blender/source/blender/collada/DocumentImporter.cpp

3972 lines
113 KiB
C++

/**
* $Id$
*
* ***** BEGIN GPL LICENSE BLOCK *****
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software Foundation,
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*
* Contributor(s): Chingiz Dyussenov, Arystanbek Dyussenov, Nathan Letwory.
*
* ***** END GPL LICENSE BLOCK *****
*/
// TODO:
// * name imported objects
// * import object rotation as euler
#include "COLLADAFWRoot.h"
#include "COLLADAFWIWriter.h"
#include "COLLADAFWStableHeaders.h"
#include "COLLADAFWAnimationCurve.h"
#include "COLLADAFWAnimationList.h"
#include "COLLADAFWCamera.h"
#include "COLLADAFWColorOrTexture.h"
#include "COLLADAFWEffect.h"
#include "COLLADAFWFloatOrDoubleArray.h"
#include "COLLADAFWGeometry.h"
#include "COLLADAFWImage.h"
#include "COLLADAFWIndexList.h"
#include "COLLADAFWInstanceGeometry.h"
#include "COLLADAFWLight.h"
#include "COLLADAFWMaterial.h"
#include "COLLADAFWMesh.h"
#include "COLLADAFWMeshPrimitiveWithFaceVertexCount.h"
#include "COLLADAFWNode.h"
#include "COLLADAFWPolygons.h"
#include "COLLADAFWSampler.h"
#include "COLLADAFWSkinController.h"
#include "COLLADAFWSkinControllerData.h"
#include "COLLADAFWTransformation.h"
#include "COLLADAFWTranslate.h"
#include "COLLADAFWRotate.h"
#include "COLLADAFWScale.h"
#include "COLLADAFWMatrix.h"
#include "COLLADAFWTypes.h"
#include "COLLADAFWVisualScene.h"
#include "COLLADAFWFileInfo.h"
#include "COLLADAFWArrayPrimitiveType.h"
#include "COLLADAFWLibraryNodes.h"
#include "COLLADASaxFWLLoader.h"
// TODO move "extern C" into header files
extern "C"
{
#include "ED_keyframing.h"
#include "ED_armature.h"
#include "ED_mesh.h" // ED_vgroup_vert_add, ...
#include "ED_anim_api.h"
#include "ED_object.h"
#include "WM_types.h"
#include "WM_api.h"
#include "BKE_main.h"
#include "BKE_customdata.h"
#include "BKE_library.h"
#include "BKE_texture.h"
#include "BKE_fcurve.h"
#include "BKE_depsgraph.h"
#include "BLI_path_util.h"
#include "BKE_displist.h"
#include "BLI_math.h"
#include "BKE_scene.h"
}
#include "BKE_armature.h"
#include "BKE_mesh.h"
#include "BKE_global.h"
#include "BKE_context.h"
#include "BKE_object.h"
#include "BKE_image.h"
#include "BKE_material.h"
#include "BKE_utildefines.h"
#include "BKE_action.h"
#include "BLI_math.h"
#include "BLI_listbase.h"
#include "BLI_string.h"
#include "DNA_lamp_types.h"
#include "DNA_armature_types.h"
#include "DNA_anim_types.h"
#include "DNA_curve_types.h"
#include "DNA_texture_types.h"
#include "DNA_camera_types.h"
#include "DNA_object_types.h"
#include "DNA_meshdata_types.h"
#include "DNA_mesh_types.h"
#include "DNA_material_types.h"
#include "DNA_scene_types.h"
#include "DNA_modifier_types.h"
#include "MEM_guardedalloc.h"
#include "DocumentImporter.h"
#include "collada_internal.h"
#include <string>
#include <map>
#include <algorithm> // sort()
#include <math.h>
#include <float.h>
// #define COLLADA_DEBUG
// creates empties for each imported bone on layer 2, for debugging
// #define ARMATURE_TEST
char *CustomData_get_layer_name(const struct CustomData *data, int type, int n);
static const char *primTypeToStr(COLLADAFW::MeshPrimitive::PrimitiveType type)
{
using namespace COLLADAFW;
switch (type) {
case MeshPrimitive::LINES:
return "LINES";
case MeshPrimitive::LINE_STRIPS:
return "LINESTRIPS";
case MeshPrimitive::POLYGONS:
return "POLYGONS";
case MeshPrimitive::POLYLIST:
return "POLYLIST";
case MeshPrimitive::TRIANGLES:
return "TRIANGLES";
case MeshPrimitive::TRIANGLE_FANS:
return "TRIANGLE_FANS";
case MeshPrimitive::TRIANGLE_STRIPS:
return "TRIANGLE_FANS";
case MeshPrimitive::POINTS:
return "POINTS";
case MeshPrimitive::UNDEFINED_PRIMITIVE_TYPE:
return "UNDEFINED_PRIMITIVE_TYPE";
}
return "UNKNOWN";
}
static const char *geomTypeToStr(COLLADAFW::Geometry::GeometryType type)
{
switch (type) {
case COLLADAFW::Geometry::GEO_TYPE_MESH:
return "MESH";
case COLLADAFW::Geometry::GEO_TYPE_SPLINE:
return "SPLINE";
case COLLADAFW::Geometry::GEO_TYPE_CONVEX_MESH:
return "CONVEX_MESH";
case COLLADAFW::Geometry::GEO_TYPE_UNKNOWN:
default:
return "UNKNOWN";
}
}
// works for COLLADAFW::Node, COLLADAFW::Geometry
template<class T>
static const char *get_dae_name(T *node)
{
const std::string& name = node->getName();
return name.size() ? name.c_str() : node->getOriginalId().c_str();
}
// use this for retrieving bone names, since these must be unique
template<class T>
static const char *get_joint_name(T *node)
{
const std::string& id = node->getOriginalId();
return id.size() ? id.c_str() : node->getName().c_str();
}
static float get_float_value(const COLLADAFW::FloatOrDoubleArray& array, unsigned int index)
{
if (index >= array.getValuesCount())
return 0.0f;
if (array.getType() == COLLADAFW::MeshVertexData::DATA_TYPE_FLOAT)
return array.getFloatValues()->getData()[index];
else
return array.getDoubleValues()->getData()[index];
}
// copied from /editors/object/object_relations.c
static int test_parent_loop(Object *par, Object *ob)
{
/* test if 'ob' is a parent somewhere in par's parents */
if(par == NULL) return 0;
if(ob == par) return 1;
return test_parent_loop(par->parent, ob);
}
// a shortened version of parent_set_exec()
// if is_parent_space is true then ob->obmat will be multiplied by par->obmat before parenting
static int set_parent(Object *ob, Object *par, bContext *C, bool is_parent_space=true)
{
if (!par || test_parent_loop(par, ob))
return false;
Object workob;
Main *bmain = CTX_data_main(C);
Scene *sce = CTX_data_scene(C);
ob->parent = par;
ob->partype = PAROBJECT;
ob->parsubstr[0] = 0;
if (is_parent_space) {
// calc par->obmat
where_is_object(sce, par);
// move child obmat into world space
float mat[4][4];
mul_m4_m4m4(mat, ob->obmat, par->obmat);
copy_m4_m4(ob->obmat, mat);
}
// apply child obmat (i.e. decompose it into rot/loc/size)
object_apply_mat4(ob, ob->obmat);
// compute parentinv
what_does_parent(sce, ob, &workob);
invert_m4_m4(ob->parentinv, workob.obmat);
ob->recalc |= OB_RECALC_OB | OB_RECALC_DATA;
par->recalc |= OB_RECALC_OB;
DAG_scene_sort(bmain, sce);
DAG_ids_flush_update(bmain, 0);
WM_event_add_notifier(C, NC_OBJECT|ND_TRANSFORM, NULL);
return true;
}
typedef std::map<COLLADAFW::TextureMapId, std::vector<MTex*> > TexIndexTextureArrayMap;
class TransformReader : public TransformBase
{
protected:
UnitConverter *unit_converter;
struct Animation {
Object *ob;
COLLADAFW::Node *node;
COLLADAFW::Transformation *tm; // which transform is animated by an AnimationList->id
};
public:
TransformReader(UnitConverter* conv) : unit_converter(conv) {}
void get_node_mat(float mat[][4], COLLADAFW::Node *node, std::map<COLLADAFW::UniqueId, Animation> *animation_map,
Object *ob)
{
float cur[4][4];
float copy[4][4];
unit_m4(mat);
for (unsigned int i = 0; i < node->getTransformations().getCount(); i++) {
COLLADAFW::Transformation *tm = node->getTransformations()[i];
COLLADAFW::Transformation::TransformationType type = tm->getTransformationType();
switch(type) {
case COLLADAFW::Transformation::TRANSLATE:
dae_translate_to_mat4(tm, cur);
break;
case COLLADAFW::Transformation::ROTATE:
dae_rotate_to_mat4(tm, cur);
break;
case COLLADAFW::Transformation::SCALE:
dae_scale_to_mat4(tm, cur);
break;
case COLLADAFW::Transformation::MATRIX:
dae_matrix_to_mat4(tm, cur);
break;
case COLLADAFW::Transformation::LOOKAT:
case COLLADAFW::Transformation::SKEW:
fprintf(stderr, "LOOKAT and SKEW transformations are not supported yet.\n");
break;
}
copy_m4_m4(copy, mat);
mul_m4_m4m4(mat, cur, copy);
if (animation_map) {
// AnimationList that drives this Transformation
const COLLADAFW::UniqueId& anim_list_id = tm->getAnimationList();
// store this so later we can link animation data with ob
Animation anim = {ob, node, tm};
(*animation_map)[anim_list_id] = anim;
}
}
}
void dae_rotate_to_mat4(COLLADAFW::Transformation *tm, float m[][4])
{
COLLADAFW::Rotate *ro = (COLLADAFW::Rotate*)tm;
COLLADABU::Math::Vector3& axis = ro->getRotationAxis();
float angle = (float)(ro->getRotationAngle() * M_PI / 180.0f);
float ax[] = {axis[0], axis[1], axis[2]};
// float quat[4];
// axis_angle_to_quat(quat, axis, angle);
// quat_to_mat4(m, quat);
axis_angle_to_mat4(m, ax, angle);
}
void dae_translate_to_mat4(COLLADAFW::Transformation *tm, float m[][4])
{
COLLADAFW::Translate *tra = (COLLADAFW::Translate*)tm;
COLLADABU::Math::Vector3& t = tra->getTranslation();
unit_m4(m);
m[3][0] = (float)t[0];
m[3][1] = (float)t[1];
m[3][2] = (float)t[2];
}
void dae_scale_to_mat4(COLLADAFW::Transformation *tm, float m[][4])
{
COLLADABU::Math::Vector3& s = ((COLLADAFW::Scale*)tm)->getScale();
float size[3] = {(float)s[0], (float)s[1], (float)s[2]};
size_to_mat4(m, size);
}
void dae_matrix_to_mat4(COLLADAFW::Transformation *tm, float m[][4])
{
unit_converter->dae_matrix_to_mat4(m, ((COLLADAFW::Matrix*)tm)->getMatrix());
}
void dae_translate_to_v3(COLLADAFW::Transformation *tm, float v[3])
{
dae_vector3_to_v3(((COLLADAFW::Translate*)tm)->getTranslation(), v);
}
void dae_scale_to_v3(COLLADAFW::Transformation *tm, float v[3])
{
dae_vector3_to_v3(((COLLADAFW::Scale*)tm)->getScale(), v);
}
void dae_vector3_to_v3(const COLLADABU::Math::Vector3 &v3, float v[3])
{
v[0] = v3.x;
v[1] = v3.y;
v[2] = v3.z;
}
};
// only for ArmatureImporter to "see" MeshImporter::get_object_by_geom_uid
class MeshImporterBase
{
public:
virtual Object *get_object_by_geom_uid(const COLLADAFW::UniqueId& geom_uid) = 0;
};
// ditto as above
class AnimationImporterBase
{
public:
// virtual void change_eul_to_quat(Object *ob, bAction *act) = 0;
};
class ArmatureImporter : private TransformReader
{
private:
Scene *scene;
UnitConverter *unit_converter;
// std::map<int, JointData> joint_index_to_joint_info_map;
// std::map<COLLADAFW::UniqueId, int> joint_id_to_joint_index_map;
struct LeafBone {
// COLLADAFW::Node *node;
EditBone *bone;
char name[32];
float mat[4][4]; // bone matrix, derived from inv_bind_mat
};
std::vector<LeafBone> leaf_bones;
// int bone_direction_row; // XXX not used
float leaf_bone_length;
int totbone;
// XXX not used
// float min_angle; // minimum angle between bone head-tail and a row of bone matrix
#if 0
struct ArmatureJoints {
Object *ob_arm;
std::vector<COLLADAFW::Node*> root_joints;
};
std::vector<ArmatureJoints> armature_joints;
#endif
Object *empty; // empty for leaf bones
std::map<COLLADAFW::UniqueId, COLLADAFW::UniqueId> geom_uid_by_controller_uid;
std::map<COLLADAFW::UniqueId, COLLADAFW::Node*> joint_by_uid; // contains all joints
std::vector<COLLADAFW::Node*> root_joints;
std::map<COLLADAFW::UniqueId, Object*> joint_parent_map;
MeshImporterBase *mesh_importer;
AnimationImporterBase *anim_importer;
// This is used to store data passed in write_controller_data.
// Arrays from COLLADAFW::SkinControllerData lose ownership, so do this class members
// so that arrays don't get freed until we free them explicitly.
