In line with this good advice, I've opted to preserve the test in 7/8:
https://hg.python.org/jython/rev/dc1ffb710882
While the mode of traversal is available to users, we should test it (which we do since we run on Java 7 onwards). But maybe --add-opens is what we should have here, once we do not support 8?
On the topic of --add-opens, and going back to "... thinking of a way to preserve the option python.security.respectJavaAccessibility=false", this change provides an example of a test in which I chose to let the illegal reflective access go ahead, but tell java to expect it:
https://hg.python.org/jython/rev/677009e86ce2
I have also fixed the warnings that arise from codec in:
https://hg.python.org/jython/rev/5820f8a915f1
Here, the "respectJavaAccessibility" had been made unnecessary by earlier work (or was never necessary, maybe). The test had a number of unrelated problems that I've fixed too.
The only reflective access warnings now are from jnr.posix, and what I called "fundamental". That is, we explicitly ask for information Java just doesn't give up any other way. |
