RFR: 8231558: [macos] Platform.exit causes assertion error on macOS 10.15 or later

[Pankaj Bansal]

On Tue, 22 Jun 2021 20:42:48 GMT, Kevin Rushforth <kcr at openjdk.org> wrote:

> This is a fix for the assertion error message that is printed to the console on macOS 10.15 or later when an application calls `Platform.exit` while a `Stage` is showing.
> 
> The root cause is a latent bug in the JavaFX glass code that was revealed by an apparent change of behavior in macOS. A few of the object deallocation methods, which are called by the Objective C auto-release mechanism, use the standard `GET_MAIN_JENV` macro to get the JNI environment. The macro will print an assertion warning if Java has been detached. I instrumented the code and can see that `GlassViewDelegate::dealloc` is now called after the `GlassApplication` main loop has detached Java. Since we don't control when the dealloc method is called, it is not correct to do the assertion check in those cases. Some of the dealloc methods already skip this assertion check by grabbing the jEnv pointer directly, so we need to fix the others. I added a new variant of the macro called `GET_MAIN_JENV_NOWARN` with a comment indicating that is suitable for use by the dealloc methods.
> 
> In addition to verifying that the test program attached to JBS now exits cleanly with no assertion failure message, I added an automated system test that fails on macOS before the fix and passes after the fix. On other platforms it passes already.

The fix looks fine. The attached test case fails without the fix and passes after the fix. I also ran the full test on Mac 10.15 and I don't see issue.

tests/system/src/test/java/test/launchertest/Constants.java line 70:

> 68:     static final int ERROR_ASSERTION_FAILURE = 28;
> 69: 
> 70:     static final int ERROR_TIMEOUT = 28;

Should not this be assigned error code 29 instead of 28?

-------------

PR: https://git.openjdk.java.net/jfx/pull/540