class SkinInfo
{
private:
// to build armature bones from inverse bind matrices
struct JointData {
float inv_bind_mat[4][4]; // joint inverse bind matrix
COLLADAFW::UniqueId joint_uid; // joint node UID
// Object *ob_arm; // armature object
};
float bind_shape_matrix[4][4];
// data from COLLADAFW::SkinControllerData, each array should be freed
COLLADAFW::UIntValuesArray joints_per_vertex;
COLLADAFW::UIntValuesArray weight_indices;
COLLADAFW::IntValuesArray joint_indices;
// COLLADAFW::FloatOrDoubleArray weights;
std::vector<float> weights;
std::vector<JointData> joint_data; // index to this vector is joint index
UnitConverter *unit_converter;
Object *ob_arm;
COLLADAFW::UniqueId controller_uid;
Object *parent;
public:
SkinInfo() {}
SkinInfo(const SkinInfo& skin) : weights(skin.weights),
joint_data(skin.joint_data),
unit_converter(skin.unit_converter),
ob_arm(skin.ob_arm),
controller_uid(skin.controller_uid),
parent(skin.parent)
{
copy_m4_m4(bind_shape_matrix, (float (*)[4])skin.bind_shape_matrix);
transfer_uint_array_data_const(skin.joints_per_vertex, joints_per_vertex);
transfer_uint_array_data_const(skin.weight_indices, weight_indices);
transfer_int_array_data_const(skin.joint_indices, joint_indices);
}
SkinInfo(UnitConverter *conv) : unit_converter(conv), ob_arm(NULL), parent(NULL) {}
// nobody owns the data after this, so it should be freed manually with releaseMemory
template <class T>
void transfer_array_data(T& src, T& dest)
{
dest.setData(src.getData(), src.getCount());
src.yieldOwnerShip();
dest.yieldOwnerShip();
}
// when src is const we cannot src.yieldOwnerShip, this is used by copy constructor
void transfer_int_array_data_const(const COLLADAFW::IntValuesArray& src, COLLADAFW::IntValuesArray& dest)
{
dest.setData((int*)src.getData(), src.getCount());
dest.yieldOwnerShip();
}
void transfer_uint_array_data_const(const COLLADAFW::UIntValuesArray& src, COLLADAFW::UIntValuesArray& dest)
{
dest.setData((unsigned int*)src.getData(), src.getCount());
dest.yieldOwnerShip();
}
void borrow_skin_controller_data(const COLLADAFW::SkinControllerData* skin)
{
transfer_array_data((COLLADAFW::UIntValuesArray&)skin->getJointsPerVertex(), joints_per_vertex);
transfer_array_data((COLLADAFW::UIntValuesArray&)skin->getWeightIndices(), weight_indices);
transfer_array_data((COLLADAFW::IntValuesArray&)skin->getJointIndices(), joint_indices);
// transfer_array_data(skin->getWeights(), weights);
// cannot transfer data for FloatOrDoubleArray, copy values manually
const COLLADAFW::FloatOrDoubleArray& weight = skin->getWeights();
for (unsigned int i = 0; i < weight.getValuesCount(); i++)
weights.push_back(get_float_value(weight, i));
unit_converter->dae_matrix_to_mat4(bind_shape_matrix, skin->getBindShapeMatrix());
}
void free()
{
joints_per_vertex.releaseMemory();
weight_indices.releaseMemory();
joint_indices.releaseMemory();
// weights.releaseMemory();
}
// using inverse bind matrices to construct armature
// it is safe to invert them to get the original matrices
// because if they are inverse matrices, they can be inverted
void add_joint(const COLLADABU::Math::Matrix4& matrix)
{
JointData jd;
unit_converter->dae_matrix_to_mat4(jd.inv_bind_mat, matrix);
joint_data.push_back(jd);
}
void set_controller(const COLLADAFW::SkinController* co)
{
controller_uid = co->getUniqueId();
// fill in joint UIDs
const COLLADAFW::UniqueIdArray& joint_uids = co->getJoints();
for (unsigned int i = 0; i < joint_uids.getCount(); i++) {
joint_data[i].joint_uid = joint_uids[i];
// // store armature pointer
// JointData& jd = joint_index_to_joint_info_map[i];
// jd.ob_arm = ob_arm;
// now we'll be able to get inv bind matrix from joint id
// joint_id_to_joint_index_map[joint_ids[i]] = i;
}
}
// called from write_controller
Object *create_armature(Scene *scene)
{
ob_arm = add_object(scene, OB_ARMATURE);
return ob_arm;
}
Object* set_armature(Object *ob_arm)
{
if (this->ob_arm)
return this->ob_arm;
this->ob_arm = ob_arm;
return ob_arm;
}
bool get_joint_inv_bind_matrix(float inv_bind_mat[][4], COLLADAFW::Node *node)
{
const COLLADAFW::UniqueId& uid = node->getUniqueId();
std::vector<JointData>::iterator it;
for (it = joint_data.begin(); it != joint_data.end(); it++) {
if ((*it).joint_uid == uid) {
copy_m4_m4(inv_bind_mat, (*it).inv_bind_mat);
return true;
}
}
return false;
}
Object *get_armature()
{
return ob_arm;
}
const COLLADAFW::UniqueId& get_controller_uid()
{
return controller_uid;
}
// check if this skin controller references a joint or any descendant of it
//
// some nodes may not be referenced by SkinController,
// in this case to determine if the node belongs to this armature,
// we need to search down the tree
bool uses_joint_or_descendant(COLLADAFW::Node *node)
{
const COLLADAFW::UniqueId& uid = node->getUniqueId();
std::vector<JointData>::iterator it;
for (it = joint_data.begin(); it != joint_data.end(); it++) {
if ((*it).joint_uid == uid)
return true;
}
COLLADAFW::NodePointerArray& children = node->getChildNodes();
for (unsigned int i = 0; i < children.getCount(); i++) {
if (uses_joint_or_descendant(children[i]))
return true;
}
return false;
}
void link_armature(bContext *C, Object *ob, std::map<COLLADAFW::UniqueId, COLLADAFW::Node*>& joint_by_uid,
TransformReader *tm)
{
Main *bmain = CTX_data_main(C);
Scene *scene = CTX_data_scene(C);
ModifierData *md = ED_object_modifier_add(NULL, bmain, scene, ob, NULL, eModifierType_Armature);
((ArmatureModifierData *)md)->object = ob_arm;
copy_m4_m4(ob->obmat, bind_shape_matrix);
object_apply_mat4(ob, ob->obmat);
#if 1
::set_parent(ob, ob_arm, C);
#else
Object workob;
ob->parent = ob_arm;
ob->partype = PAROBJECT;
what_does_parent(scene, ob, &workob);
invert_m4_m4(ob->parentinv, workob.obmat);
ob->recalc |= OB_RECALC_OB|OB_RECALC_DATA;
DAG_scene_sort(bmain, scene);
DAG_ids_flush_update(bmain, 0);
WM_event_add_notifier(C, NC_OBJECT|ND_TRANSFORM, NULL);
#endif
((bArmature*)ob_arm->data)->deformflag = ARM_DEF_VGROUP;
// create all vertex groups
std::vector<JointData>::iterator it;
int joint_index;
for (it = joint_data.begin(), joint_index = 0; it != joint_data.end(); it++, joint_index++) {
const char *name = "Group";
// name group by joint node name
if (joint_by_uid.find((*it).joint_uid) != joint_by_uid.end()) {
name = get_joint_name(joint_by_uid[(*it).joint_uid]);
}
ED_vgroup_add_name(ob, (char*)name);
}
// <vcount> - number of joints per vertex - joints_per_vertex
// <v> - [[bone index, weight index] * joints per vertex] * vertices - weight indices
// ^ bone index can be -1 meaning weight toward bind shape, how to express this in Blender?
// for each vertex in weight indices
// for each bone index in vertex
// add vertex to group at group index
// treat group index -1 specially
// get def group by index with BLI_findlink
for (unsigned int vertex = 0, weight = 0; vertex < joints_per_vertex.getCount(); vertex++) {
unsigned int limit = weight + joints_per_vertex[vertex];
for ( ; weight < limit; weight++) {
int joint = joint_indices[weight], joint_weight = weight_indices[weight];
// -1 means "weight towards the bind shape", we just don't assign it to any group
if (joint != -1) {
bDeformGroup *def = (bDeformGroup*)BLI_findlink(&ob->defbase, joint);
ED_vgroup_vert_add(ob, def, vertex, weights[joint_weight], WEIGHT_REPLACE);
}
}
}
}
bPoseChannel *get_pose_channel_from_node(COLLADAFW::Node *node)
{
return get_pose_channel(ob_arm->pose, get_joint_name(node));
}
void set_parent(Object *_parent)
{
parent = _parent;
}
Object* get_parent()
{
return parent;
}
void find_root_joints(const std::vector<COLLADAFW::Node*> &root_joints,
std::map<COLLADAFW::UniqueId, COLLADAFW::Node*>& joint_by_uid,
std::vector<COLLADAFW::Node*>& result)
{
std::vector<COLLADAFW::Node*>::const_iterator it;
for (it = root_joints.begin(); it != root_joints.end(); it++) {
COLLADAFW::Node *root = *it;
std::vector<JointData>::iterator ji;
for (ji = joint_data.begin(); ji != joint_data.end(); ji++) {
COLLADAFW::Node *joint = joint_by_uid[(*ji).joint_uid];
if (find_node_in_tree(joint, root)) {
if (std::find(result.begin(), result.end(), root) == result.end())
result.push_back(root);
}
}
}
}
bool find_node_in_tree(COLLADAFW::Node *node, COLLADAFW::Node *tree_root)
{
if (node == tree_root)
return true;
COLLADAFW::NodePointerArray& children = tree_root->getChildNodes();
for (unsigned int i = 0; i < children.getCount(); i++) {
if (find_node_in_tree(node, children[i]))
return true;
}
return false;
}
};
std::map<COLLADAFW::UniqueId, SkinInfo> skin_by_data_uid; // data UID = skin controller data UID
#if 0
JointData *get_joint_data(COLLADAFW::Node *node)
{
const COLLADAFW::UniqueId& joint_id = node->getUniqueId();
if (joint_id_to_joint_index_map.find(joint_id) == joint_id_to_joint_index_map.end()) {
fprintf(stderr, "Cannot find a joint index by joint id for %s.\n",
node->getOriginalId().c_str());
return NULL;
}
int joint_index = joint_id_to_joint_index_map[joint_id];
return &joint_index_to_joint_info_map[joint_index];
}
#endif
void create_bone(SkinInfo& skin, COLLADAFW::Node *node, EditBone *parent, int totchild,
float parent_mat[][4], bArmature *arm)
{
float joint_inv_bind_mat[4][4];
// JointData* jd = get_joint_data(node);
float mat[4][4];
if (skin.get_joint_inv_bind_matrix(joint_inv_bind_mat, node)) {
// get original world-space matrix
invert_m4_m4(mat, joint_inv_bind_mat);
}
// create a bone even if there's no joint data for it (i.e. it has no influence)
else {
float obmat[4][4];
// object-space
get_node_mat(obmat, node, NULL, NULL);
// get world-space
if (parent)
mul_m4_m4m4(mat, obmat, parent_mat);
else
copy_m4_m4(mat, obmat);
}
// TODO rename from Node "name" attrs later
EditBone *bone = ED_armature_edit_bone_add(arm, (char*)get_joint_name(node));
totbone++;
if (parent) bone->parent = parent;
// set head
copy_v3_v3(bone->head, mat[3]);
// set tail, don't set it to head because 0-length bones are not allowed
float vec[3] = {0.0f, 0.5f, 0.0f};
add_v3_v3v3(bone->tail, bone->head, vec);
// set parent tail
if (parent && totchild == 1) {
copy_v3_v3(parent->tail, bone->head);
// not setting BONE_CONNECTED because this would lock child bone location with respect to parent
// bone->flag |= BONE_CONNECTED;
// XXX increase this to prevent "very" small bones?
const float epsilon = 0.000001f;
// derive leaf bone length
float length = len_v3v3(parent->head, parent->tail);
if ((length < leaf_bone_length || totbone == 0) && length > epsilon) {
leaf_bone_length = length;
}
// treat zero-sized bone like a leaf bone
if (length <= epsilon) {
add_leaf_bone(parent_mat, parent);
}
/*
#if 0
// and which row in mat is bone direction
float vec[3];
sub_v3_v3v3(vec, parent->tail, parent->head);
#ifdef COLLADA_DEBUG
print_v3("tail - head", vec);
print_m4("matrix", parent_mat);
#endif
for (int i = 0; i < 3; i++) {
#ifdef COLLADA_DEBUG
char *axis_names[] = {"X", "Y", "Z"};
printf("%s-axis length is %f\n", axis_names[i], len_v3(parent_mat[i]));
#endif
float angle = angle_v2v2(vec, parent_mat[i]);
if (angle < min_angle) {
#ifdef COLLADA_DEBUG
print_v3("picking", parent_mat[i]);
printf("^ %s axis of %s's matrix\n", axis_names[i], get_dae_name(node));
#endif
bone_direction_row = i;
min_angle = angle;
}
}
#endif
*/
}
COLLADAFW::NodePointerArray& children = node->getChildNodes();
for (unsigned int i = 0; i < children.getCount(); i++) {
create_bone(skin, children[i], bone, children.getCount(), mat, arm);
}
// in second case it's not a leaf bone, but we handle it the same way
if (!children.getCount() || children.getCount() > 1) {
add_leaf_bone(mat, bone);
}
}
void add_leaf_bone(float mat[][4], EditBone *bone)
{
LeafBone leaf;
leaf.bone = bone;
copy_m4_m4(leaf.mat, mat);
BLI_strncpy(leaf.name, bone->name, sizeof(leaf.name));
leaf_bones.push_back(leaf);
}
void fix_leaf_bones()
{
// just setting tail for leaf bones here
std::vector<LeafBone>::iterator it;
for (it = leaf_bones.begin(); it != leaf_bones.end(); it++) {
LeafBone& leaf = *it;
// pointing up
float vec[3] = {0.0f, 0.0f, 1.0f};
mul_v3_fl(vec, leaf_bone_length);
copy_v3_v3(leaf.bone->tail, leaf.bone->head);
add_v3_v3v3(leaf.bone->tail, leaf.bone->head, vec);
}
}
void set_leaf_bone_shapes(Object *ob_arm)
{
bPose *pose = ob_arm->pose;
std::vector<LeafBone>::iterator it;
for (it = leaf_bones.begin(); it != leaf_bones.end(); it++) {
LeafBone& leaf = *it;
bPoseChannel *pchan = get_pose_channel(pose, leaf.name);
if (pchan) {
pchan->custom = get_empty_for_leaves();
}
else {
fprintf(stderr, "Cannot find a pose channel for leaf bone %s\n", leaf.name);
}
}
}
#if 0
void set_euler_rotmode()
{
// just set rotmode = ROT_MODE_EUL on pose channel for each joint
std::map<COLLADAFW::UniqueId, COLLADAFW::Node*>::iterator it;
for (it = joint_by_uid.begin(); it != joint_by_uid.end(); it++) {
COLLADAFW::Node *joint = it->second;
std::map<COLLADAFW::UniqueId, SkinInfo>::iterator sit;
for (sit = skin_by_data_uid.begin(); sit != skin_by_data_uid.end(); sit++) {
SkinInfo& skin = sit->second;
if (skin.uses_joint_or_descendant(joint)) {
bPoseChannel *pchan = skin.get_pose_channel_from_node(joint);
if (pchan) {
pchan->rotmode = ROT_MODE_EUL;
}
else {
fprintf(stderr, "Cannot find pose channel for %s.\n", get_joint_name(joint));
}
break;
}
}
}
}
#endif
Object *get_empty_for_leaves()
{
if (empty) return empty;
empty = add_object(scene, OB_EMPTY);
empty->empty_drawtype = OB_EMPTY_SPHERE;
return empty;
}
#if 0
Object *find_armature(COLLADAFW::Node *node)
{
JointData* jd = get_joint_data(node);
if (jd) return jd->ob_arm;
COLLADAFW::NodePointerArray& children = node->getChildNodes();
for (int i = 0; i < children.getCount(); i++) {
Object *ob_arm = find_armature(children[i]);
if (ob_arm) return ob_arm;
}
return NULL;
}
ArmatureJoints& get_armature_joints(Object *ob_arm)
{
// try finding it
std::vector<ArmatureJoints>::iterator it;
for (it = armature_joints.begin(); it != armature_joints.end(); it++) {
if ((*it).ob_arm == ob_arm) return *it;
}
// not found, create one
ArmatureJoints aj;
aj.ob_arm = ob_arm;
armature_joints.push_back(aj);
return armature_joints.back();
}
#endif
void create_armature_bones(SkinInfo& skin)
{
// just do like so:
// - get armature
// - enter editmode
// - add edit bones and head/tail properties using matrices and parent-child info
// - exit edit mode
// - set a sphere shape to leaf bones
Object *ob_arm = NULL;
/*
* find if there's another skin sharing at least one bone with this skin
* if so, use that skin's armature
*/
/*
Pseudocode:
find_node_in_tree(node, root_joint)
skin::find_root_joints(root_joints):
std::vector root_joints;
for each root in root_joints:
for each joint in joints:
if find_node_in_tree(joint, root):
if (std::find(root_joints.begin(), root_joints.end(), root) == root_joints.end())
root_joints.push_back(root);
for (each skin B with armature) {
find all root joints for skin B
for each joint X in skin A:
for each root joint R in skin B:
if (find_node_in_tree(X, R)) {
shared = 1;
goto endloop;
}
}
endloop:
*/
SkinInfo *a = &skin;
Object *shared = NULL;
std::vector<COLLADAFW::Node*> skin_root_joints;
std::map<COLLADAFW::UniqueId, SkinInfo>::iterator it;
for (it = skin_by_data_uid.begin(); it != skin_by_data_uid.end(); it++) {
SkinInfo *b = &it->second;
if (b == a || b->get_armature() == NULL)
continue;
skin_root_joints.clear();
