[svn:parrot] r47560 - in trunk/src: . pmc

[whiteknight at svn.parrot.org]

Author: whiteknight
Date: Sat Jun 12 01:14:00 2010
New Revision: 47560
URL: https://trac.parrot.org/parrot/changeset/47560

Log:
[gc] Gix order-of-destruction bug discovered by Chandon++. The check for aliveness of the scheduler in Parrot_cx_delete_taskwas failing, so in Scheduler.destroy I null out the ->scheduler field of the interpreter to ensure it gets around the segfault here. I don't think gc finalization adds PMCs back onto the free list, so the test there was failing

Modified:
   trunk/src/pmc/scheduler.pmc
   trunk/src/scheduler.c

Modified: trunk/src/pmc/scheduler.pmc
==============================================================================
--- trunk/src/pmc/scheduler.pmc	Fri Jun 11 20:54:21 2010	(r47559)
+++ trunk/src/pmc/scheduler.pmc	Sat Jun 12 01:14:00 2010	(r47560)
@@ -251,6 +251,7 @@
 */
     VTABLE void destroy() {
         Parrot_Scheduler_attributes * const core_struct = PARROT_SCHEDULER(SELF);
+        core_struct->interp->scheduler = NULL;
         /* TT #946: this line is causing an order-of-destruction error
            because the scheduler is being freed before its tasks.
            Commenting this out till we get a real fix (although it's a hack) */

Modified: trunk/src/scheduler.c
==============================================================================
--- trunk/src/scheduler.c	Fri Jun 11 20:54:21 2010	(r47559)
+++ trunk/src/scheduler.c	Sat Jun 12 01:14:00 2010	(r47560)
@@ -403,7 +403,7 @@
 Parrot_cx_delete_task(PARROT_INTERP, ARGIN(PMC *task))
 {
     ASSERT_ARGS(Parrot_cx_delete_task)
-    if (interp->scheduler && !PObj_on_free_list_TEST(interp->scheduler)) {
+    if (interp->scheduler) {
         const INTVAL tid = VTABLE_get_integer(interp, task);
         VTABLE_delete_keyed_int(interp, interp->scheduler, tid);
     }