[g-a-devel] GUI Restrictions on an Accessibility App?

["Quiring, Sam" <Sam Quiring windriver com>]

Greetings,

 

To date, our at-spi applications do not have GUI interfaces, but we are now trying to add one.  I seem to be experiencing some restrictions and am wondering if there are certain rules that need to be followed?

 

Our app runs the SPI event loop in the main thread, so during startup the Gtk+ GUI and its event loop are started in a new thread.  The effective end of the main thread looks like this:

 

createListeners();

registerListeners();

SPI_event_main();

SPI_deregisterAll(genericListener);

 

If execution of the main thread makes it past registerListeners() before the GUI is initialized, then the GUI never comes up, it hangs in the call to show the app's main window: gtk_widget_show(window).  If I stop the main thread before registerListeners() and wait until gtk_widget_show(window) returns, then the GUI comes up and mostly works.  registerListeners just consists of a series of calls on SPI_registerGlobalEventListener, SPI_registerDeviceEventListener, and SPI_registerAccessibleKeystrokeListener.

 

Does anyone have an explanation for this behavior?  What is it in the SPI_register* routines that would cause gtk_widget_show(window) to hang?

 

Sometimes, when the GUI fails to initialize and the app tries an orderly shutdown, I get this error:

(<unknown>:10225): Bonobo-CRITICAL **: bonobo_main_quit: assertion `bonobo_main_loops != NULL' failed

-Sam

 

PS: Here's a possibly related fact.   Right now the GUI has three buttons.  Two just print a message. If I click these two over and over I see their messages.  After 8 seconds a timer kills the entire process by calling SPI_event_quit and the process exits normally: SPI_event_main returns and the listeners are all deregistered.

 

The third button, "exit", exits the GUI by calling gtk_widget_destroy(window) and gtk_main_quit.  The GUI goes away.  However when the timer tries to kill the process something goes wrong.  A message that is supposed to be displayed in the terminal with a printf is not displayed, even though logging events that happen before and after the printf are logged.  SPI_event_main() (shown above) returns, but the call on SPI_deregisterAll hangs.