SVN Revision: 33356 (compiled with scons)
Ubuntu 10.04
nVidia GeForce 8400 GS

I've noticed a lot of inconsistent and unusual behavior when trying to bake smoke simulations. See the attached .blend file. Here is a demonstration of "Free All Bakes" not doing what I would expect.

- Starting on Domain1, Smoke Cache -> Bake All Dynamics
- After baking, Alt-A and play one cycle, then stop and return to frame 1
- Under Smoke, enable Dissolve and set the Time to 2.
- Smoke Cache -> Free All Bakes
- Smoke Cache -> Bake All Dynamics (finishes quick)
- Using Alt-A to animate shows no change
- Delete the ~/blendcache_smokelab1-4 directory
- Smoke Cache -> Free All Bakes
- Smoke Cache -> Bake All Dynamics (actually bakes this time)
- Using Alt-A to animate shows the dissolve working as expected


Also of note is that the Domain2 smoke cache (type 3, flags 4) isn't a disk cache, unlike the Domain1 smoke cache (type 3, flags 68). In the blendcache_smokelab1-4 directory, I see cache1_s files, but not cache2_s files, although I don't see any obvious differences between the two.
#define PTCACHE_DISK_CACHE 64


Index: source/blender/editors/physics/physics_pointcache.c
===================================================================
--- source/blender/editors/physics/physics_pointcache.c (revision 33356)
+++ source/blender/editors/physics/physics_pointcache.c (working copy)
@@ -135,6 +135,7 @@

for(pid=pidlist.first; pid; pid=pid->next) {
pid->cache->flag &= ~PTCACHE_BAKED;
+ printf("here at ptcache_free_bake_all, freeing '%s' type %d, flags %d\n", pid->ob->id.name, pid->type, pid->cache->flag);
}

here at ptcache_free_bake_all, freeing 'OBEmitter2' type 1, flags 4
here at ptcache_free_bake_all, freeing 'OBEmitter1' type 1, flags 4
here at ptcache_free_bake_all, freeing 'OBDomain2' type 3, flags 4
here at ptcache_free_bake_all, freeing 'OBDomain2' type 4, flags 68
here at ptcache_free_bake_all, freeing 'OBDomain1' type 3, flags 68
here at ptcache_free_bake_all, freeing 'OBDomain1' type 4, flags 68