1 files changed, 57 insertions, 0 deletions
diff --git a/ipc/glue/test/browser/browser_utility_hard_kill_delayed.js b/ipc/glue/test/browser/browser_utility_hard_kill_delayed.js
new file mode 100644
index 0000000000..ffda4d3988
--- /dev/null
+++ b/ipc/glue/test/browser/browser_utility_hard_kill_delayed.js
@@ -0,0 +1,57 @@
+/* Any copyright is dedicated to the Public Domain.
+ * http://creativecommons.org/publicdomain/zero/1.0/ */
+
+"use strict";
+
+add_task(async () => {
+  const utilityPid = await startUtilityProcess();
+
+  SimpleTest.expectChildProcessCrash();
+
+  const utilityProcessGone = TestUtils.topicObserved("ipc:utility-shutdown");
+
+  info("Hard kill Utility Process");
+  const ProcessTools = Cc["@mozilla.org/processtools-service;1"].getService(
+    Ci.nsIProcessToolsService
+  );
+
+  // Here we really want to exercise the fact that kill() might not be done
+  // right now but a bit later, and we should wait for the process to be dead
+  // before considering the test is finished.
+  //
+  // Without this, we get into bug 1754572 (where there was no setTimeout nor
+  // the wait) where the kill() operation ends up really killing the child a
+  // bit after the current test has been finished ; unfortunately, this happened
+  // right after the next test, browser_utility_memoryReport.js did start and
+  // even worse, after it thought it had started a new utility process. We were
+  // in fact re-using the one we started here, and when we wanted to query its
+  // pid in the browser_utility_memoryReport.js then the kill() happened, so
+  // no more process and the test intermittently failed.
+  //
+  // The timeout value of 50ms should be long enough to allow the test to finish
+  // and the next one to start and get a reference on the process we launched,
+  // and yet allow us to kill the process in the middle of the next test. Higher
+  // values would allow browser_utility_memoryReport.js to complete without
+  // reproducing the issue (both locally and on try).
+  //
+  // eslint-disable-next-line mozilla/no-arbitrary-setTimeout
+  setTimeout(() => {
+    ProcessTools.kill(utilityPid);
+  }, 50);
+
+  info(`Waiting for utility process ${utilityPid} to go away.`);
+  let [subject, data] = await utilityProcessGone;
+  ok(
+    subject instanceof Ci.nsIPropertyBag2,
+    "Subject needs to be a nsIPropertyBag2 to clean up properly"
+  );
+  is(
+    parseInt(data, 10),
+    utilityPid,
+    `Should match the crashed PID ${utilityPid} with ${data}`
+  );
+
+  // Make sure the process is dead, otherwise there is a risk of race for
+  // writing leak logs
+  utilityProcessTest().noteIntentionalCrash(utilityPid);
+});