21210 to 21300 Note to self: the next merge will be messy again, since 2 commits occurred this one was complete
1260 lines
42 KiB
C
1260 lines
42 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., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
|
|
*
|
|
* The Original Code is Copyright (C) 2001-2002 by NaN Holding BV.
|
|
* All rights reserved.
|
|
*
|
|
* This is a new part of Blender.
|
|
*
|
|
* Contributor(s): Joseph Gilbert, Campbell Barton
|
|
*
|
|
* ***** END GPL LICENSE BLOCK *****
|
|
*/
|
|
|
|
#include "Mathutils.h"
|
|
|
|
#include "BLI_arithb.h"
|
|
#include "PIL_time.h"
|
|
#include "BLI_rand.h"
|
|
#include "BKE_utildefines.h"
|
|
|
|
//-------------------------DOC STRINGS ---------------------------
|
|
static char M_Mathutils_doc[] = "The Blender Mathutils module\n\n";
|
|
static char M_Mathutils_Rand_doc[] = "() - return a random number";
|
|
static char M_Mathutils_AngleBetweenVecs_doc[] = "() - returns the angle between two vectors in degrees";
|
|
static char M_Mathutils_MidpointVecs_doc[] = "() - return the vector to the midpoint between two vectors";
|
|
static char M_Mathutils_ProjectVecs_doc[] = "() - returns the projection vector from the projection of vecA onto vecB";
|
|
static char M_Mathutils_RotationMatrix_doc[] = "() - construct a rotation matrix from an angle and axis of rotation";
|
|
static char M_Mathutils_ScaleMatrix_doc[] = "() - construct a scaling matrix from a scaling factor";
|
|
static char M_Mathutils_OrthoProjectionMatrix_doc[] = "() - construct a orthographic projection matrix from a selected plane";
|
|
static char M_Mathutils_ShearMatrix_doc[] = "() - construct a shearing matrix from a plane of shear and a shear factor";
|
|
static char M_Mathutils_TranslationMatrix_doc[] = "(vec) - create a translation matrix from a vector";
|
|
static char M_Mathutils_Slerp_doc[] = "() - returns the interpolation between two quaternions";
|
|
static char M_Mathutils_DifferenceQuats_doc[] = "() - return the angular displacment difference between two quats";
|
|
static char M_Mathutils_Intersect_doc[] = "(v1, v2, v3, ray, orig, clip=1) - returns the intersection between a ray and a triangle, if possible, returns None otherwise";
|
|
static char M_Mathutils_TriangleArea_doc[] = "(v1, v2, v3) - returns the area size of the 2D or 3D triangle defined";
|
|
static char M_Mathutils_TriangleNormal_doc[] = "(v1, v2, v3) - returns the normal of the 3D triangle defined";
|
|
static char M_Mathutils_QuadNormal_doc[] = "(v1, v2, v3, v4) - returns the normal of the 3D quad defined";
|
|
static char M_Mathutils_LineIntersect_doc[] = "(v1, v2, v3, v4) - returns a tuple with the points on each line respectively closest to the other";
|
|
//-----------------------METHOD DEFINITIONS ----------------------
|
|
|
|
static PyObject *M_Mathutils_Rand(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_AngleBetweenVecs(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_MidpointVecs(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_ProjectVecs(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_RotationMatrix(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_TranslationMatrix(PyObject * self, VectorObject * value);
