On Fri, 14 Jan 2022 19:09:53 GMT, Phil Race <p...@openjdk.org> wrote:
> Some more signal handler related warning strings from the VM need to be > excluded from what this test considers a failure > > See the bug for more info. In hotspot, we want to explicitly ignore SIGPIPE and SIGXFSZ (see https://bugs.openjdk.java.net/browse/JDK-4229104 and https://bugs.openjdk.java.net/browse/JDK-6499219). But we don't do SIG_IGN, since we still need to give chained handlers the opportunity to handle them. See https://github.com/openjdk/jdk/blob/fef8f2d30004ea75c663752ae6c61b214fb8abc8/src/hotspot/os/posix/signals_posix.cpp#L618-L623. So now, from the OS perspective, we handle SIGPIPE. We just don't do anything but return from the signal handler (ignoring chained handlers for now). If someone outside sets SIGPIPE to SIG_IGN, then we get subtle changes: - IO calls would now yield EPIPE, whereas now I am not sure what happens? Does the IO call report success? - Also, chained handlers would not get called. It's all pretty unexciting. Nothing would crash. There may be an argument for skipping the test on SIGPIPE if it is SIG_IGN and we don't have chained handlers. If it seems such a standard thing to do for third party libraries. Idk. Maybe open a separate bug? ------------- PR: https://git.openjdk.java.net/jdk/pull/7091
