Bullet Integration

This adds rigid body simulation to blender, developed during Aligorith's and my GSoC projects.
Should apply and compile with current trunk.
Only limitation is that WITH_BULLET needs to be true.
More info here:
http://wiki.blender.org/index.php/User:Sergof/GSoC2012/Design

[ideasman42, 2013-01-15 15:47:09]

LGTM to go into trunk, however a few issues I noticed.

Summery

- If 'Physics World' id disabled, Id like this to work exactly the same as if
physics wasn't ever added. Currently there are a few places where disabling
physics wont stop pointcache, syncing functions and depsgraph from using changed
behavior. I've noted the ones that I found in the review.

- Mesh conversion could be sped up by not adding triangles one at a time with
double-vertex checks, but this can be done after merge.

- I noticed scale is always removed from objects involved with physics, IIRC the
game engine manages to have scaled meshes with physics so it could be supported.

- Id like to allow building with WITH_BULLET disabled, any hints as to which
branch I should look to getting the this change?

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
File source/blender/blenkernel/intern/depsgraph.c (right):

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
source/blender/blenkernel/intern/depsgraph.c:2353: if (ob->rigidbody_object)
ob->recalc |= OB_RECALC_OB;
Firstly, shouldn't this check if physics world is enabled or muted?

When the physics world is enabled:
Should this check if the physics object is sleeping before assuming it needs to
be recalculated.

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
File source/blender/blenkernel/intern/rigidbody.c (right):

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
source/blender/blenkernel/intern/rigidbody.c:910: MEM_freeN(rbw->objects);
suggest to re-alloc here and only when needed. likely in most cases the array
wont resize and existing array can be used.

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
source/blender/blenkernel/intern/rigidbody.c:1099: void
BKE_rigidbody_sync_transforms(Scene *scene, Object *ob, float ctime)
I've noticed this function runs even when (rbw->flag & RBW_FLAG_MUTED), Id
prefer if un-checking the physics option panel disables physics functions like
this.

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
source/blender/blenkernel/intern/rigidbody.c:1147: if (rbw->physics_world ==
NULL || rbw->numbodies != BLI_countlist(&rbw->group->gobject)) {
counting the number of objects in the group might not be totally reliable
(objects could be swapped out).

Perhaps existing methods such as depsgraph flags or notifiers can be used to set
the cache as invalid.

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
File source/blender/blenkernel/intern/scene.c (right):

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
source/blender/blenkernel/intern/scene.c:1151: /* quick point cache updates */
This runs even when physics is muted.

also - would be nice to add note that this is for rigidbody sim.

https://codereview.appspot.com/7092051/diff/1/source/blender/editors/object/o...
File source/blender/editors/object/object_add.c (right):

https://codereview.appspot.com/7092051/diff/1/source/blender/editors/object/o...
source/blender/editors/object/object_add.c:947: /* remove rigid body constraint
from world before removing object */
There are other places objects are removed from a scene: make_proxy_exec().
rna_Scene_object_unlink(), I think 'BLI_remlink(&scene->base, base)' should be
replaced with a BKE_scene_base_unlink() function that handles low level stuff
like freeing rigid body data.

Then all can call this.

https://codereview.appspot.com/7092051/diff/1/source/blender/rigidbody/rb_bul...
File source/blender/rigidbody/rb_bullet_api.cpp (right):

https://codereview.appspot.com/7092051/diff/1/source/blender/rigidbody/rb_bul...
source/blender/rigidbody/rb_bullet_api.cpp:419: v_out[0] = (float)vec[0];
*picky* suggest to add some local static functions to handle conversions.

https://codereview.appspot.com/7092051/diff/1/source/blender/rigidbody/rb_bul...
source/blender/rigidbody/rb_bullet_api.cpp:657: void
RB_trimesh_add_triangle(rbMeshData *mesh, const float v1[3], const float v2[3],
const float v3[3])
calling add triangle on each mesh face isnt that optimal, this isnt a big deal
since its not so hard to change this after the merge.

[ideasman42, 2013-01-15 16:08:58]

Some things I should have mentioned last review...

- The way groups is used I find a bit strange, instead it could flag objects and
store in an array? - it works as is and may be annoying to change right now but
noting here.
Maybe other devs can explain why doing it this way is good.