b->find_root_joints(root_joints, joint_by_uid, skin_root_joints);
std::vector<COLLADAFW::Node*>::iterator ri;
for (ri = skin_root_joints.begin(); ri != skin_root_joints.end(); ri++) {
if (a->uses_joint_or_descendant(*ri)) {
shared = b->get_armature();
break;
}
}
if (shared != NULL)
break;
}
if (shared)
ob_arm = skin.set_armature(shared);
else
ob_arm = skin.create_armature(scene);
// enter armature edit mode
ED_armature_to_edit(ob_arm);
leaf_bones.clear();
totbone = 0;
// bone_direction_row = 1; // TODO: don't default to Y but use asset and based on it decide on default row
leaf_bone_length = 0.1f;
// min_angle = 360.0f; // minimum angle between bone head-tail and a row of bone matrix
// create bones
/*
TODO:
check if bones have already been created for a given joint
*/
std::vector<COLLADAFW::Node*>::iterator ri;
for (ri = root_joints.begin(); ri != root_joints.end(); ri++) {
// for shared armature check if bone tree is already created
if (shared && std::find(skin_root_joints.begin(), skin_root_joints.end(), *ri) != skin_root_joints.end())
continue;
// since root_joints may contain joints for multiple controllers, we need to filter
if (skin.uses_joint_or_descendant(*ri)) {
create_bone(skin, *ri, NULL, (*ri)->getChildNodes().getCount(), NULL, (bArmature*)ob_arm->data);
if (joint_parent_map.find((*ri)->getUniqueId()) != joint_parent_map.end() && !skin.get_parent())
skin.set_parent(joint_parent_map[(*ri)->getUniqueId()]);
}
}
fix_leaf_bones();
// exit armature edit mode
ED_armature_from_edit(ob_arm);
ED_armature_edit_free(ob_arm);
DAG_id_flush_update(&ob_arm->id, OB_RECALC_OB|OB_RECALC_DATA);
set_leaf_bone_shapes(ob_arm);
// set_euler_rotmode();
}
public:
ArmatureImporter(UnitConverter *conv, MeshImporterBase *mesh, AnimationImporterBase *anim, Scene *sce) :
TransformReader(conv), scene(sce), empty(NULL), mesh_importer(mesh), anim_importer(anim) {}
~ArmatureImporter()
{
// free skin controller data if we forget to do this earlier
std::map<COLLADAFW::UniqueId, SkinInfo>::iterator it;
for (it = skin_by_data_uid.begin(); it != skin_by_data_uid.end(); it++) {
it->second.free();
}
}
// root - if this joint is the top joint in hierarchy, if a joint
// is a child of a node (not joint), root should be true since
// this is where we build armature bones from
void add_joint(COLLADAFW::Node *node, bool root, Object *parent)
{
joint_by_uid[node->getUniqueId()] = node;
if (root) {
root_joints.push_back(node);
if (parent)
joint_parent_map[node->getUniqueId()] = parent;
}
}
#if 0
void add_root_joint(COLLADAFW::Node *node)
{
// root_joints.push_back(node);
Object *ob_arm = find_armature(node);
if (ob_arm) {
get_armature_joints(ob_arm).root_joints.push_back(node);
}
#ifdef COLLADA_DEBUG
else {
fprintf(stderr, "%s cannot be added to armature.\n", get_joint_name(node));
}
#endif
}
#endif
// here we add bones to armatures, having armatures previously created in write_controller
void make_armatures(bContext *C)
{
std::map<COLLADAFW::UniqueId, SkinInfo>::iterator it;
for (it = skin_by_data_uid.begin(); it != skin_by_data_uid.end(); it++) {
SkinInfo& skin = it->second;
create_armature_bones(skin);
// link armature with a mesh object
Object *ob = mesh_importer->get_object_by_geom_uid(*get_geometry_uid(skin.get_controller_uid()));
if (ob)
skin.link_armature(C, ob, joint_by_uid, this);
else
fprintf(stderr, "Cannot find object to link armature with.\n");
// set armature parent if any
Object *par = skin.get_parent();
if (par)
set_parent(skin.get_armature(), par, C, false);
// free memory stolen from SkinControllerData
skin.free();
}
}
#if 0
// link with meshes, create vertex groups, assign weights
void link_armature(Object *ob_arm, const COLLADAFW::UniqueId& geom_id, const COLLADAFW::UniqueId& controller_data_id)
{
Object *ob = mesh_importer->get_object_by_geom_uid(geom_id);
if (!ob) {
fprintf(stderr, "Cannot find object by geometry UID.\n");
return;
}
if (skin_by_data_uid.find(controller_data_id) == skin_by_data_uid.end()) {
fprintf(stderr, "Cannot find skin info by controller data UID.\n");
return;
}
SkinInfo& skin = skin_by_data_uid[conroller_data_id];
// create vertex groups
}
#endif
bool write_skin_controller_data(const COLLADAFW::SkinControllerData* data)
{
// at this stage we get vertex influence info that should go into me->verts and ob->defbase
// there's no info to which object this should be long so we associate it with skin controller data UID
// don't forget to call defgroup_unique_name before we copy
// controller data uid -> [armature] -> joint data,
// [mesh object]
//
SkinInfo skin(unit_converter);
skin.borrow_skin_controller_data(data);
// store join inv bind matrix to use it later in armature construction
const COLLADAFW::Matrix4Array& inv_bind_mats = data->getInverseBindMatrices();
for (unsigned int i = 0; i < data->getJointsCount(); i++) {
skin.add_joint(inv_bind_mats[i]);
}
skin_by_data_uid[data->getUniqueId()] = skin;
return true;
}
bool write_controller(const COLLADAFW::Controller* controller)
{
// - create and store armature object
const COLLADAFW::UniqueId& skin_id = controller->getUniqueId();
if (controller->getControllerType() == COLLADAFW::Controller::CONTROLLER_TYPE_SKIN) {
COLLADAFW::SkinController *co = (COLLADAFW::SkinController*)controller;
// to be able to find geom id by controller id
geom_uid_by_controller_uid[skin_id] = co->getSource();
const COLLADAFW::UniqueId& data_uid = co->getSkinControllerData();
if (skin_by_data_uid.find(data_uid) == skin_by_data_uid.end()) {
fprintf(stderr, "Cannot find skin by controller data UID.\n");
return true;
}
skin_by_data_uid[data_uid].set_controller(co);
}
// morph controller
else {
// shape keys? :)
fprintf(stderr, "Morph controller is not supported yet.\n");
}
return true;
}
COLLADAFW::UniqueId *get_geometry_uid(const COLLADAFW::UniqueId& controller_uid)
{
if (geom_uid_by_controller_uid.find(controller_uid) == geom_uid_by_controller_uid.end())
return NULL;
return &geom_uid_by_controller_uid[controller_uid];
}
Object *get_armature_for_joint(COLLADAFW::Node *node)
{
std::map<COLLADAFW::UniqueId, SkinInfo>::iterator it;
for (it = skin_by_data_uid.begin(); it != skin_by_data_uid.end(); it++) {
SkinInfo& skin = it->second;
if (skin.uses_joint_or_descendant(node))
return skin.get_armature();
}
return NULL;
}
void get_rna_path_for_joint(COLLADAFW::Node *node, char *joint_path, size_t count)
{
BLI_snprintf(joint_path, count, "pose.bones[\"%s\"]", get_joint_name(node));
}
// gives a world-space mat
bool get_joint_bind_mat(float m[][4], COLLADAFW::Node *joint)
{
std::map<COLLADAFW::UniqueId, SkinInfo>::iterator it;
bool found = false;
for (it = skin_by_data_uid.begin(); it != skin_by_data_uid.end(); it++) {
SkinInfo& skin = it->second;
if ((found = skin.get_joint_inv_bind_matrix(m, joint))) {
invert_m4(m);
break;
}
}
return found;
}
};
class MeshImporter : public MeshImporterBase
{
private:
Scene *scene;
ArmatureImporter *armature_importer;
std::map<COLLADAFW::UniqueId, Mesh*> uid_mesh_map; // geometry unique id-to-mesh map
std::map<COLLADAFW::UniqueId, Object*> uid_object_map; // geom uid-to-object
// this structure is used to assign material indices to faces
// it holds a portion of Mesh faces and corresponds to a DAE primitive list (<triangles>, <polylist>, etc.)
struct Primitive {
MFace *mface;
unsigned int totface;
};
typedef std::map<COLLADAFW::MaterialId, std::vector<Primitive> > MaterialIdPrimitiveArrayMap;
std::map<COLLADAFW::UniqueId, MaterialIdPrimitiveArrayMap> geom_uid_mat_mapping_map; // crazy name!
class UVDataWrapper
{
COLLADAFW::MeshVertexData *mVData;
public:
UVDataWrapper(COLLADAFW::MeshVertexData& vdata) : mVData(&vdata)
{}
#ifdef COLLADA_DEBUG
void print()
{
fprintf(stderr, "UVs:\n");
switch(mVData->getType()) {
case COLLADAFW::MeshVertexData::DATA_TYPE_FLOAT:
{
COLLADAFW::ArrayPrimitiveType<float>* values = mVData->getFloatValues();
if (values->getCount()) {
for (int i = 0; i < values->getCount(); i += 2) {
fprintf(stderr, "%.1f, %.1f\n", (*values)[i], (*values)[i+1]);
}
}
}
break;
case COLLADAFW::MeshVertexData::DATA_TYPE_DOUBLE:
{
COLLADAFW::ArrayPrimitiveType<double>* values = mVData->getDoubleValues();
if (values->getCount()) {
for (int i = 0; i < values->getCount(); i += 2) {
fprintf(stderr, "%.1f, %.1f\n", (float)(*values)[i], (float)(*values)[i+1]);
}
}
}
break;
}
fprintf(stderr, "\n");
}
#endif
void getUV(int uv_index[2], float *uv)
{
switch(mVData->getType()) {
case COLLADAFW::MeshVertexData::DATA_TYPE_FLOAT:
{
COLLADAFW::ArrayPrimitiveType<float>* values = mVData->getFloatValues();
if (values->empty()) return;
uv[0] = (*values)[uv_index[0]];
uv[1] = (*values)[uv_index[1]];
}
break;
case COLLADAFW::MeshVertexData::DATA_TYPE_DOUBLE:
{
COLLADAFW::ArrayPrimitiveType<double>* values = mVData->getDoubleValues();
if (values->empty()) return;
uv[0] = (float)(*values)[uv_index[0]];
uv[1] = (float)(*values)[uv_index[1]];
}
break;
case COLLADAFW::MeshVertexData::DATA_TYPE_UNKNOWN:
default:
fprintf(stderr, "MeshImporter.getUV(): unknown data type\n");
}
}
};
void set_face_indices(MFace *mface, unsigned int *indices, bool quad)
{
mface->v1 = indices[0];
mface->v2 = indices[1];
mface->v3 = indices[2];
if (quad) mface->v4 = indices[3];
else mface->v4 = 0;
#ifdef COLLADA_DEBUG
// fprintf(stderr, "%u, %u, %u \n", indices[0], indices[1], indices[2]);
#endif
}
// not used anymore, test_index_face from blenkernel is better
#if 0
// change face indices order so that v4 is not 0
void rotate_face_indices(MFace *mface) {
mface->v4 = mface->v1;
mface->v1 = mface->v2;
mface->v2 = mface->v3;
mface->v3 = 0;
}
#endif
void set_face_uv(MTFace *mtface, UVDataWrapper &uvs,
COLLADAFW::IndexList& index_list, unsigned int *tris_indices)
{
int uv_indices[4][2];
// per face vertex indices, this means for quad we have 4 indices, not 8
COLLADAFW::UIntValuesArray& indices = index_list.getIndices();
// make indices into FloatOrDoubleArray
for (int i = 0; i < 3; i++) {
int uv_index = indices[tris_indices[i]];
uv_indices[i][0] = uv_index * 2;
uv_indices[i][1] = uv_index * 2 + 1;
}
uvs.getUV(uv_indices[0], mtface->uv[0]);
uvs.getUV(uv_indices[1], mtface->uv[1]);
uvs.getUV(uv_indices[2], mtface->uv[2]);
}
void set_face_uv(MTFace *mtface, UVDataWrapper &uvs,
COLLADAFW::IndexList& index_list, int index, bool quad)
{
int uv_indices[4][2];
// per face vertex indices, this means for quad we have 4 indices, not 8
COLLADAFW::UIntValuesArray& indices = index_list.getIndices();
// make indices into FloatOrDoubleArray
for (int i = 0; i < (quad ? 4 : 3); i++) {
int uv_index = indices[index + i];
uv_indices[i][0] = uv_index * 2;
uv_indices[i][1] = uv_index * 2 + 1;
}
uvs.getUV(uv_indices[0], mtface->uv[0]);
uvs.getUV(uv_indices[1], mtface->uv[1]);
uvs.getUV(uv_indices[2], mtface->uv[2]);
if (quad) uvs.getUV(uv_indices[3], mtface->uv[3]);
#ifdef COLLADA_DEBUG
/*if (quad) {
fprintf(stderr, "face uv:\n"
"((%d, %d), (%d, %d), (%d, %d), (%d, %d))\n"
"((%.1f, %.1f), (%.1f, %.1f), (%.1f, %.1f), (%.1f, %.1f))\n",
uv_indices[0][0], uv_indices[0][1],
uv_indices[1][0], uv_indices[1][1],
uv_indices[2][0], uv_indices[2][1],
uv_indices[3][0], uv_indices[3][1],
mtface->uv[0][0], mtface->uv[0][1],
mtface->uv[1][0], mtface->uv[1][1],
mtface->uv[2][0], mtface->uv[2][1],
mtface->uv[3][0], mtface->uv[3][1]);
}
else {
fprintf(stderr, "face uv:\n"
"((%d, %d), (%d, %d), (%d, %d))\n"
"((%.1f, %.1f), (%.1f, %.1f), (%.1f, %.1f))\n",
uv_indices[0][0], uv_indices[0][1],
uv_indices[1][0], uv_indices[1][1],
uv_indices[2][0], uv_indices[2][1],
mtface->uv[0][0], mtface->uv[0][1],
mtface->uv[1][0], mtface->uv[1][1],
mtface->uv[2][0], mtface->uv[2][1]);
}*/
#endif
}
#ifdef COLLADA_DEBUG
void print_index_list(COLLADAFW::IndexList& index_list)
{
fprintf(stderr, "Index list for \"%s\":\n", index_list.getName().c_str());
for (int i = 0; i < index_list.getIndicesCount(); i += 2) {
fprintf(stderr, "%u, %u\n", index_list.getIndex(i), index_list.getIndex(i + 1));
}
fprintf(stderr, "\n");
}
#endif
bool is_nice_mesh(COLLADAFW::Mesh *mesh)
{
COLLADAFW::MeshPrimitiveArray& prim_arr = mesh->getMeshPrimitives();
const char *name = get_dae_name(mesh);
for (unsigned i = 0; i < prim_arr.getCount(); i++) {
COLLADAFW::MeshPrimitive *mp = prim_arr[i];
COLLADAFW::MeshPrimitive::PrimitiveType type = mp->getPrimitiveType();
const char *type_str = primTypeToStr(type);
// OpenCollada passes POLYGONS type for <polylist>
if (type == COLLADAFW::MeshPrimitive::POLYLIST || type == COLLADAFW::MeshPrimitive::POLYGONS) {
COLLADAFW::Polygons *mpvc = (COLLADAFW::Polygons*)mp;
COLLADAFW::Polygons::VertexCountArray& vca = mpvc->getGroupedVerticesVertexCountArray();
for(unsigned int j = 0; j < vca.getCount(); j++){
int count = vca[j];
if (count < 3) {
fprintf(stderr, "Primitive %s in %s has at least one face with vertex count < 3\n",
type_str, name);
return false;
}
}
}
else if(type != COLLADAFW::MeshPrimitive::TRIANGLES) {
fprintf(stderr, "Primitive type %s is not supported.\n", type_str);
return false;
}
}
if (mesh->getPositions().empty()) {
fprintf(stderr, "Mesh %s has no vertices.\n", name);
return false;
}
return true;
}
void read_vertices(COLLADAFW::Mesh *mesh, Mesh *me)
{
// vertices
me->totvert = mesh->getPositions().getFloatValues()->getCount() / 3;
me->mvert = (MVert*)CustomData_add_layer(&me->vdata, CD_MVERT, CD_CALLOC, NULL, me->totvert);
COLLADAFW::MeshVertexData& pos = mesh->getPositions();
MVert *mvert;
int i;
for (i = 0, mvert = me->mvert; i < me->totvert; i++, mvert++)
get_vector(mvert->co, pos, i);
}
int triangulate_poly(unsigned int *indices, int totvert, MVert *verts, std::vector<unsigned int>& tri)
{
ListBase dispbase;
DispList *dl;
float *vert;
int i = 0;
dispbase.first = dispbase.last = NULL;
dl = (DispList*)MEM_callocN(sizeof(DispList), "poly disp");
dl->nr = totvert;
dl->type = DL_POLY;
dl->parts = 1;
dl->verts = vert = (float*)MEM_callocN(totvert * 3 * sizeof(float), "poly verts");
dl->index = (int*)MEM_callocN(sizeof(int) * 3 * totvert, "dl index");
BLI_addtail(&dispbase, dl);
for (i = 0; i < totvert; i++) {
copy_v3_v3(vert, verts[indices[i]].co);
vert += 3;
}
filldisplist(&dispbase, &dispbase, 0);
int tottri = 0;
dl= (DispList*)dispbase.first;
if (dl->type == DL_INDEX3) {
tottri = dl->parts;
int *index = dl->index;
for (i= 0; i < tottri; i++) {
int t[3]= {*index, *(index + 1), *(index + 2)};
std::sort(t, t + 3);
tri.push_back(t[0]);
tri.push_back(t[1]);
tri.push_back(t[2]);
index += 3;
}
}
freedisplist(&dispbase);
return tottri;
}
int count_new_tris(COLLADAFW::Mesh *mesh, Mesh *me)
{
COLLADAFW::MeshPrimitiveArray& prim_arr = mesh->getMeshPrimitives();
unsigned int i;
int tottri = 0;
for (i = 0; i < prim_arr.getCount(); i++) {
COLLADAFW::MeshPrimitive *mp = prim_arr[i];
int type = mp->getPrimitiveType();
size_t prim_totface = mp->getFaceCount();
unsigned int *indices = mp->getPositionIndices().getData();
if (type == COLLADAFW::MeshPrimitive::POLYLIST ||
type == COLLADAFW::MeshPrimitive::POLYGONS) {
COLLADAFW::Polygons *mpvc = (COLLADAFW::Polygons*)mp;
COLLADAFW::Polygons::VertexCountArray& vcounta = mpvc->getGroupedVerticesVertexCountArray();
for (unsigned int j = 0; j < prim_totface; j++) {
int vcount = vcounta[j];
if (vcount > 4) {
std::vector<unsigned int> tri;
// tottri += triangulate_poly(indices, vcount, me->mvert, tri) - 1; // XXX why - 1?!
