[g-a-devel]exceptionally bad behavior ...

[Michael Meeks <michael ximian com>]

Hi Bill,

	Ok - so it turns out the oddness on exception is a _really_ silly
at-spi/cspi bug ;-) [ you'll be pleased to know (and the ORB is doing
the right thing ) ], but it throws up a more important / concerning bug
IMHO - I'll see if it's feasible to knock up a patch for that shortly.

	get_child_at_index does:

  retval = cspi_object_add (
    Accessibility_Accessible_getChildAtIndex (CSPI_OBJREF (obj),
					      childIndex, cspi_ev ()));

  cspi_return_val_if_ev ("getChildAtIndex", NULL);
		

	Which is fine: cspi_object_add does:

  else if (!cspi_check_ev ("pre method check: add"))
	retval = NULL;

	which is also correct;

SPIBoolean
cspi_check_ev (const char *error_string)
{
  CORBA_Environment *ev = cspi_ev ();

  if (ev->_major != CORBA_NO_EXCEPTION)
    {

	Which is _fine_ _BUT_ cspi_ev does:

CORBA_Environment *
cspi_ev (void)
{
  CORBA_exception_init (&ev);
  return &ev;
}

	ie. it wipes out the exception as soon as we check it. Thus AFAICS in
cspi we have not been returning defined values from CORBA calls which
fire exceptions for literally ages. [ it looks like this is my fault ].

	So - an immediate fix to get it in is this:

Index: ChangeLog
===================================================================
RCS file: /cvs/gnome/at-spi/ChangeLog,v
retrieving revision 1.317
diff -u -p -u -r1.317 ChangeLog
--- ChangeLog	11 Jun 2003 14:08:11 -0000	1.317
+++ ChangeLog	13 Jun 2003 11:58:26 -0000
@@ -1,3 +1,9 @@
+2003-06-13  Michael Meeks  <michael ximian com>
+
+	* cspi/bonobo/cspi-bonobo.c (cspi_check_ev): use it.
+
+	* cspi/spi_main.c (cspi_peek_ev): impl.
+
 2003-06-11  Padraig O'Briain <padraig obriain sun com>
 
 	* cspi/spi-roletypes.h: Add role SPI_ROLE_AUTOCOMPLETE
Index: cspi/spi-private.h
===================================================================
RCS file: /cvs/gnome/at-spi/cspi/spi-private.h,v
retrieving revision 1.14
diff -u -p -u -r1.14 spi-private.h
--- cspi/spi-private.h	11 Jun 2003 12:07:38 -0000	1.14
+++ cspi/spi-private.h	13 Jun 2003 11:58:26 -0000
@@ -68,6 +68,7 @@ struct _SPIException {
 #define CSPI_OBJREF(a) (((Accessible *)(a))->objref)
 
 CORBA_Environment     *cspi_ev                (void);
+CORBA_Environment     *cspi_peek_ev           (void);
 SPIBoolean             cspi_exception         (void);
 Accessibility_Registry cspi_registry          (void);
 Accessible            *cspi_object_add        (CORBA_Object
corba_object);
Index: cspi/spi_main.c
===================================================================
RCS file: /cvs/gnome/at-spi/cspi/spi_main.c,v
retrieving revision 1.35
diff -u -p -u -r1.35 spi_main.c
--- cspi/spi_main.c	11 Jun 2003 12:07:39 -0000	1.35
+++ cspi/spi_main.c	13 Jun 2003 11:58:27 -0000
@@ -157,6 +157,12 @@ cspi_get_live_refs (void)
 }
 
 CORBA_Environment *
+cspi_peek_ev (void)
+{
+  return &ev;
+}
+
+CORBA_Environment *
 cspi_ev (void)
 {
   CORBA_exception_init (&ev);
Index: cspi/bonobo/cspi-bonobo.c
===================================================================
RCS file: /cvs/gnome/at-spi/cspi/bonobo/cspi-bonobo.c,v
retrieving revision 1.9
diff -u -p -u -r1.9 cspi-bonobo.c
--- cspi/bonobo/cspi-bonobo.c	9 Jun 2003 22:32:50 -0000	1.9
+++ cspi/bonobo/cspi-bonobo.c	13 Jun 2003 11:58:27 -0000
@@ -41,7 +41,7 @@ cspi_release_unref (CORBA_Object object)
 SPIBoolean
 cspi_check_ev (const char *error_string)
 {
-  CORBA_Environment *ev = cspi_ev ();
+  CORBA_Environment *ev = cspi_peek_ev ();
 
   if (ev->_major != CORBA_NO_EXCEPTION)
     {


	This exposes a rather broader issue, which is that the continual re-use
of that CORBA_Environment is quite badly broken;

	Consider:

	a) outgoing method with cspi_ev();
		b) incoming event
			c) outgoing method with cspi_ev returns
			   an exception - sets cspi_ev.
		even returns
	method (apparently) fails with exception - although it completed
cleanly - resulting in a leak / bad result.

	So - we really need to retro-fit the code en-masse to instantiate /
initialize CORBA_Environments where we need them, rather than trying to
re-use the system one.

	Anyhow - I hope the fix will solve the immediate problem.

	Regards,

		Michael.

-- 
 michael ximian com  <><, Pseudo Engineer, itinerant idiot