|
|
static PyObject *M_Mathutils_ScaleMatrix(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_OrthoProjectionMatrix(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_ShearMatrix(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_DifferenceQuats(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_Slerp(PyObject * self, PyObject * args);
|
|
static PyObject *M_Mathutils_Intersect( PyObject * self, PyObject * args );
|
|
static PyObject *M_Mathutils_TriangleArea( PyObject * self, PyObject * args );
|
|
static PyObject *M_Mathutils_TriangleNormal( PyObject * self, PyObject * args );
|
|
static PyObject *M_Mathutils_QuadNormal( PyObject * self, PyObject * args );
|
|
static PyObject *M_Mathutils_LineIntersect( PyObject * self, PyObject * args );
|
|
|
|
struct PyMethodDef M_Mathutils_methods[] = {
|
|
{"Rand", (PyCFunction) M_Mathutils_Rand, METH_VARARGS, M_Mathutils_Rand_doc},
|
|
{"AngleBetweenVecs", (PyCFunction) M_Mathutils_AngleBetweenVecs, METH_VARARGS, M_Mathutils_AngleBetweenVecs_doc},
|
|
{"MidpointVecs", (PyCFunction) M_Mathutils_MidpointVecs, METH_VARARGS, M_Mathutils_MidpointVecs_doc},
|
|
{"ProjectVecs", (PyCFunction) M_Mathutils_ProjectVecs, METH_VARARGS, M_Mathutils_ProjectVecs_doc},
|
|
{"RotationMatrix", (PyCFunction) M_Mathutils_RotationMatrix, METH_VARARGS, M_Mathutils_RotationMatrix_doc},
|
|
{"ScaleMatrix", (PyCFunction) M_Mathutils_ScaleMatrix, METH_VARARGS, M_Mathutils_ScaleMatrix_doc},
|
|
{"ShearMatrix", (PyCFunction) M_Mathutils_ShearMatrix, METH_VARARGS, M_Mathutils_ShearMatrix_doc},
|
|
{"TranslationMatrix", (PyCFunction) M_Mathutils_TranslationMatrix, METH_O, M_Mathutils_TranslationMatrix_doc},
|
|
{"OrthoProjectionMatrix", (PyCFunction) M_Mathutils_OrthoProjectionMatrix, METH_VARARGS, M_Mathutils_OrthoProjectionMatrix_doc},
|
|
{"DifferenceQuats", (PyCFunction) M_Mathutils_DifferenceQuats, METH_VARARGS,M_Mathutils_DifferenceQuats_doc},
|
|
{"Slerp", (PyCFunction) M_Mathutils_Slerp, METH_VARARGS, M_Mathutils_Slerp_doc},
|
|
{"Intersect", ( PyCFunction ) M_Mathutils_Intersect, METH_VARARGS, M_Mathutils_Intersect_doc},
|
|
{"TriangleArea", ( PyCFunction ) M_Mathutils_TriangleArea, METH_VARARGS, M_Mathutils_TriangleArea_doc},
|
|
{"TriangleNormal", ( PyCFunction ) M_Mathutils_TriangleNormal, METH_VARARGS, M_Mathutils_TriangleNormal_doc},
|
|
{"QuadNormal", ( PyCFunction ) M_Mathutils_QuadNormal, METH_VARARGS, M_Mathutils_QuadNormal_doc},
|
|
{"LineIntersect", ( PyCFunction ) M_Mathutils_LineIntersect, METH_VARARGS, M_Mathutils_LineIntersect_doc},
|
|
{NULL, NULL, 0, NULL}
|
|
};
|
|
|
|
/*----------------------------MODULE INIT-------------------------*/
|
|
/* from can be Blender.Mathutils or GameLogic.Mathutils for the BGE */
|
|
|
|
#if (PY_VERSION_HEX >= 0x03000000)
|
|
static struct PyModuleDef M_Mathutils_module_def = {
|
|
PyModuleDef_HEAD_INIT,
|
|
"Mathutils", /* m_name */