tottri += triangulate_poly(indices, vcount, me->mvert, tri);
}
indices += vcount;
}
}
}
return tottri;
}
// TODO: import uv set names
void read_faces(COLLADAFW::Mesh *mesh, Mesh *me, int new_tris)
{
unsigned int i;
// allocate faces
me->totface = mesh->getFacesCount() + new_tris;
me->mface = (MFace*)CustomData_add_layer(&me->fdata, CD_MFACE, CD_CALLOC, NULL, me->totface);
// allocate UV layers
unsigned int totuvset = mesh->getUVCoords().getInputInfosArray().getCount();
// for (i = 0; i < totuvset; i++) {
// if (mesh->getUVCoords().getLength(i) == 0) {
// totuvset = 0;
// break;
// }
// }
for (i = 0; i < totuvset; i++) {
if (mesh->getUVCoords().getLength(i) == 0) {
totuvset = 0;
break;
}
}
for (i = 0; i < totuvset; i++) {
CustomData_add_layer(&me->fdata, CD_MTFACE, CD_CALLOC, NULL, me->totface);
//this->set_layername_map[i] = CustomData_get_layer_name(&me->fdata, CD_MTFACE, i);
}
// activate the first uv layer
if (totuvset) me->mtface = (MTFace*)CustomData_get_layer_n(&me->fdata, CD_MTFACE, 0);
UVDataWrapper uvs(mesh->getUVCoords());
#ifdef COLLADA_DEBUG
// uvs.print();
#endif
MFace *mface = me->mface;
MaterialIdPrimitiveArrayMap mat_prim_map;
int face_index = 0;
COLLADAFW::MeshPrimitiveArray& prim_arr = mesh->getMeshPrimitives();
bool has_normals = mesh->hasNormals();
COLLADAFW::MeshVertexData& nor = mesh->getNormals();
for (i = 0; i < prim_arr.getCount(); i++) {
COLLADAFW::MeshPrimitive *mp = prim_arr[i];
// faces
size_t prim_totface = mp->getFaceCount();
unsigned int *indices = mp->getPositionIndices().getData();
unsigned int *nind = mp->getNormalIndices().getData();
unsigned int j, k;
int type = mp->getPrimitiveType();
int index = 0;
// since we cannot set mface->mat_nr here, we store a portion of me->mface in Primitive
Primitive prim = {mface, 0};
COLLADAFW::IndexListArray& index_list_array = mp->getUVCoordIndicesArray();
#ifdef COLLADA_DEBUG
/*
fprintf(stderr, "Primitive %d:\n", i);
for (int j = 0; j < totuvset; j++) {
print_index_list(*index_list_array[j]);
}
*/
#endif
if (type == COLLADAFW::MeshPrimitive::TRIANGLES) {
for (j = 0; j < prim_totface; j++){
set_face_indices(mface, indices, false);
indices += 3;
#if 0
for (k = 0; k < totuvset; k++) {
if (!index_list_array.empty() && index_list_array[k]) {
// get mtface by face index and uv set index
MTFace *mtface = (MTFace*)CustomData_get_layer_n(&me->fdata, CD_MTFACE, k);
set_face_uv(&mtface[face_index], uvs, k, *index_list_array[k], index, false);
}
}
#else
for (k = 0; k < index_list_array.getCount(); k++) {
int uvset_index = index_list_array[k]->getSetIndex();
// get mtface by face index and uv set index
MTFace *mtface = (MTFace*)CustomData_get_layer_n(&me->fdata, CD_MTFACE, uvset_index);
set_face_uv(&mtface[face_index], uvs, *index_list_array[k], index, false);
}
#endif
test_index_face(mface, &me->fdata, face_index, 3);
if (has_normals) {
if (!flat_face(nind, nor, 3))
mface->flag |= ME_SMOOTH;
nind += 3;
}
index += 3;
mface++;
face_index++;
prim.totface++;
}
}
else if (type == COLLADAFW::MeshPrimitive::POLYLIST || type == COLLADAFW::MeshPrimitive::POLYGONS) {
COLLADAFW::Polygons *mpvc = (COLLADAFW::Polygons*)mp;
COLLADAFW::Polygons::VertexCountArray& vcounta = mpvc->getGroupedVerticesVertexCountArray();
for (j = 0; j < prim_totface; j++) {
// face
int vcount = vcounta[j];
if (vcount == 3 || vcount == 4) {
set_face_indices(mface, indices, vcount == 4);
// set mtface for each uv set
// it is assumed that all primitives have equal number of UV sets
#if 0
for (k = 0; k < totuvset; k++) {
if (!index_list_array.empty() && index_list_array[k]) {
// get mtface by face index and uv set index
MTFace *mtface = (MTFace*)CustomData_get_layer_n(&me->fdata, CD_MTFACE, k);
set_face_uv(&mtface[face_index], uvs, k, *index_list_array[k], index, mface->v4 != 0);
}
}
#else
for (k = 0; k < index_list_array.getCount(); k++) {
int uvset_index = index_list_array[k]->getSetIndex();
// get mtface by face index and uv set index
MTFace *mtface = (MTFace*)CustomData_get_layer_n(&me->fdata, CD_MTFACE, uvset_index);
set_face_uv(&mtface[face_index], uvs, *index_list_array[k], index, mface->v4 != 0);
}
#endif
test_index_face(mface, &me->fdata, face_index, vcount);
if (has_normals) {
if (!flat_face(nind, nor, vcount))
mface->flag |= ME_SMOOTH;
nind += vcount;
}
mface++;
face_index++;
prim.totface++;
}
else {
std::vector<unsigned int> tri;
triangulate_poly(indices, vcount, me->mvert, tri);
for (k = 0; k < tri.size() / 3; k++) {
int v = k * 3;
unsigned int uv_indices[3] = {
index + tri[v],
index + tri[v + 1],
index + tri[v + 2]
};
unsigned int tri_indices[3] = {
indices[tri[v]],
indices[tri[v + 1]],
indices[tri[v + 2]]
};
set_face_indices(mface, tri_indices, false);
#if 0
for (unsigned int l = 0; l < totuvset; l++) {
if (!index_list_array.empty() && index_list_array[l]) {
// get mtface by face index and uv set index
MTFace *mtface = (MTFace*)CustomData_get_layer_n(&me->fdata, CD_MTFACE, l);
set_face_uv(&mtface[face_index], uvs, l, *index_list_array[l], uv_indices);
}
}
#else
for (unsigned int l = 0; l < index_list_array.getCount(); l++) {
int uvset_index = index_list_array[l]->getSetIndex();
// get mtface by face index and uv set index
MTFace *mtface = (MTFace*)CustomData_get_layer_n(&me->fdata, CD_MTFACE, uvset_index);
set_face_uv(&mtface[face_index], uvs, *index_list_array[l], uv_indices);
}
#endif
test_index_face(mface, &me->fdata, face_index, 3);
if (has_normals) {
unsigned int ntri[3] = {nind[tri[v]], nind[tri[v + 1]], nind[tri[v + 2]]};
if (!flat_face(ntri, nor, 3))
mface->flag |= ME_SMOOTH;
}
mface++;
face_index++;
prim.totface++;
}
if (has_normals)
nind += vcount;
}
index += vcount;
indices += vcount;
}
}
mat_prim_map[mp->getMaterialId()].push_back(prim);
}
geom_uid_mat_mapping_map[mesh->getUniqueId()] = mat_prim_map;
}
void get_vector(float v[3], COLLADAFW::MeshVertexData& arr, int i)
{
i *= 3;
switch(arr.getType()) {
case COLLADAFW::MeshVertexData::DATA_TYPE_FLOAT:
{
COLLADAFW::ArrayPrimitiveType<float>* values = arr.getFloatValues();
if (values->empty()) return;
v[0] = (*values)[i++];
v[1] = (*values)[i++];
v[2] = (*values)[i];
}
break;
case COLLADAFW::MeshVertexData::DATA_TYPE_DOUBLE:
{
COLLADAFW::ArrayPrimitiveType<double>* values = arr.getDoubleValues();
if (values->empty()) return;
v[0] = (float)(*values)[i++];
v[1] = (float)(*values)[i++];
v[2] = (float)(*values)[i];
}
break;
default:
break;
}
}
bool flat_face(unsigned int *nind, COLLADAFW::MeshVertexData& nor, int count)
{
float a[3], b[3];
get_vector(a, nor, *nind);
normalize_v3(a);
nind++;
for (int i = 1; i < count; i++, nind++) {
get_vector(b, nor, *nind);
normalize_v3(b);
float dp = dot_v3v3(a, b);
if (dp < 0.99999f || dp > 1.00001f)
return false;
}
return true;
}
public:
MeshImporter(ArmatureImporter *arm, Scene *sce) : scene(sce), armature_importer(arm) {}
virtual Object *get_object_by_geom_uid(const COLLADAFW::UniqueId& geom_uid)
{
if (uid_object_map.find(geom_uid) != uid_object_map.end())
return uid_object_map[geom_uid];
return NULL;
}
MTex *assign_textures_to_uvlayer(COLLADAFW::TextureCoordinateBinding &ctexture,
Mesh *me, TexIndexTextureArrayMap& texindex_texarray_map,
MTex *color_texture)
{
COLLADAFW::TextureMapId texture_index = ctexture.getTextureMapId();
char *uvname = CustomData_get_layer_name(&me->fdata, CD_MTFACE, ctexture.getSetIndex());
if (texindex_texarray_map.find(texture_index) == texindex_texarray_map.end()) {
fprintf(stderr, "Cannot find texture array by texture index.\n");
return color_texture;
}
std::vector<MTex*> textures = texindex_texarray_map[texture_index];
std::vector<MTex*>::iterator it;
for (it = textures.begin(); it != textures.end(); it++) {
MTex *texture = *it;
if (texture) {
strcpy(texture->uvname, uvname);
if (texture->mapto == MAP_COL) color_texture = texture;
}
}
return color_texture;
}
MTFace *assign_material_to_geom(COLLADAFW::MaterialBinding cmaterial,
std::map<COLLADAFW::UniqueId, Material*>& uid_material_map,
Object *ob, const COLLADAFW::UniqueId *geom_uid,
MTex **color_texture, char *layername, MTFace *texture_face,
std::map<Material*, TexIndexTextureArrayMap>& material_texture_mapping_map, int mat_index)
{
Mesh *me = (Mesh*)ob->data;
const COLLADAFW::UniqueId& ma_uid = cmaterial.getReferencedMaterial();
// do we know this material?
if (uid_material_map.find(ma_uid) == uid_material_map.end()) {
fprintf(stderr, "Cannot find material by UID.\n");
return NULL;
}
Material *ma = uid_material_map[ma_uid];
assign_material(ob, ma, ob->totcol + 1);
COLLADAFW::TextureCoordinateBindingArray& tex_array =
cmaterial.getTextureCoordinateBindingArray();
TexIndexTextureArrayMap texindex_texarray_map = material_texture_mapping_map[ma];
unsigned int i;
// loop through <bind_vertex_inputs>
for (i = 0; i < tex_array.getCount(); i++) {
*color_texture = assign_textures_to_uvlayer(tex_array[i], me, texindex_texarray_map,
*color_texture);
}
// set texture face
if (*color_texture &&
strlen((*color_texture)->uvname) &&
strcmp(layername, (*color_texture)->uvname) != 0) {
texture_face = (MTFace*)CustomData_get_layer_named(&me->fdata, CD_MTFACE,
(*color_texture)->uvname);
strcpy(layername, (*color_texture)->uvname);
}
MaterialIdPrimitiveArrayMap& mat_prim_map = geom_uid_mat_mapping_map[*geom_uid];
COLLADAFW::MaterialId mat_id = cmaterial.getMaterialId();
// assign material indices to mesh faces
if (mat_prim_map.find(mat_id) != mat_prim_map.end()) {
std::vector<Primitive>& prims = mat_prim_map[mat_id];
std::vector<Primitive>::iterator it;
for (it = prims.begin(); it != prims.end(); it++) {
Primitive& prim = *it;
i = 0;
while (i++ < prim.totface) {
prim.mface->mat_nr = mat_index;
prim.mface++;
// bind texture images to faces
if (texture_face && (*color_texture)) {
texture_face->mode = TF_TEX;
texture_face->tpage = (Image*)(*color_texture)->tex->ima;
texture_face++;
}
}
}
}
return texture_face;
}
Object *create_mesh_object(COLLADAFW::Node *node, COLLADAFW::InstanceGeometry *geom,
bool isController,
std::map<COLLADAFW::UniqueId, Material*>& uid_material_map,
std::map<Material*, TexIndexTextureArrayMap>& material_texture_mapping_map)
{
const COLLADAFW::UniqueId *geom_uid = &geom->getInstanciatedObjectId();
// check if node instanciates controller or geometry
if (isController) {
geom_uid = armature_importer->get_geometry_uid(*geom_uid);
if (!geom_uid) {
fprintf(stderr, "Couldn't find a mesh UID by controller's UID.\n");
return NULL;
}
}
else {
if (uid_mesh_map.find(*geom_uid) == uid_mesh_map.end()) {
// this could happen if a mesh was not created
// (e.g. if it contains unsupported geometry)
fprintf(stderr, "Couldn't find a mesh by UID.\n");
return NULL;
}
}
if (!uid_mesh_map[*geom_uid]) return NULL;
Object *ob = add_object(scene, OB_MESH);
// store object pointer for ArmatureImporter
uid_object_map[*geom_uid] = ob;
// name Object
const std::string& id = node->getOriginalId();
if (id.length())
rename_id(&ob->id, (char*)id.c_str());
// replace ob->data freeing the old one
Mesh *old_mesh = (Mesh*)ob->data;
set_mesh(ob, uid_mesh_map[*geom_uid]);
if (old_mesh->id.us == 0) free_libblock(&G.main->mesh, old_mesh);
char layername[100];
MTFace *texture_face = NULL;
MTex *color_texture = NULL;
COLLADAFW::MaterialBindingArray& mat_array =
geom->getMaterialBindings();
// loop through geom's materials
for (unsigned int i = 0; i < mat_array.getCount(); i++) {
texture_face = assign_material_to_geom(mat_array[i], uid_material_map, ob, geom_uid,
&color_texture, layername, texture_face,
material_texture_mapping_map, i);
}
return ob;
}
// create a mesh storing a pointer in a map so it can be retrieved later by geometry UID
bool write_geometry(const COLLADAFW::Geometry* geom)
{
// TODO: import also uvs, normals
// XXX what to do with normal indices?