- Think it would be good to have a check on the scene if physics should run
something like;
'bool BKE_scene_rigidbody_check(Scene *scene)',
The check only needs to see if the scene has a physics pointer and its un-muted.
Then any place physics is done through the code this check can be called.
Currently each function has to check in its body which I find less easy to
follow.

[sergof, 2013-01-15 20:05:02]

Thanks for the review, a few quick comments:

> - Mesh conversion could be sped up by not adding triangles one at a time with
> double-vertex checks, but this can be done after merge.

Right, I haven't touched that actually. Ideally we could avoid allocating the
extra mesh by using blender's mesh data directly in bullet, would need to test
if that could cause problems in cases where the mesh changes while the
simulation is running.

> - I noticed scale is always removed from objects involved with physics, IIRC
the
> game engine manages to have scaled meshes with physics so it could be
supported.

Scaling is supported, we create unscaled collision shapes because we scale them
afterwards. This also allows for dynamic scaling while the simulation is
running.
> 
> - Id like to allow building with WITH_BULLET disabled, any hints as to which
> branch I should look to getting the this change?

Sure, I'd do it myself but if you want, you can look here:
https://gitorious.org/~sergof/blenderprojects/sergofs-blender

> - The way groups is used I find a bit strange, instead it could flag objects
and
> store in an array? - it works as is and may be annoying to change right now
but
> noting here.
> Maybe other devs can explain why doing it this way is good.

Personally I'd prefer to get rid of the groups and just use arrays instead,
since we need them for the cache anyway. It would also make the code much more
clear.
However, Ton insists on using groups. I guess if we'd add multiple rigid body
worlds someday it would be easier to keep track of objects.

Will fix the other issues.

[sergof, 2013-01-15 20:15:27]

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
File source/blender/blenkernel/intern/scene.c (right):

https://codereview.appspot.com/7092051/diff/1/source/blender/blenkernel/inter...
source/blender/blenkernel/intern/scene.c:1151: /* quick point cache updates */
On 2013/01/15 15:47:09, ideasman42 wrote:
> This runs even when physics is muted.
> 
> also - would be nice to add note that this is for rigidbody sim.
Oh, this is a merge mistake (ridibdoy sim doesn't even use quick cache).

[ideasman42, 2013-01-17 11:16:03]

On 2013/01/15 20:05:02, sergof wrote:
> Thanks for the review, a few quick comments:
> 
> > - Mesh conversion could be sped up by not adding triangles one at a time
with
> > double-vertex checks, but this can be done after merge.
> 
> Right, I haven't touched that actually. Ideally we could avoid allocating the
> extra mesh by using blender's mesh data directly in bullet, would need to test
> if that could cause problems in cases where the mesh changes while the
> simulation is running.
> 
> > - I noticed scale is always removed from objects involved with physics, IIRC
> the
> > game engine manages to have scaled meshes with physics so it could be
> supported.
> 
> Scaling is supported, we create unscaled collision shapes because we scale
them
> afterwards. This also allows for dynamic scaling while the simulation is
> running.
> > 
> > - Id like to allow building with WITH_BULLET disabled, any hints as to which
> > branch I should look to getting the this change?
> 
> Sure, I'd do it myself but if you want, you can look here:
> https://gitorious.org/%7Esergof/blenderprojects/sergofs-blender
> 
> > - The way groups is used I find a bit strange, instead it could flag objects
> and
> > store in an array? - it works as is and may be annoying to change right now
> but
> > noting here.
> > Maybe other devs can explain why doing it this way is good.
> 
> Personally I'd prefer to get rid of the groups and just use arrays instead,
> since we need them for the cache anyway. It would also make the code much more
> clear.
> However, Ton insists on using groups. I guess if we'd add multiple rigid body
> worlds someday it would be easier to keep track of objects.

Using groups can be fine, its the way they are integrated, I dont make any
groups - the objects are not green - yet there are 2 groups in the dropdowns.

Since I've used groups a fair bit in blender, I found the automatic creation of
groups odd, also how the objects are in groups but not colored green?

To me its not really clear whats happening and blender is using groups in the
code but hiding it a bit from the user.

Personally - Id prefer this either uses groups like the rest of blender
(dupli-groups, group particles etc), or have some internal datastructure the
user doesnt have to know about.

If we need we could have some utility operator 'Add to physics group' if that is
a hassle to do with existing operators.


> Will fix the other issues.