, and then I decided to avoid any side-effects, however benign, of calling the setter.

Most of my time was spent formatting the Java code.

Worth adding that this class could be written in Scala.

[som-snytt]

Looks like I have an opportunity edit it again anyway. I requested your review because of your proximate interest in the build passing correctly. By proximate, I mean near and dear to your heart.