// XXX num_normals may be != num verts, then what to do?
// check geometry->getType() first
if (geom->getType() != COLLADAFW::Geometry::GEO_TYPE_MESH) {
// TODO: report warning
fprintf(stderr, "Mesh type %s is not supported\n", geomTypeToStr(geom->getType()));
return true;
}
COLLADAFW::Mesh *mesh = (COLLADAFW::Mesh*)geom;
if (!is_nice_mesh(mesh)) {
fprintf(stderr, "Ignoring mesh %s\n", get_dae_name(mesh));
return true;
}
const std::string& str_geom_id = mesh->getOriginalId();
Mesh *me = add_mesh((char*)str_geom_id.c_str());
// store the Mesh pointer to link it later with an Object
this->uid_mesh_map[mesh->getUniqueId()] = me;
int new_tris = 0;
read_vertices(mesh, me);
new_tris = count_new_tris(mesh, me);
read_faces(mesh, me, new_tris);
make_edges(me, 0);
mesh_calc_normals(me->mvert, me->totvert, me->mface, me->totface, NULL);
return true;
}
};
class AnimationImporter : private TransformReader, public AnimationImporterBase
{
private:
ArmatureImporter *armature_importer;
Scene *scene;
std::map<COLLADAFW::UniqueId, std::vector<FCurve*> > curve_map;
std::map<COLLADAFW::UniqueId, TransformReader::Animation> uid_animated_map;
// std::map<bActionGroup*, std::vector<FCurve*> > fcurves_actionGroup_map;
std::map<COLLADAFW::UniqueId, const COLLADAFW::AnimationList*> animlist_map;
std::vector<FCurve*> unused_curves;
std::map<COLLADAFW::UniqueId, Object*> joint_objects;
FCurve *create_fcurve(int array_index, const char *rna_path)
{
FCurve *fcu = (FCurve*)MEM_callocN(sizeof(FCurve), "FCurve");
fcu->flag = (FCURVE_VISIBLE|FCURVE_AUTO_HANDLES|FCURVE_SELECTED);
fcu->rna_path = BLI_strdupn(rna_path, strlen(rna_path));
fcu->array_index = array_index;
return fcu;
}
void create_bezt(FCurve *fcu, float frame, float output)
{
BezTriple bez;
memset(&bez, 0, sizeof(BezTriple));
bez.vec[1][0] = frame;
bez.vec[1][1] = output;
bez.ipo = U.ipo_new; /* use default interpolation mode here... */
bez.f1 = bez.f2 = bez.f3 = SELECT;
bez.h1 = bez.h2 = HD_AUTO;
insert_bezt_fcurve(fcu, &bez, 0);
calchandles_fcurve(fcu);
}
// create one or several fcurves depending on the number of parameters being animated
void animation_to_fcurves(COLLADAFW::AnimationCurve *curve)
{
COLLADAFW::FloatOrDoubleArray& input = curve->getInputValues();
COLLADAFW::FloatOrDoubleArray& output = curve->getOutputValues();
// COLLADAFW::FloatOrDoubleArray& intan = curve->getInTangentValues();
// COLLADAFW::FloatOrDoubleArray& outtan = curve->getOutTangentValues();
float fps = (float)FPS;
size_t dim = curve->getOutDimension();
unsigned int i;
std::vector<FCurve*>& fcurves = curve_map[curve->getUniqueId()];
switch (dim) {
case 1: // X, Y, Z or angle
case 3: // XYZ
case 16: // matrix
{
for (i = 0; i < dim; i++ ) {
FCurve *fcu = (FCurve*)MEM_callocN(sizeof(FCurve), "FCurve");
fcu->flag = (FCURVE_VISIBLE|FCURVE_AUTO_HANDLES|FCURVE_SELECTED);
// fcu->rna_path = BLI_strdupn(path, strlen(path));
fcu->array_index = 0;
//fcu->totvert = curve->getKeyCount();
// create beztriple for each key
for (unsigned int j = 0; j < curve->getKeyCount(); j++) {
BezTriple bez;
memset(&bez, 0, sizeof(BezTriple));
// intangent
// bez.vec[0][0] = get_float_value(intan, j * 6 + i + i) * fps;
// bez.vec[0][1] = get_float_value(intan, j * 6 + i + i + 1);
// input, output
bez.vec[1][0] = get_float_value(input, j) * fps;
bez.vec[1][1] = get_float_value(output, j * dim + i);
// outtangent
// bez.vec[2][0] = get_float_value(outtan, j * 6 + i + i) * fps;
// bez.vec[2][1] = get_float_value(outtan, j * 6 + i + i + 1);
bez.ipo = U.ipo_new; /* use default interpolation mode here... */
bez.f1 = bez.f2 = bez.f3 = SELECT;
bez.h1 = bez.h2 = HD_AUTO;
insert_bezt_fcurve(fcu, &bez, 0);
}
calchandles_fcurve(fcu);
fcurves.push_back(fcu);
}
}
break;
default:
fprintf(stderr, "Output dimension of %d is not yet supported (animation id = %s)\n", dim, curve->getOriginalId().c_str());
}
for (std::vector<FCurve*>::iterator it = fcurves.begin(); it != fcurves.end(); it++)
unused_curves.push_back(*it);
}
void fcurve_deg_to_rad(FCurve *cu)
{
for (unsigned int i = 0; i < cu->totvert; i++) {
// TODO convert handles too
cu->bezt[i].vec[1][1] *= M_PI / 180.0f;
}
}
void add_fcurves_to_object(Object *ob, std::vector<FCurve*>& curves, char *rna_path, int array_index, Animation *animated)
{
bAction *act;
if (!ob->adt || !ob->adt->action) act = verify_adt_action((ID*)&ob->id, 1);
else act = ob->adt->action;
std::vector<FCurve*>::iterator it;
int i;
#if 0
char *p = strstr(rna_path, "rotation_euler");
bool is_rotation = p && *(p + strlen("rotation_euler")) == '\0';
// convert degrees to radians for rotation
if (is_rotation)
fcurve_deg_to_rad(fcu);
#endif
for (it = curves.begin(), i = 0; it != curves.end(); it++, i++) {
FCurve *fcu = *it;
fcu->rna_path = BLI_strdupn(rna_path, strlen(rna_path));
if (array_index == -1) fcu->array_index = i;
else fcu->array_index = array_index;
if (ob->type == OB_ARMATURE) {
bActionGroup *grp = NULL;
const char *bone_name = get_joint_name(animated->node);
if (bone_name) {
/* try to find group */
grp = action_groups_find_named(act, bone_name);
/* no matching groups, so add one */
if (grp == NULL) {
/* Add a new group, and make it active */
grp = (bActionGroup*)MEM_callocN(sizeof(bActionGroup), "bActionGroup");
grp->flag = AGRP_SELECTED;
BLI_strncpy(grp->name, bone_name, sizeof(grp->name));
BLI_addtail(&act->groups, grp);
BLI_uniquename(&act->groups, grp, "Group", '.', offsetof(bActionGroup, name), 64);
}
/* add F-Curve to group */
action_groups_add_channel(act, grp, fcu);
}
#if 0
if (is_rotation) {
fcurves_actionGroup_map[grp].push_back(fcu);
}
#endif
}
else {
BLI_addtail(&act->curves, fcu);
}
// curve is used, so remove it from unused_curves
unused_curves.erase(std::remove(unused_curves.begin(), unused_curves.end(), fcu), unused_curves.end());
}
}
public:
AnimationImporter(UnitConverter *conv, ArmatureImporter *arm, Scene *scene) :
TransformReader(conv), armature_importer(arm), scene(scene) { }
~AnimationImporter()
{
// free unused FCurves
for (std::vector<FCurve*>::iterator it = unused_curves.begin(); it != unused_curves.end(); it++)
free_fcurve(*it);
if (unused_curves.size())
fprintf(stderr, "removed %u unused curves\n", unused_curves.size());
}
bool write_animation(const COLLADAFW::Animation* anim)
{
if (anim->getAnimationType() == COLLADAFW::Animation::ANIMATION_CURVE) {
COLLADAFW::AnimationCurve *curve = (COLLADAFW::AnimationCurve*)anim;
// XXX Don't know if it's necessary
// Should we check outPhysicalDimension?
if (curve->getInPhysicalDimension() != COLLADAFW::PHYSICAL_DIMENSION_TIME) {
fprintf(stderr, "Inputs physical dimension is not time. \n");
return true;
}
// a curve can have mixed interpolation type,
// in this case curve->getInterpolationTypes returns a list of interpolation types per key
COLLADAFW::AnimationCurve::InterpolationType interp = curve->getInterpolationType();
if (interp != COLLADAFW::AnimationCurve::INTERPOLATION_MIXED) {
switch (interp) {
case COLLADAFW::AnimationCurve::INTERPOLATION_LINEAR:
case COLLADAFW::AnimationCurve::INTERPOLATION_BEZIER:
animation_to_fcurves(curve);
break;
default:
// TODO there're also CARDINAL, HERMITE, BSPLINE and STEP types
fprintf(stderr, "CARDINAL, HERMITE, BSPLINE and STEP anim interpolation types not supported yet.\n");
break;
}
}
else {
// not supported yet
fprintf(stderr, "MIXED anim interpolation type is not supported yet.\n");
}
}
else {
fprintf(stderr, "FORMULA animation type is not supported yet.\n");
}
return true;
}
// called on post-process stage after writeVisualScenes
bool write_animation_list(const COLLADAFW::AnimationList* animlist)
{
const COLLADAFW::UniqueId& animlist_id = animlist->getUniqueId();
animlist_map[animlist_id] = animlist;
#if 0
// should not happen
if (uid_animated_map.find(animlist_id) == uid_animated_map.end()) {
return true;
}
// for bones rna_path is like: pose.bones["bone-name"].rotation
// what does this AnimationList animate?
Animation& animated = uid_animated_map[animlist_id];
Object *ob = animated.ob;
char rna_path[100];
char joint_path[100];
bool is_joint = false;
// if ob is NULL, it should be a JOINT
if (!ob) {
ob = armature_importer->get_armature_for_joint(animated.node);
if (!ob) {
fprintf(stderr, "Cannot find armature for node %s\n", get_joint_name(animated.node));
return true;
}
armature_importer->get_rna_path_for_joint(animated.node, joint_path, sizeof(joint_path));
is_joint = true;
}
const COLLADAFW::AnimationList::AnimationBindings& bindings = animlist->getAnimationBindings();
switch (animated.tm->getTransformationType()) {
case COLLADAFW::Transformation::TRANSLATE:
case COLLADAFW::Transformation::SCALE:
{
bool loc = animated.tm->getTransformationType() == COLLADAFW::Transformation::TRANSLATE;
if (is_joint)
BLI_snprintf(rna_path, sizeof(rna_path), "%s.%s", joint_path, loc ? "location" : "scale");
else
BLI_strncpy(rna_path, loc ? "location" : "scale", sizeof(rna_path));
for (int i = 0; i < bindings.getCount(); i++) {
const COLLADAFW::AnimationList::AnimationBinding& binding = bindings[i];
COLLADAFW::UniqueId anim_uid = binding.animation;
if (curve_map.find(anim_uid) == curve_map.end()) {
fprintf(stderr, "Cannot find FCurve by animation UID.\n");
continue;
}
std::vector<FCurve*>& fcurves = curve_map[anim_uid];
switch (binding.animationClass) {
case COLLADAFW::AnimationList::POSITION_X:
add_fcurves_to_object(ob, fcurves, rna_path, 0, &animated);
break;
case COLLADAFW::AnimationList::POSITION_Y:
add_fcurves_to_object(ob, fcurves, rna_path, 1, &animated);
break;
case COLLADAFW::AnimationList::POSITION_Z:
add_fcurves_to_object(ob, fcurves, rna_path, 2, &animated);
break;
case COLLADAFW::AnimationList::POSITION_XYZ:
add_fcurves_to_object(ob, fcurves, rna_path, -1, &animated);
break;
default:
fprintf(stderr, "AnimationClass %d is not supported for %s.\n",
binding.animationClass, loc ? "TRANSLATE" : "SCALE");
}
}
}
break;
case COLLADAFW::Transformation::ROTATE:
{
if (is_joint)
BLI_snprintf(rna_path, sizeof(rna_path), "%s.rotation_euler", joint_path);
else
BLI_strncpy(rna_path, "rotation_euler", sizeof(rna_path));
COLLADAFW::Rotate* rot = (COLLADAFW::Rotate*)animated.tm;
COLLADABU::Math::Vector3& axis = rot->getRotationAxis();
for (int i = 0; i < bindings.getCount(); i++) {
const COLLADAFW::AnimationList::AnimationBinding& binding = bindings[i];
COLLADAFW::UniqueId anim_uid = binding.animation;
if (curve_map.find(anim_uid) == curve_map.end()) {
fprintf(stderr, "Cannot find FCurve by animation UID.\n");
continue;
}
std::vector<FCurve*>& fcurves = curve_map[anim_uid];
switch (binding.animationClass) {
case COLLADAFW::AnimationList::ANGLE:
if (COLLADABU::Math::Vector3::UNIT_X == axis) {
add_fcurves_to_object(ob, fcurves, rna_path, 0, &animated);
}
else if (COLLADABU::Math::Vector3::UNIT_Y == axis) {
add_fcurves_to_object(ob, fcurves, rna_path, 1, &animated);
}
else if (COLLADABU::Math::Vector3::UNIT_Z == axis) {
add_fcurves_to_object(ob, fcurves, rna_path, 2, &animated);
}
break;
case COLLADAFW::AnimationList::AXISANGLE:
// TODO convert axis-angle to quat? or XYZ?
