[qtcomp] Random crash in Mir input when running AP tests: [terminate called after throwing an instance of '...' what(): assign: File exists] when constructing a mir::AsioMainLoop::FDHandler

The crash happens because the autopilot test sends key events very fast, one right next to the other. If we change it so that it waits a bit before generating the next key event, there's no crash:

--- /usr/lib/python2.7/dist-packages/ubuntu_html5_ui_toolkit/tests/__init__.py 2014-07-23 13:16:36.000000000 +0000
+++ /usr/lib/python2.7/dist-packages/ubuntu_html5_ui_toolkit/tests/__init__modified.py 2014-07-23 13:17:28.000000000 +0000
@@ -187,7 +187,7 @@
         addressbar = self.get_addressbar()
         self.assertThat(addressbar.activeFocus, Eventually(Equals(True)))

- self.keyboard.type(url, 0.001)
+ self.keyboard.type(url, 0.2)

         self.pointer.click_object(self.get_webview())
         time.sleep(1)

With that patch there's no crash but the typing is too slow (to the tests take way longer to complete). Some inbetween value should still work. Haven't tried any other though.

This crash is caused through interleaved register and unregister calls of fds. mirserver unregisters the FD as soon as the surface disappears or when all outstanding input responds have been received. The exception is thrown when the registration of an FD happens before the old FDHandler has been removed.

For this bug to work the following things are necessary:
 * user input is handled and dispatched in multiple threads - i.e. input responds are currently handled in main loop (we discussed about moving that into a separate thread - would be no difference since input is read in a dedicated thread)
 * user input and scene reconfigurations happen in multiple threads - imagine a surface gets removed while a user input event is sent
 * fd registration happens out of order in the callers thread, while the removals happen in main thread.

I am tempted to make fd registration like the removal, it would solve that issue but that might not guarantee the right order and have different side effects.

Fix committed into lp:mir/devel at revision None, scheduled for release in mir, milestone Unknown

This bug was fixed in the package mir - 0.5.1+14.10.20140728-0ubuntu1

---------------
mir (0.5.1+14.10.20140728-0ubuntu1) utopic; urgency=medium

  [ Kevin Gunn ]
  * Fixed: Crash due to racing input registration & surface removal
    (LP: #1346952)

  [ Ubuntu daily release ]
  * New rebuild forced
 -- Ubuntu daily release <email address hidden> Mon, 28 Jul 2014 02:49:50 +0000