Make sure the self tasks get a drift even when we don't need to split them at the top level.

src/scheduler.c

@@ -173,6 +173,13 @@ static void scheduler_splittask(struct task *t, struct scheduler *s) {
           /* convert to a self-subtask. */
           t->type = task_type_sub_self;
 
+          /* Make sure we have a drift task */
+          lock_lock(&ci->lock);
+          if (ci->drift == NULL)
+            ci->drift = scheduler_addtask(s, task_type_drift, task_subtype_none,
+                                          0, 0, ci, NULL, 0);
+          lock_unlock_blind(&ci->lock);
+
           /* Depend on local sorts on this cell. */
           if (ci->sorts != NULL) scheduler_addunlock(s, ci->sorts, t);
 

[Pedro Gonnet]

Why do the sub-cell self tasks need the drift to be there? They don't depend on it (at least they shouldn't), as they drift the particles on the fly.

[Matthieu Schaller]

At the moment my gravity tasks don't trigger a drift of the particles. They need a drift task to exist.

[Matthieu Schaller]

I will need to look back at splittask and very likely make an independent version for the gravity tasks. I did not want to break functionalities for now. Let me make that drift task creation only for gravity sub-types.

[Pedro Gonnet]

OK, good for me then!