[Bf-blender-cvs] [cc31722] master: Fix T38000: vertex parent crash due to threading issues

[Sergey Sharybin]

Commit: cc31722a412b19645aebedb8192502d3a3e73cdb
Author: Sergey Sharybin
Date:   Thu Jan 9 16:19:51 2014 +0600
https://developer.blender.org/rBcc31722a412b19645aebedb8192502d3a3e73cdb

Fix T38000: vertex parent crash due to threading issues

Issue is causes by vertex parent modifies original BMesh from
a multiple threads. Ideally this is to be done as a separate
update callback for mesh datablock, but it's not so much simple
now (would need to do some re-arranges to DAG which might conflict
with the work from Ali or will double amount of work we did).

So for now use simple solution with mutex lock.

Based on the patch from Campbell Barton with some fixes to make
changes really thread-safe.

Differential Revision: https://developer.blender.org/D168

===================================================================

M	source/blender/blenkernel/intern/object.c

===================================================================

diff --git a/source/blender/blenkernel/intern/object.c b/source/blender/blenkernel/intern/object.c
index 4dccffe..d0d4c6f 100644
--- a/source/blender/blenkernel/intern/object.c
+++ b/source/blender/blenkernel/intern/object.c
@@ -62,6 +62,7 @@
 
 #include "BLI_blenlib.h"
 #include "BLI_math.h"
+#include "BLI_threads.h"
 #include "BLI_utildefines.h"
 #include "BLI_linklist.h"
 #include "BLI_kdtree.h"
@@ -122,6 +123,18 @@
 
 #include "GPU_material.h"
 
+/* Vertex parent modifies original BMesh which is not safe for threading.
+ * Ideally such a modification should be handled as a separate DAG update
+ * callback for mesh datablock, but for until it is actually supported use
+ * simpler solution with a mutex lock.
+ *                                               - sergey -
+ */
+#define VPARENT_THREADING_HACK
+
+#ifdef VPARENT_THREADING_HACK
+static ThreadMutex vparent_lock = BLI_MUTEX_INITIALIZER;
+#endif
+
 void BKE_object_workob_clear(Object *workob)
 {
 	memset(workob, 0, sizeof(Object));
@@ -2081,32 +2094,13 @@ static void ob_parbone(Object *ob, Object *par, float mat[4][4])
 
 static void give_parvert(Object *par, int nr, float vec[3])
 {
-	BMEditMesh *em;
-
 	zero_v3(vec);
 	
 	if (par->type == OB_MESH) {
 		Mesh *me = par->data;
+		BMEditMesh *em = me->edit_btmesh;
 		DerivedMesh *dm;
 
-		em = me->edit_btmesh;
-
-#if 0   /* this was bmesh only, better, evaluate why this was needed - campbell*/
-		if (em) {
-			BMVert *eve;
-			BMIter iter;
-
-			BM_ITER_MESH (eve, &iter, em->bm, BM_VERTS_OF_MESH) {
-				int *keyindex = CustomData_bmesh_get(&em->bm->vdata, eve->head.data, CD_SHAPE_KEYINDEX);
-				
-				if (keyindex && *keyindex == nr) {
-					copy_v3_v3(vec, eve->co);
-					break;
-				}
-			}
-		}
-#endif
-
 		dm = (em) ? em->derivedFinal : par->derivedFinal;
 			
 		if (dm) {
@@ -2114,22 +2108,41 @@ static void give_parvert(Object *par, int nr, float vec[3])
 			int numVerts = dm->getNumVerts(dm);
 
 			if (nr < numVerts) {
-				MVert *mvert = dm->getVertArray(dm);
-				int   *index = (int *)dm->getVertDataArray(dm, CD_ORIGINDEX);
+				/* avoid dm->getVertDataArray() since it allocates arrays in the dm (not thread safe) */
 				int i;
 
+				if (em && dm->type == DM_TYPE_EDITBMESH) {
+					if (em->bm->elem_table_dirty & BM_VERT) {
+#ifdef VPARENT_THREADING_HACK
+						BLI_mutex_lock(&vparent_lock);
+						if (em->bm->elem_table_dirty & BM_VERT) {
+							BM_mesh_elem_table_ensure(em->bm, BM_VERT);
+						}
+						BLI_mutex_unlock(&vparent_lock);
+#else
+						BLI_assert(!"Not safe for threading");
+						BM_mesh_elem_table_ensure(em->bm, BM_VERT);
+#endif
+					}
+				}
+
 				/* get the average of all verts with (original index == nr) */
-				if (index) {
+				if (CustomData_has_layer(&dm->vertData, CD_ORIGINDEX)) {
 					for (i = 0; i < numVerts; i++) {
-						if (index[i] == nr) {
-							add_v3_v3(vec, mvert[i].co);
+						const int *index = dm->getVertData(dm, i, CD_ORIGINDEX);
+						if (*index == nr) {
+							float co[3];
+							dm->getVertCo(dm, i, co);
+							add_v3_v3(vec, co);
 							count++;
 						}
 					}
 				}
 				else {
 					if (nr < numVerts) {
-						add_v3_v3(vec, mvert[nr].co);
+						float co[3];
+						dm->getVertCo(dm, nr, co);
+						add_v3_v3(vec, co);
 						count++;
 					}
 				}