default:
fprintf(stderr, "AnimationClass %d is not supported for ROTATE transformation.\n",
binding.animationClass);
}
}
}
break;
case COLLADAFW::Transformation::MATRIX:
case COLLADAFW::Transformation::SKEW:
case COLLADAFW::Transformation::LOOKAT:
fprintf(stderr, "Animation of MATRIX, SKEW and LOOKAT transformations is not supported yet.\n");
break;
}
#endif
return true;
}
void read_node_transform(COLLADAFW::Node *node, Object *ob)
{
float mat[4][4];
TransformReader::get_node_mat(mat, node, &uid_animated_map, ob);
if (ob) {
copy_m4_m4(ob->obmat, mat);
object_apply_mat4(ob, ob->obmat);
}
}
#if 0
virtual void change_eul_to_quat(Object *ob, bAction *act)
{
bActionGroup *grp;
int i;
for (grp = (bActionGroup*)act->groups.first; grp; grp = grp->next) {
FCurve *eulcu[3] = {NULL, NULL, NULL};
if (fcurves_actionGroup_map.find(grp) == fcurves_actionGroup_map.end())
continue;
std::vector<FCurve*> &rot_fcurves = fcurves_actionGroup_map[grp];
if (rot_fcurves.size() > 3) continue;
for (i = 0; i < rot_fcurves.size(); i++)
eulcu[rot_fcurves[i]->array_index] = rot_fcurves[i];
char joint_path[100];
char rna_path[100];
BLI_snprintf(joint_path, sizeof(joint_path), "pose.bones[\"%s\"]", grp->name);
BLI_snprintf(rna_path, sizeof(rna_path), "%s.rotation_quaternion", joint_path);
FCurve *quatcu[4] = {
create_fcurve(0, rna_path),
create_fcurve(1, rna_path),
create_fcurve(2, rna_path),
create_fcurve(3, rna_path)
};
bPoseChannel *chan = get_pose_channel(ob->pose, grp->name);
float m4[4][4], irest[3][3];
invert_m4_m4(m4, chan->bone->arm_mat);
copy_m3_m4(irest, m4);
for (i = 0; i < 3; i++) {
FCurve *cu = eulcu[i];
if (!cu) continue;
for (int j = 0; j < cu->totvert; j++) {
float frame = cu->bezt[j].vec[1][0];
float eul[3] = {
eulcu[0] ? evaluate_fcurve(eulcu[0], frame) : 0.0f,
eulcu[1] ? evaluate_fcurve(eulcu[1], frame) : 0.0f,
eulcu[2] ? evaluate_fcurve(eulcu[2], frame) : 0.0f
};
// make eul relative to bone rest pose
float rot[3][3], rel[3][3], quat[4];
/*eul_to_mat3(rot, eul);
mul_m3_m3m3(rel, irest, rot);
mat3_to_quat(quat, rel);
*/
eul_to_quat(quat, eul);
for (int k = 0; k < 4; k++)
create_bezt(quatcu[k], frame, quat[k]);
}
}
// now replace old Euler curves
for (i = 0; i < 3; i++) {
if (!eulcu[i]) continue;
action_groups_remove_channel(act, eulcu[i]);
free_fcurve(eulcu[i]);
}
chan->rotmode = ROT_MODE_QUAT;
for (i = 0; i < 4; i++)
action_groups_add_channel(act, grp, quatcu[i]);
}
bPoseChannel *pchan;
for (pchan = (bPoseChannel*)ob->pose->chanbase.first; pchan; pchan = pchan->next) {
pchan->rotmode = ROT_MODE_QUAT;
}
}
#endif
// prerequisites:
// animlist_map - map animlist id -> animlist
// curve_map - map anim id -> curve(s)
Object *translate_animation(COLLADAFW::Node *node,
std::map<COLLADAFW::UniqueId, Object*>& object_map,
std::map<COLLADAFW::UniqueId, COLLADAFW::Node*>& root_map,
COLLADAFW::Transformation::TransformationType tm_type,
Object *par_job = NULL)
{
bool is_rotation = tm_type == COLLADAFW::Transformation::ROTATE;
bool is_matrix = tm_type == COLLADAFW::Transformation::MATRIX;
bool is_joint = node->getType() == COLLADAFW::Node::JOINT;
COLLADAFW::Node *root = root_map.find(node->getUniqueId()) == root_map.end() ? node : root_map[node->getUniqueId()];
Object *ob = is_joint ? armature_importer->get_armature_for_joint(node) : object_map[node->getUniqueId()];
const char *bone_name = is_joint ? get_joint_name(node) : NULL;
if (!ob) {
fprintf(stderr, "cannot find Object for Node with id=\"%s\"\n", node->getOriginalId().c_str());
return NULL;
}
// frames at which to sample
std::vector<float> frames;
// for each <rotate>, <translate>, etc. there is a separate Transformation
const COLLADAFW::TransformationPointerArray& tms = node->getTransformations();
unsigned int i;
// find frames at which to sample plus convert all rotation keys to radians
for (i = 0; i < tms.getCount(); i++) {
COLLADAFW::Transformation *tm = tms[i];
COLLADAFW::Transformation::TransformationType type = tm->getTransformationType();
if (type == tm_type) {
const COLLADAFW::UniqueId& listid = tm->getAnimationList();
if (animlist_map.find(listid) != animlist_map.end()) {
const COLLADAFW::AnimationList *animlist = animlist_map[listid];
const COLLADAFW::AnimationList::AnimationBindings& bindings = animlist->getAnimationBindings();
if (bindings.getCount()) {
for (unsigned int j = 0; j < bindings.getCount(); j++) {
std::vector<FCurve*>& curves = curve_map[bindings[j].animation];
bool xyz = ((type == COLLADAFW::Transformation::TRANSLATE || type == COLLADAFW::Transformation::SCALE) && bindings[j].animationClass == COLLADAFW::AnimationList::POSITION_XYZ);
if ((!xyz && curves.size() == 1) || (xyz && curves.size() == 3) || is_matrix) {
std::vector<FCurve*>::iterator iter;
for (iter = curves.begin(); iter != curves.end(); iter++) {
FCurve *fcu = *iter;
if (is_rotation)
fcurve_deg_to_rad(fcu);
for (unsigned int k = 0; k < fcu->totvert; k++) {
float fra = fcu->bezt[k].vec[1][0];
if (std::find(frames.begin(), frames.end(), fra) == frames.end())
frames.push_back(fra);
}
}
}
else {
fprintf(stderr, "expected %d curves, got %u\n", xyz ? 3 : 1, curves.size());
}
}
}
}
}
}
float irest_dae[4][4];
float rest[4][4], irest[4][4];
if (is_joint) {
get_joint_rest_mat(irest_dae, root, node);
invert_m4(irest_dae);
Bone *bone = get_named_bone((bArmature*)ob->data, bone_name);
if (!bone) {
fprintf(stderr, "cannot find bone \"%s\"\n", bone_name);
return NULL;
}
unit_m4(rest);
copy_m4_m4(rest, bone->arm_mat);
invert_m4_m4(irest, rest);
}
Object *job = NULL;
#ifdef ARMATURE_TEST
FCurve *job_curves[10];
job = get_joint_object(root, node, par_job);
#endif
if (frames.size() == 0)
return job;
std::sort(frames.begin(), frames.end());
const char *tm_str = NULL;
switch (tm_type) {
case COLLADAFW::Transformation::ROTATE:
tm_str = "rotation_quaternion";
break;
case COLLADAFW::Transformation::SCALE:
tm_str = "scale";
break;
case COLLADAFW::Transformation::TRANSLATE:
tm_str = "location";
break;
case COLLADAFW::Transformation::MATRIX:
break;
default:
return job;
}
char rna_path[200];
char joint_path[200];
if (is_joint)
armature_importer->get_rna_path_for_joint(node, joint_path, sizeof(joint_path));
// new curves
FCurve *newcu[10]; // if tm_type is matrix, then create 10 curves: 4 rot, 3 loc, 3 scale
unsigned int totcu = is_matrix ? 10 : (is_rotation ? 4 : 3);
for (i = 0; i < totcu; i++) {
int axis = i;
if (is_matrix) {
if (i < 4) {
tm_str = "rotation_quaternion";
axis = i;
}
else if (i < 7) {
tm_str = "location";
axis = i - 4;
}
else {
tm_str = "scale";
axis = i - 7;
}
}
if (is_joint)
BLI_snprintf(rna_path, sizeof(rna_path), "%s.%s", joint_path, tm_str);
else
strcpy(rna_path, tm_str);
newcu[i] = create_fcurve(axis, rna_path);
#ifdef ARMATURE_TEST
if (is_joint)
job_curves[i] = create_fcurve(axis, tm_str);
#endif
}
std::vector<float>::iterator it;
// sample values at each frame
for (it = frames.begin(); it != frames.end(); it++) {
float fra = *it;
float mat[4][4];
float matfra[4][4];
unit_m4(matfra);
// calc object-space mat
evaluate_transform_at_frame(matfra, node, fra);
// for joints, we need a special matrix
if (is_joint) {
// special matrix: iR * M * iR_dae * R
// where R, iR are bone rest and inverse rest mats in world space (Blender bones),
// iR_dae is joint inverse rest matrix (DAE) and M is an evaluated joint world-space matrix (DAE)
float temp[4][4], par[4][4];
// calc M
calc_joint_parent_mat_rest(par, NULL, root, node);
mul_m4_m4m4(temp, matfra, par);
// evaluate_joint_world_transform_at_frame(temp, NULL, , node, fra);
// calc special matrix
mul_serie_m4(mat, irest, temp, irest_dae, rest, NULL, NULL, NULL, NULL);
}
else {
copy_m4_m4(mat, matfra);
}
float val[4], rot[4], loc[3], scale[3];
switch (tm_type) {
case COLLADAFW::Transformation::ROTATE:
mat4_to_quat(val, mat);
break;
case COLLADAFW::Transformation::SCALE:
mat4_to_size(val, mat);
break;
case COLLADAFW::Transformation::TRANSLATE:
copy_v3_v3(val, mat[3]);
break;
case COLLADAFW::Transformation::MATRIX:
mat4_to_quat(rot, mat);
copy_v3_v3(loc, mat[3]);
mat4_to_size(scale, mat);
break;
default:
break;
}
// add keys
for (i = 0; i < totcu; i++) {
if (is_matrix) {
if (i < 4)
add_bezt(newcu[i], fra, rot[i]);
else if (i < 7)
add_bezt(newcu[i], fra, loc[i - 4]);
else
add_bezt(newcu[i], fra, scale[i - 7]);
}
else {
add_bezt(newcu[i], fra, val[i]);
}
}
#ifdef ARMATURE_TEST
if (is_joint) {
switch (tm_type) {
case COLLADAFW::Transformation::ROTATE:
mat4_to_quat(val, matfra);
break;
case COLLADAFW::Transformation::SCALE:
mat4_to_size(val, matfra);
break;
case COLLADAFW::Transformation::TRANSLATE:
copy_v3_v3(val, matfra[3]);
break;
case MATRIX:
mat4_to_quat(rot, matfra);
copy_v3_v3(loc, matfra[3]);
mat4_to_size(scale, matfra);
break;
default:
break;
}
for (i = 0; i < totcu; i++) {
if (is_matrix) {
if (i < 4)
add_bezt(job_curves[i], fra, rot[i]);
else if (i < 7)
add_bezt(job_curves[i], fra, loc[i - 4]);
else
add_bezt(job_curves[i], fra, scale[i - 7]);
}
else {
add_bezt(job_curves[i], fra, val[i]);
}
}
}
#endif
}
verify_adt_action((ID*)&ob->id, 1);
ListBase *curves = &ob->adt->action->curves;
// add curves
for (i = 0; i < totcu; i++) {
if (is_joint)
add_bone_fcurve(ob, node, newcu[i]);
else
BLI_addtail(curves, newcu[i]);
#ifdef ARMATURE_TEST
if (is_joint)
BLI_addtail(&job->adt->action->curves, job_curves[i]);
#endif
}
if (is_rotation || is_matrix) {
if (is_joint) {
bPoseChannel *chan = get_pose_channel(ob->pose, bone_name);
chan->rotmode = ROT_MODE_QUAT;
}
else {
ob->rotmode = ROT_MODE_QUAT;
}
}
return job;
}
// internal, better make it private
// warning: evaluates only rotation
// prerequisites: animlist_map, curve_map
void evaluate_transform_at_frame(float mat[4][4], COLLADAFW::Node *node, float fra)
{
const COLLADAFW::TransformationPointerArray& tms = node->getTransformations();
unit_m4(mat);
for (unsigned int i = 0; i < tms.getCount(); i++) {
COLLADAFW::Transformation *tm = tms[i];
COLLADAFW::Transformation::TransformationType type = tm->getTransformationType();
float m[4][4];
unit_m4(m);
if (!evaluate_animation(tm, m, fra, node->getOriginalId().c_str())) {
switch (type) {
case COLLADAFW::Transformation::ROTATE:
dae_rotate_to_mat4(tm, m);
break;
case COLLADAFW::Transformation::TRANSLATE:
dae_translate_to_mat4(tm, m);
break;
case COLLADAFW::Transformation::SCALE:
dae_scale_to_mat4(tm, m);
break;
case COLLADAFW::Transformation::MATRIX:
dae_matrix_to_mat4(tm, m);
break;
default:
fprintf(stderr, "unsupported transformation type %d\n", type);
}
}
float temp[4][4];
copy_m4_m4(temp, mat);
mul_m4_m4m4(mat, m, temp);
}
}
// return true to indicate that mat contains a sane value
bool evaluate_animation(COLLADAFW::Transformation *tm, float mat[4][4], float fra, const char *node_id)
{
const COLLADAFW::UniqueId& listid = tm->getAnimationList();
COLLADAFW::Transformation::TransformationType type = tm->getTransformationType();
if (type != COLLADAFW::Transformation::ROTATE &&
type != COLLADAFW::Transformation::SCALE &&
type != COLLADAFW::Transformation::TRANSLATE &&
type != COLLADAFW::Transformation::MATRIX) {
fprintf(stderr, "animation of transformation %d is not supported yet\n", type);
return false;
}
if (animlist_map.find(listid) == animlist_map.end())
return false;
const COLLADAFW::AnimationList *animlist = animlist_map[listid];
const COLLADAFW::AnimationList::AnimationBindings& bindings = animlist->getAnimationBindings();
if (bindings.getCount()) {
float vec[3];
bool is_scale = (type == COLLADAFW::Transformation::SCALE);
bool is_translate = (type == COLLADAFW::Transformation::TRANSLATE);
if (type == COLLADAFW::Transformation::SCALE)
dae_scale_to_v3(tm, vec);
else if (type == COLLADAFW::Transformation::TRANSLATE)
dae_translate_to_v3(tm, vec);
for (unsigned int j = 0; j < bindings.getCount(); j++) {
const COLLADAFW::AnimationList::AnimationBinding& binding = bindings[j];
std::vector<FCurve*>& curves = curve_map[binding.animation];
COLLADAFW::AnimationList::AnimationClass animclass = binding.animationClass;
char path[100];
switch (type) {
case COLLADAFW::Transformation::ROTATE:
BLI_snprintf(path, sizeof(path), "%s.rotate (binding %u)", node_id, j);
break;
case COLLADAFW::Transformation::SCALE:
BLI_snprintf(path, sizeof(path), "%s.scale (binding %u)", node_id, j);
break;
case COLLADAFW::Transformation::TRANSLATE:
BLI_snprintf(path, sizeof(path), "%s.translate (binding %u)", node_id, j);
break;
case COLLADAFW::Transformation::MATRIX:
BLI_snprintf(path, sizeof(path), "%s.matrix (binding %u)", node_id, j);
break;
default:
break;
}
if (animclass == COLLADAFW::AnimationList::UNKNOWN_CLASS) {
fprintf(stderr, "%s: UNKNOWN animation class\n", path);
continue;
}
if (type == COLLADAFW::Transformation::ROTATE) {
if (curves.size() != 1) {
fprintf(stderr, "expected 1 curve, got %u\n", curves.size());
return false;
}
// TODO support other animclasses
if (animclass != COLLADAFW::AnimationList::ANGLE) {
fprintf(stderr, "%s: animation class %d is not supported yet\n", path, animclass);
return false;
}
COLLADABU::Math::Vector3& axis = ((COLLADAFW::Rotate*)tm)->getRotationAxis();
float ax[3] = {axis[0], axis[1], axis[2]};
float angle = evaluate_fcurve(curves[0], fra);
axis_angle_to_mat4(mat, ax, angle);
return true;
}
else if (is_scale || is_translate) {
bool is_xyz = animclass == COLLADAFW::AnimationList::POSITION_XYZ;
if ((!is_xyz && curves.size() != 1) || (is_xyz && curves.size() != 3)) {
if (is_xyz)
fprintf(stderr, "%s: expected 3 curves, got %u\n", path, curves.size());
else
fprintf(stderr, "%s: expected 1 curve, got %u\n", path, curves.size());
return false;
}
switch (animclass) {
case COLLADAFW::AnimationList::POSITION_X:
vec[0] = evaluate_fcurve(curves[0], fra);
break;
case COLLADAFW::AnimationList::POSITION_Y:
vec[1] = evaluate_fcurve(curves[0], fra);
break;
