On Thu, 17 Jul 2025 01:32:52 GMT, David Holmes <dhol...@openjdk.org> wrote:
> Okay, but are there other tests we have missed here? I'd say this tests was missed in https://bugs.openjdk.org/browse/JDK-8354461 fix. > > I'm wondering what the value of "streaming output" is when it can lead to > deadlocks? Is this just a test problem? "Streaming output" means attach operation output is not buffered, but is sent to the client while the handler produce it. It's useful in real cases, when user runs jcmd or other tool against external target VM (see https://bugs.openjdk.org/browse/JDK-8319055 and related issues). In tests we often use scenarios when test attaches to itself or like in this case use attach tool to attach to the test process. And we can get deadlock because producer (operation handler, often executed in safepoint) and consumer (test reads redirected output of the tool) are executed in the same process. It's just test problem (to reproduce real scenario tests should run target VM as a separate process) ------------- PR Comment: https://git.openjdk.org/jdk/pull/26358#issuecomment-3085091259
