This is a really weird one...

While working on a backport for the new inspect-free NameError logic (#8538) I had some trouble getting the issues in #8384 to reproduce on Java 21. It reproduced fine on Java 8.

Debugging through code and setting some breakpoints, I was unable to determine why it did not call the memory-exploding inspect on the NameError object. Instead, it seemed to fall back on the basic inspect logic that just prints the name of the class and an ID for the object:

$ jruby blah.rb
generating large obj
done
NameError: undefined local variable or method `nameerror' for #<Test:0x2a04ab05>
   error at blah.rb:9
  <main> at blah.rb:16

I need to investigate this to understand why a different path was taken on Java 21.