case COLLADAFW::AnimationList::POSITION_Z:
vec[2] = evaluate_fcurve(curves[0], fra);
break;
case COLLADAFW::AnimationList::POSITION_XYZ:
vec[0] = evaluate_fcurve(curves[0], fra);
vec[1] = evaluate_fcurve(curves[1], fra);
vec[2] = evaluate_fcurve(curves[2], fra);
break;
default:
fprintf(stderr, "%s: animation class %d is not supported yet\n", path, animclass);
break;
}
}
else if (type == COLLADAFW::Transformation::MATRIX) {
// for now, of matrix animation, support only the case when all values are packed into one animation
if (curves.size() != 16) {
fprintf(stderr, "%s: expected 16 curves, got %u\n", path, curves.size());
return false;
}
COLLADABU::Math::Matrix4 matrix;
int i = 0, j = 0;
for (std::vector<FCurve*>::iterator it = curves.begin(); it != curves.end(); it++) {
matrix.setElement(i, j, evaluate_fcurve(*it, fra));
j++;
if (j == 4) {
i++;
j = 0;
}
}
COLLADAFW::Matrix tm(matrix);
dae_matrix_to_mat4(&tm, mat);
return true;
}
}
if (is_scale)
size_to_mat4(mat, vec);
else
copy_v3_v3(mat[3], vec);
return is_scale || is_translate;
}
return false;
}
// gives a world-space mat of joint at rest position
void get_joint_rest_mat(float mat[4][4], COLLADAFW::Node *root, COLLADAFW::Node *node)
{
// if bind mat is not available,
// use "current" node transform, i.e. all those tms listed inside <node>
if (!armature_importer->get_joint_bind_mat(mat, node)) {
float par[4][4], m[4][4];
calc_joint_parent_mat_rest(par, NULL, root, node);
get_node_mat(m, node, NULL, NULL);
mul_m4_m4m4(mat, m, par);
}
}
// gives a world-space mat, end's mat not included
bool calc_joint_parent_mat_rest(float mat[4][4], float par[4][4], COLLADAFW::Node *node, COLLADAFW::Node *end)
{
float m[4][4];
if (node == end) {
par ? copy_m4_m4(mat, par) : unit_m4(mat);
return true;
}
// use bind matrix if available or calc "current" world mat
if (!armature_importer->get_joint_bind_mat(m, node)) {
if (par) {
float temp[4][4];
get_node_mat(temp, node, NULL, NULL);
mul_m4_m4m4(m, temp, par);
}
else {
get_node_mat(m, node, NULL, NULL);
}
}
COLLADAFW::NodePointerArray& children = node->getChildNodes();
for (unsigned int i = 0; i < children.getCount(); i++) {
if (calc_joint_parent_mat_rest(mat, m, children[i], end))
return true;
}
return false;
}
#ifdef ARMATURE_TEST
Object *get_joint_object(COLLADAFW::Node *root, COLLADAFW::Node *node, Object *par_job)
{
if (joint_objects.find(node->getUniqueId()) == joint_objects.end()) {
Object *job = add_object(scene, OB_EMPTY);
rename_id((ID*)&job->id, (char*)get_joint_name(node));
job->lay = object_in_scene(job, scene)->lay = 2;
mul_v3_fl(job->size, 0.5f);
job->recalc |= OB_RECALC_OB;
verify_adt_action((ID*)&job->id, 1);
job->rotmode = ROT_MODE_QUAT;
float mat[4][4];
get_joint_rest_mat(mat, root, node);
if (par_job) {
float temp[4][4], ipar[4][4];
invert_m4_m4(ipar, par_job->obmat);
copy_m4_m4(temp, mat);
mul_m4_m4m4(mat, temp, ipar);
}
TransformBase::decompose(mat, job->loc, NULL, job->quat, job->size);
if (par_job) {
job->parent = par_job;
par_job->recalc |= OB_RECALC_OB;
job->parsubstr[0] = 0;
}
where_is_object(scene, job);
// after parenting and layer change
DAG_scene_sort(CTX_data_main(C), scene);
joint_objects[node->getUniqueId()] = job;
}
return joint_objects[node->getUniqueId()];
}
#endif
#if 0
// recursively evaluates joint tree until end is found, mat then is world-space matrix of end
// mat must be identity on enter, node must be root
bool evaluate_joint_world_transform_at_frame(float mat[4][4], float par[4][4], COLLADAFW::Node *node, COLLADAFW::Node *end, float fra)
{
float m[4][4];
if (par) {
float temp[4][4];
evaluate_transform_at_frame(temp, node, node == end ? fra : 0.0f);
mul_m4_m4m4(m, temp, par);
}
else {
evaluate_transform_at_frame(m, node, node == end ? fra : 0.0f);
}
if (node == end) {
copy_m4_m4(mat, m);
return true;
}
else {
COLLADAFW::NodePointerArray& children = node->getChildNodes();
for (int i = 0; i < children.getCount(); i++) {
if (evaluate_joint_world_transform_at_frame(mat, m, children[i], end, fra))
return true;
}
}
return false;
}
#endif
void add_bone_fcurve(Object *ob, COLLADAFW::Node *node, FCurve *fcu)
{
const char *bone_name = get_joint_name(node);
bAction *act = ob->adt->action;
/* try to find group */
bActionGroup *grp = action_groups_find_named(act, bone_name);
/* no matching groups, so add one */
if (grp == NULL) {
/* Add a new group, and make it active */
grp = (bActionGroup*)MEM_callocN(sizeof(bActionGroup), "bActionGroup");
grp->flag = AGRP_SELECTED;
BLI_strncpy(grp->name, bone_name, sizeof(grp->name));
BLI_addtail(&act->groups, grp);
BLI_uniquename(&act->groups, grp, "Group", '.', offsetof(bActionGroup, name), 64);
}
/* add F-Curve to group */
action_groups_add_channel(act, grp, fcu);
}
void add_bezt(FCurve *fcu, float fra, float value)
{
BezTriple bez;
memset(&bez, 0, sizeof(BezTriple));
bez.vec[1][0] = fra;
bez.vec[1][1] = value;
bez.ipo = U.ipo_new; /* use default interpolation mode here... */
bez.f1 = bez.f2 = bez.f3 = SELECT;
bez.h1 = bez.h2 = HD_AUTO;
insert_bezt_fcurve(fcu, &bez, 0);
calchandles_fcurve(fcu);
}
};
/*
COLLADA Importer limitations:
- no multiple scene import, all objects are added to active scene
*/
/** Class that needs to be implemented by a writer.
IMPORTANT: The write functions are called in arbitrary order.*/
class Writer: public COLLADAFW::IWriter
{
private:
std::string mFilename;
bContext *mContext;
UnitConverter unit_converter;
ArmatureImporter armature_importer;
MeshImporter mesh_importer;
AnimationImporter anim_importer;
std::map<COLLADAFW::UniqueId, Image*> uid_image_map;
std::map<COLLADAFW::UniqueId, Material*> uid_material_map;
std::map<COLLADAFW::UniqueId, Material*> uid_effect_map;
std::map<COLLADAFW::UniqueId, Camera*> uid_camera_map;
std::map<COLLADAFW::UniqueId, Lamp*> uid_lamp_map;
std::map<Material*, TexIndexTextureArrayMap> material_texture_mapping_map;
std::map<COLLADAFW::UniqueId, Object*> object_map;
std::map<COLLADAFW::UniqueId, COLLADAFW::Node*> node_map;
std::vector<const COLLADAFW::VisualScene*> vscenes;
std::vector<Object*> libnode_ob;
std::map<COLLADAFW::UniqueId, COLLADAFW::Node*> root_map; // find root joint by child joint uid, for bone tree evaluation during resampling
// animation
// std::map<COLLADAFW::UniqueId, std::vector<FCurve*> > uid_fcurve_map;
// Nodes don't share AnimationLists (Arystan)
// std::map<COLLADAFW::UniqueId, Animation> uid_animated_map; // AnimationList->uniqueId to AnimatedObject map
public:
/** Constructor. */
Writer(bContext *C, const char *filename) : mFilename(filename), mContext(C),
armature_importer(&unit_converter, &mesh_importer, &anim_importer, CTX_data_scene(C)),
mesh_importer(&armature_importer, CTX_data_scene(C)),
anim_importer(&unit_converter, &armature_importer, CTX_data_scene(C)) {}
/** Destructor. */
~Writer() {}
bool write()
{
COLLADASaxFWL::Loader loader;
COLLADAFW::Root root(&loader, this);
// XXX report error
if (!root.loadDocument(mFilename))
return false;
return true;
}
/** This method will be called if an error in the loading process occurred and the loader cannot
continue to load. The writer should undo all operations that have been performed.
@param errorMessage A message containing informations about the error that occurred.
*/
virtual void cancel(const COLLADAFW::String& errorMessage)
{
// TODO: if possible show error info
//
// Should we get rid of invisible Meshes that were created so far
// or maybe create objects at coordinate space origin?
//
// The latter sounds better.
}
/** This is the method called. The writer hast to prepare to receive data.*/
virtual void start()
{
}
/** This method is called after the last write* method. No other methods will be called after this.*/
virtual void finish()
{
std::vector<const COLLADAFW::VisualScene*>::iterator it;
for (it = vscenes.begin(); it != vscenes.end(); it++) {
// TODO: create a new scene except the selected <visual_scene> - use current blender scene for it
Scene *sce = CTX_data_scene(mContext);
const COLLADAFW::NodePointerArray& roots = (*it)->getRootNodes();
for (unsigned int i = 0; i < roots.getCount(); i++) {
write_node(roots[i], NULL, sce, NULL, false);
}
}
armature_importer.make_armatures(mContext);
#if 0
armature_importer.fix_animation();
#endif
for (std::vector<const COLLADAFW::VisualScene*>::iterator it = vscenes.begin(); it != vscenes.end(); it++) {
const COLLADAFW::NodePointerArray& roots = (*it)->getRootNodes();
for (unsigned int i = 0; i < roots.getCount(); i++)
translate_anim_recursive(roots[i]);
}
if (libnode_ob.size()) {
Scene *sce = CTX_data_scene(mContext);
fprintf(stderr, "got %u library nodes to free\n", libnode_ob.size());
// free all library_nodes
std::vector<Object*>::iterator it;
for (it = libnode_ob.begin(); it != libnode_ob.end(); it++) {
Object *ob = *it;
Base *base = object_in_scene(ob, sce);
if (base) {
BLI_remlink(&sce->base, base);
free_libblock_us(&G.main->object, base->object);
if (sce->basact==base)
sce->basact= NULL;
MEM_freeN(base);
}
}
libnode_ob.clear();
DAG_scene_sort(CTX_data_main(mContext), sce);
DAG_ids_flush_update(CTX_data_main(mContext), 0);
}
}
void translate_anim_recursive(COLLADAFW::Node *node, COLLADAFW::Node *par = NULL, Object *parob = NULL)
{
if (par && par->getType() == COLLADAFW::Node::JOINT) {
// par is root if there's no corresp. key in root_map
if (root_map.find(par->getUniqueId()) == root_map.end())
root_map[node->getUniqueId()] = par;
else
root_map[node->getUniqueId()] = root_map[par->getUniqueId()];
}
COLLADAFW::Transformation::TransformationType types[] = {
COLLADAFW::Transformation::ROTATE,
COLLADAFW::Transformation::SCALE,
COLLADAFW::Transformation::TRANSLATE,
COLLADAFW::Transformation::MATRIX
};
unsigned int i;
Object *ob;
for (i = 0; i < 4; i++)
ob = anim_importer.translate_animation(node, object_map, root_map, types[i]);
COLLADAFW::NodePointerArray &children = node->getChildNodes();
for (i = 0; i < children.getCount(); i++) {
translate_anim_recursive(children[i], node, ob);
}
}
/** When this method is called, the writer must write the global document asset.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeGlobalAsset ( const COLLADAFW::FileInfo* asset )
{
// XXX take up_axis, unit into account
// COLLADAFW::FileInfo::Unit unit = asset->getUnit();
// COLLADAFW::FileInfo::UpAxisType upAxis = asset->getUpAxisType();
unit_converter.read_asset(asset);
return true;
}
/** When this method is called, the writer must write the scene.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeScene ( const COLLADAFW::Scene* scene )
{
// XXX could store the scene id, but do nothing for now
return true;
}
Object *create_camera_object(COLLADAFW::InstanceCamera *camera, Scene *sce)
{
const COLLADAFW::UniqueId& cam_uid = camera->getInstanciatedObjectId();
if (uid_camera_map.find(cam_uid) == uid_camera_map.end()) {
fprintf(stderr, "Couldn't find camera by UID.\n");
return NULL;
}
Object *ob = add_object(sce, OB_CAMERA);
Camera *cam = uid_camera_map[cam_uid];
Camera *old_cam = (Camera*)ob->data;
ob->data = cam;
old_cam->id.us--;
if (old_cam->id.us == 0)
free_libblock(&G.main->camera, old_cam);
return ob;
}
Object *create_lamp_object(COLLADAFW::InstanceLight *lamp, Scene *sce)
{
const COLLADAFW::UniqueId& lamp_uid = lamp->getInstanciatedObjectId();
if (uid_lamp_map.find(lamp_uid) == uid_lamp_map.end()) {
fprintf(stderr, "Couldn't find lamp by UID. \n");
return NULL;
}
Object *ob = add_object(sce, OB_LAMP);
Lamp *la = uid_lamp_map[lamp_uid];
Lamp *old_lamp = (Lamp*)ob->data;
ob->data = la;
old_lamp->id.us--;
if (old_lamp->id.us == 0)
free_libblock(&G.main->lamp, old_lamp);
return ob;
}
Object *create_instance_node(Object *source_ob, COLLADAFW::Node *source_node, COLLADAFW::Node *instance_node, Scene *sce, bool is_library_node)
{
Object *obn = copy_object(source_ob);
obn->recalc |= OB_RECALC_ALL;
scene_add_base(sce, obn);
if (instance_node)
anim_importer.read_node_transform(instance_node, obn);
else
anim_importer.read_node_transform(source_node, obn);
DAG_scene_sort(CTX_data_main(mContext), sce);
DAG_ids_flush_update(CTX_data_main(mContext), 0);
COLLADAFW::NodePointerArray &children = source_node->getChildNodes();
if (children.getCount()) {
for (unsigned int i = 0; i < children.getCount(); i++) {
COLLADAFW::Node *child_node = children[i];
const COLLADAFW::UniqueId& child_id = child_node->getUniqueId();
if (object_map.find(child_id) == object_map.end())
continue;
COLLADAFW::InstanceNodePointerArray &inodes = child_node->getInstanceNodes();
Object *new_child = NULL;
if (inodes.getCount()) {
const COLLADAFW::UniqueId& id = inodes[0]->getInstanciatedObjectId();
new_child = create_instance_node(object_map[id], node_map[id], child_node, sce, is_library_node);
}
else {
new_child = create_instance_node(object_map[child_id], child_node, NULL, sce, is_library_node);
}
set_parent(new_child, obn, mContext, true);
if (is_library_node)
libnode_ob.push_back(new_child);
}
}
return obn;
}
void write_node (COLLADAFW::Node *node, COLLADAFW::Node *parent_node, Scene *sce, Object *par, bool is_library_node)
{
Object *ob = NULL;
bool is_joint = node->getType() == COLLADAFW::Node::JOINT;
if (is_joint) {
armature_importer.add_joint(node, parent_node == NULL || parent_node->getType() != COLLADAFW::Node::JOINT, par);
}
else {
COLLADAFW::InstanceGeometryPointerArray &geom = node->getInstanceGeometries();
COLLADAFW::InstanceCameraPointerArray &camera = node->getInstanceCameras();
COLLADAFW::InstanceLightPointerArray &lamp = node->getInstanceLights();
COLLADAFW::InstanceControllerPointerArray &controller = node->getInstanceControllers();
COLLADAFW::InstanceNodePointerArray &inst_node = node->getInstanceNodes();
// XXX linking object with the first <instance_geometry>, though a node may have more of them...
