On Tue, 4 Jun 2024 13:02:18 GMT, Alexey Semenyuk <asemen...@openjdk.org> wrote:
>> Fix MainClassTest class to use HelloApp.AppOutputVerifier class to run app >> launcher instead of raw Executor. This makes MainClassTest test run app >> launchers with retries. This change addresses the primary issue. >> >> Fix inconsistencies in HelloApp.AppOutputVerifier class. It used to provide >> API allowing to run launchers without retries. It inconsistently allowed the >> execution of launchers with suppressed output (stdout and stderr). It >> inconsistently executed launchers with/without PATH removed from the >> environment. >> >> These loopholes were eliminated: >> >> - stdout and stderr of app launchers is never suppressed; >> - PATH env variable is always deleted for app launchers on Windows. It is >> not deleted on other platforms. This change sets the correct scope of >> [JDK-8254920](https://bugs.openjdk.org/browse/JDK-8254920) fix that >> introduced the removal of PATH env variable for app launchers; >> - app launchers are always executed with retries unless the launcher is >> executed with `jpackage.test.noexit` system property set to `true` >> indicating the test app will not terminate on its own. >> >> Other changes are due to changes in HelloApp.AppOutputVerifier class. > > The crash is sporadic, hard to reproduce, and happens only on Linux. > Almost every jpackage test runs a launcher, so the alternative to "hiding" > the issue was to disable all jpackage tests or make them not run launchers on > Linux. There is no good solution except nailing and fixing the root cause. @alexeysemenyukoracle Would have been good to rename this JBS issue then to something like "rerun tests in case of crashes", since the issue, from the outside, looks like a fix for the crash itself. This is mostly to make our lives easier when it comes to backporting (which may happen years from now, when everyone forgot what this was) ------------- PR Comment: https://git.openjdk.org/jdk/pull/19502#issuecomment-2147486733