// maybe join multiple <instance_...> meshes into 1, and link object with it? not sure...
// <instance_geometry>
if (geom.getCount() != 0) {
ob = mesh_importer.create_mesh_object(node, geom[0], false, uid_material_map,
material_texture_mapping_map);
}
else if (camera.getCount() != 0) {
ob = create_camera_object(camera[0], sce);
}
else if (lamp.getCount() != 0) {
ob = create_lamp_object(lamp[0], sce);
}
else if (controller.getCount() != 0) {
COLLADAFW::InstanceGeometry *geom = (COLLADAFW::InstanceGeometry*)controller[0];
ob = mesh_importer.create_mesh_object(node, geom, true, uid_material_map, material_texture_mapping_map);
}
// XXX instance_node is not supported yet
else if (inst_node.getCount() != 0) {
const COLLADAFW::UniqueId& node_id = inst_node[0]->getInstanciatedObjectId();
if (object_map.find(node_id) == object_map.end()) {
fprintf(stderr, "Cannot find node to instanciate.\n");
ob = NULL;
}
else {
Object *source_ob = object_map[node_id];
COLLADAFW::Node *source_node = node_map[node_id];
ob = create_instance_node(source_ob, source_node, node, sce, is_library_node);
}
}
// if node is empty - create empty object
// XXX empty node may not mean it is empty object, not sure about this
else {
ob = add_object(sce, OB_EMPTY);
rename_id(&ob->id, (char*)node->getOriginalId().c_str());
}
// check if object is not NULL
if (!ob) return;
object_map[node->getUniqueId()] = ob;
node_map[node->getUniqueId()] = node;
if (is_library_node)
libnode_ob.push_back(ob);
}
anim_importer.read_node_transform(node, ob);
if (!is_joint) {
// if par was given make this object child of the previous
if (par && ob)
set_parent(ob, par, mContext);
}
// if node has child nodes write them
COLLADAFW::NodePointerArray &child_nodes = node->getChildNodes();
for (unsigned int i = 0; i < child_nodes.getCount(); i++) {
write_node(child_nodes[i], node, sce, ob, is_library_node);
}
}
/** When this method is called, the writer must write the entire visual scene.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeVisualScene ( const COLLADAFW::VisualScene* visualScene )
{
// this method called on post process after writeGeometry, writeMaterial, etc.
// for each <node> in <visual_scene>:
// create an Object
// if Mesh (previously created in writeGeometry) to which <node> corresponds exists, link Object with that mesh
// update: since we cannot link a Mesh with Object in
// writeGeometry because <geometry> does not reference <node>,
// we link Objects with Meshes here
vscenes.push_back(visualScene);
return true;
}
/** When this method is called, the writer must handle all nodes contained in the
library nodes.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeLibraryNodes ( const COLLADAFW::LibraryNodes* libraryNodes )
{
Scene *sce = CTX_data_scene(mContext);
const COLLADAFW::NodePointerArray& nodes = libraryNodes->getNodes();
for (unsigned int i = 0; i < nodes.getCount(); i++) {
write_node(nodes[i], NULL, sce, NULL, true);
}
return true;
}
/** When this method is called, the writer must write the geometry.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeGeometry ( const COLLADAFW::Geometry* geom )
{
return mesh_importer.write_geometry(geom);
}
/** When this method is called, the writer must write the material.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeMaterial( const COLLADAFW::Material* cmat )
{
const std::string& str_mat_id = cmat->getOriginalId();
Material *ma = add_material((char*)str_mat_id.c_str());
this->uid_effect_map[cmat->getInstantiatedEffect()] = ma;
this->uid_material_map[cmat->getUniqueId()] = ma;
return true;
}
// create mtex, create texture, set texture image
MTex *create_texture(COLLADAFW::EffectCommon *ef, COLLADAFW::Texture &ctex, Material *ma,
int i, TexIndexTextureArrayMap &texindex_texarray_map)
{
COLLADAFW::SamplerPointerArray& samp_array = ef->getSamplerPointerArray();
COLLADAFW::Sampler *sampler = samp_array[ctex.getSamplerId()];
const COLLADAFW::UniqueId& ima_uid = sampler->getSourceImage();
if (uid_image_map.find(ima_uid) == uid_image_map.end()) {
fprintf(stderr, "Couldn't find an image by UID.\n");
return NULL;
}
ma->mtex[i] = add_mtex();
ma->mtex[i]->texco = TEXCO_UV;
ma->mtex[i]->tex = add_texture("Texture");
ma->mtex[i]->tex->type = TEX_IMAGE;
ma->mtex[i]->tex->imaflag &= ~TEX_USEALPHA;
ma->mtex[i]->tex->ima = uid_image_map[ima_uid];
texindex_texarray_map[ctex.getTextureMapId()].push_back(ma->mtex[i]);
return ma->mtex[i];
}
void write_profile_COMMON(COLLADAFW::EffectCommon *ef, Material *ma)
{
COLLADAFW::EffectCommon::ShaderType shader = ef->getShaderType();
// blinn
if (shader == COLLADAFW::EffectCommon::SHADER_BLINN) {
ma->spec_shader = MA_SPEC_BLINN;
ma->spec = ef->getShininess().getFloatValue();
}
// phong
else if (shader == COLLADAFW::EffectCommon::SHADER_PHONG) {
ma->spec_shader = MA_SPEC_PHONG;
// XXX setting specular hardness instead of specularity intensity
ma->har = ef->getShininess().getFloatValue() * 4;
}
// lambert
else if (shader == COLLADAFW::EffectCommon::SHADER_LAMBERT) {
ma->diff_shader = MA_DIFF_LAMBERT;
}
// default - lambert
else {
ma->diff_shader = MA_DIFF_LAMBERT;
fprintf(stderr, "Current shader type is not supported.\n");
}
// reflectivity
ma->ray_mirror = ef->getReflectivity().getFloatValue();
// index of refraction
ma->ang = ef->getIndexOfRefraction().getFloatValue();
int i = 0;
COLLADAFW::Color col;
MTex *mtex = NULL;
TexIndexTextureArrayMap texindex_texarray_map;
// DIFFUSE
// color
if (ef->getDiffuse().isColor()) {
col = ef->getDiffuse().getColor();
ma->r = col.getRed();
ma->g = col.getGreen();
ma->b = col.getBlue();
}
// texture
else if (ef->getDiffuse().isTexture()) {
COLLADAFW::Texture ctex = ef->getDiffuse().getTexture();
mtex = create_texture(ef, ctex, ma, i, texindex_texarray_map);
if (mtex != NULL) {
mtex->mapto = MAP_COL;
ma->texact = (int)i;
i++;
}
}
// AMBIENT
// color
if (ef->getAmbient().isColor()) {
col = ef->getAmbient().getColor();
ma->ambr = col.getRed();
ma->ambg = col.getGreen();
ma->ambb = col.getBlue();
}
// texture
else if (ef->getAmbient().isTexture()) {
COLLADAFW::Texture ctex = ef->getAmbient().getTexture();
mtex = create_texture(ef, ctex, ma, i, texindex_texarray_map);
if (mtex != NULL) {
mtex->mapto = MAP_AMB;
i++;
}
}
// SPECULAR
// color
if (ef->getSpecular().isColor()) {
col = ef->getSpecular().getColor();
ma->specr = col.getRed();
ma->specg = col.getGreen();
ma->specb = col.getBlue();
}
// texture
else if (ef->getSpecular().isTexture()) {
COLLADAFW::Texture ctex = ef->getSpecular().getTexture();
mtex = create_texture(ef, ctex, ma, i, texindex_texarray_map);
if (mtex != NULL) {
mtex->mapto = MAP_SPEC;
i++;
}
}
// REFLECTIVE
// color
if (ef->getReflective().isColor()) {
col = ef->getReflective().getColor();
ma->mirr = col.getRed();
ma->mirg = col.getGreen();
ma->mirb = col.getBlue();
}
// texture
else if (ef->getReflective().isTexture()) {
COLLADAFW::Texture ctex = ef->getReflective().getTexture();
mtex = create_texture(ef, ctex, ma, i, texindex_texarray_map);
if (mtex != NULL) {
mtex->mapto = MAP_REF;
i++;
}
}
// EMISSION
// color
if (ef->getEmission().isColor()) {
// XXX there is no emission color in blender
// but I am not sure
}
// texture
else if (ef->getEmission().isTexture()) {
COLLADAFW::Texture ctex = ef->getEmission().getTexture();
mtex = create_texture(ef, ctex, ma, i, texindex_texarray_map);
if (mtex != NULL) {
mtex->mapto = MAP_EMIT;
i++;
}
}
// TRANSPARENT
// color
// if (ef->getOpacity().isColor()) {
// // XXX don't know what to do here
// }
// // texture
// else if (ef->getOpacity().isTexture()) {
// ctex = ef->getOpacity().getTexture();
// if (mtex != NULL) mtex->mapto &= MAP_ALPHA;
// else {
// mtex = create_texture(ef, ctex, ma, i, texindex_texarray_map);
// if (mtex != NULL) mtex->mapto = MAP_ALPHA;
// }
// }
material_texture_mapping_map[ma] = texindex_texarray_map;
}
/** When this method is called, the writer must write the effect.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeEffect( const COLLADAFW::Effect* effect )
{
const COLLADAFW::UniqueId& uid = effect->getUniqueId();
if (uid_effect_map.find(uid) == uid_effect_map.end()) {
fprintf(stderr, "Couldn't find a material by UID.\n");
return true;
}
Material *ma = uid_effect_map[uid];
COLLADAFW::CommonEffectPointerArray common_efs = effect->getCommonEffects();
if (common_efs.getCount() < 1) {
fprintf(stderr, "Couldn't find <profile_COMMON>.\n");
return true;
}
// XXX TODO: Take all <profile_common>s
// Currently only first <profile_common> is supported
COLLADAFW::EffectCommon *ef = common_efs[0];
write_profile_COMMON(ef, ma);
return true;
}
/** When this method is called, the writer must write the camera.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeCamera( const COLLADAFW::Camera* camera )
{
Camera *cam = NULL;
std::string cam_id, cam_name;
cam_id = camera->getOriginalId();
cam_name = camera->getName();
if (cam_name.size()) cam = (Camera*)add_camera((char*)cam_name.c_str());
else cam = (Camera*)add_camera((char*)cam_id.c_str());
if (!cam) {
fprintf(stderr, "Cannot create camera. \n");
return true;
}
cam->clipsta = camera->getNearClippingPlane().getValue();
cam->clipend = camera->getFarClippingPlane().getValue();
COLLADAFW::Camera::CameraType type = camera->getCameraType();
switch(type) {
case COLLADAFW::Camera::ORTHOGRAPHIC:
{
cam->type = CAM_ORTHO;
}
break;
case COLLADAFW::Camera::PERSPECTIVE:
{
cam->type = CAM_PERSP;
}
break;
case COLLADAFW::Camera::UNDEFINED_CAMERATYPE:
{
fprintf(stderr, "Current camera type is not supported. \n");
cam->type = CAM_PERSP;
}
break;
}
this->uid_camera_map[camera->getUniqueId()] = cam;
// XXX import camera options
return true;
}
/** When this method is called, the writer must write the image.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeImage( const COLLADAFW::Image* image )
{
// XXX maybe it is necessary to check if the path is absolute or relative
const std::string& filepath = image->getImageURI().toNativePath();
const char *filename = (const char*)mFilename.c_str();
char dir[FILE_MAX];
char full_path[FILE_MAX];
BLI_split_dirfile(filename, dir, NULL);
BLI_join_dirfile(full_path, dir, filepath.c_str());
Image *ima = BKE_add_image_file(full_path, 0);
if (!ima) {
fprintf(stderr, "Cannot create image. \n");
return true;
}
this->uid_image_map[image->getUniqueId()] = ima;
return true;
}
/** When this method is called, the writer must write the light.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeLight( const COLLADAFW::Light* light )
{
Lamp *lamp = NULL;
std::string la_id, la_name;
la_id = light->getOriginalId();
la_name = light->getName();
if (la_name.size()) lamp = (Lamp*)add_lamp((char*)la_name.c_str());
else lamp = (Lamp*)add_lamp((char*)la_id.c_str());
if (!lamp) {
fprintf(stderr, "Cannot create lamp. \n");
return true;
}
if (light->getColor().isValid()) {
COLLADAFW::Color col = light->getColor();
lamp->r = col.getRed();
lamp->g = col.getGreen();
lamp->b = col.getBlue();
}
COLLADAFW::Light::LightType type = light->getLightType();
switch(type) {
case COLLADAFW::Light::AMBIENT_LIGHT:
{
lamp->type = LA_HEMI;
}
break;
case COLLADAFW::Light::SPOT_LIGHT:
{
lamp->type = LA_SPOT;
lamp->falloff_type = LA_FALLOFF_SLIDERS;
lamp->att1 = light->getLinearAttenuation().getValue();
lamp->att2 = light->getQuadraticAttenuation().getValue();
lamp->spotsize = light->getFallOffAngle().getValue();
lamp->spotblend = light->getFallOffExponent().getValue();
}
break;
case COLLADAFW::Light::DIRECTIONAL_LIGHT:
{
lamp->type = LA_SUN;
}
break;
case COLLADAFW::Light::POINT_LIGHT:
{
lamp->type = LA_LOCAL;
lamp->att1 = light->getLinearAttenuation().getValue();
lamp->att2 = light->getQuadraticAttenuation().getValue();
}
break;
case COLLADAFW::Light::UNDEFINED:
{
fprintf(stderr, "Current lamp type is not supported. \n");
lamp->type = LA_LOCAL;
}
break;
}
this->uid_lamp_map[light->getUniqueId()] = lamp;
return true;
}
// this function is called only for animations that pass COLLADAFW::validate
virtual bool writeAnimation( const COLLADAFW::Animation* anim )
{
// return true;
return anim_importer.write_animation(anim);
}
// called on post-process stage after writeVisualScenes
virtual bool writeAnimationList( const COLLADAFW::AnimationList* animationList )
{
// return true;
return anim_importer.write_animation_list(animationList);
}
/** When this method is called, the writer must write the skin controller data.
@return The writer should return true, if writing succeeded, false otherwise.*/
virtual bool writeSkinControllerData( const COLLADAFW::SkinControllerData* skin )
{
return armature_importer.write_skin_controller_data(skin);
}
// this is called on postprocess, before writeVisualScenes
virtual bool writeController( const COLLADAFW::Controller* controller )
{
return armature_importer.write_controller(controller);
}
virtual bool writeFormulas( const COLLADAFW::Formulas* formulas )
{
return true;
}
virtual bool writeKinematicsScene( const COLLADAFW::KinematicsScene* kinematicsScene )
{
return true;
}
};
void DocumentImporter::import(bContext *C, const char *filename)
{
Writer w(C, filename);
w.write();
}