|
|
M_Mathutils_doc, /* m_doc */
|
|
0, /* m_size */
|
|
M_Mathutils_methods, /* m_methods */
|
|
0, /* m_reload */
|
|
0, /* m_traverse */
|
|
0, /* m_clear */
|
|
0, /* m_free */
|
|
};
|
|
#endif
|
|
|
|
PyObject *Mathutils_Init(const char *from)
|
|
{
|
|
PyObject *submodule;
|
|
|
|
//seed the generator for the rand function
|
|
BLI_srand((unsigned int) (PIL_check_seconds_timer() * 0x7FFFFFFF));
|
|
|
|
if( PyType_Ready( &vector_Type ) < 0 )
|
|
return NULL;
|
|
if( PyType_Ready( &matrix_Type ) < 0 )
|
|
return NULL;
|
|
if( PyType_Ready( &euler_Type ) < 0 )
|
|
return NULL;
|
|
if( PyType_Ready( &quaternion_Type ) < 0 )
|
|
return NULL;
|
|
|
|
#if (PY_VERSION_HEX >= 0x03000000)
|
|
submodule = PyModule_Create(&M_Mathutils_module_def);
|
|
PyDict_SetItemString(PySys_GetObject("modules"), M_Mathutils_module_def.m_name, submodule);
|
|
#else
|
|
submodule = Py_InitModule3(from, M_Mathutils_methods, M_Mathutils_doc);
|
|
#endif
|
|
|
|
/* each type has its own new() function */
|
|
PyModule_AddObject( submodule, "Vector", (PyObject *)&vector_Type );
|
|
PyModule_AddObject( submodule, "Matrix", (PyObject *)&matrix_Type );
|
|
PyModule_AddObject( submodule, "Euler", (PyObject *)&euler_Type );
|
|
PyModule_AddObject( submodule, "Quaternion", (PyObject *)&quaternion_Type );
|
|
|
|
mathutils_matrix_vector_cb_index= Mathutils_RegisterCallback(&mathutils_matrix_vector_cb);
|
|
|
|
return (submodule);
|
|
}
|
|
|
|
//-----------------------------METHODS----------------------------
|
|
//-----------------quat_rotation (internal)-----------
|
|
//This function multiplies a vector/point * quat or vice versa
|
|
//to rotate the point/vector by the quaternion
|
|
//arguments should all be 3D
|
|
PyObject *quat_rotation(PyObject *arg1, PyObject *arg2)
|
|
{
|
|
float rot[3];
|
|
QuaternionObject *quat = NULL;
|
|
VectorObject *vec = NULL;
|
|
|
|
if(QuaternionObject_Check(arg1)){
|
|
quat = (QuaternionObject*)arg1;
|
|
if(!BaseMath_ReadCallback(quat))
|
|
return NULL;
|
|
|
|
if(VectorObject_Check(arg2)){
|
|
vec = (VectorObject*)arg2;
|
|
|
|
if(!BaseMath_ReadCallback(vec))
|
|
return NULL;
|
|
|
|
rot[0] = quat->quat[0]*quat->quat[0]*vec->vec[0] + 2*quat->quat[2]*quat->quat[0]*vec->vec[2] -
|
|
2*quat->quat[3]*quat->quat[0]*vec->vec[1] + quat->quat[1]*quat->quat[1]*vec->vec[0] +
|
|
2*quat->quat[2]*quat->quat[1]*vec->vec[1] + 2*quat->quat[3]*quat->quat[1]*vec->vec[2] -
|
|
quat->quat[3]*quat->quat[3]*vec->vec[0] - quat->quat[2]*quat->quat[2]*vec->vec[0];
|
|
rot[1] = 2*quat->quat[1]*quat->quat[2]*vec->vec[0] + quat->quat[2]*quat->quat[2]*vec->vec[1] +
|
|
2*quat->quat[3]*quat->quat[2]*vec->vec[2] + 2*quat->quat[0]*quat->quat[3]*vec->vec[0] -
|
|
quat->quat[3]*quat->quat[3]*vec->vec[1] + quat->quat[0]*quat->quat[0]*vec->vec[1] -
|
|
2*quat->quat[1]*quat->quat[0]*vec->vec[2] - quat->quat[1]*quat->quat[1]*vec->vec[1];
|
|
rot[2] = 2*quat->quat[1]*quat->quat[3]*vec->vec[0] + 2*quat->quat[2]*quat->quat[3]*vec->vec[1] +
|
|
quat->quat[3]*quat->quat[3]*vec->vec[2] - 2*quat->quat[0]*quat->quat[2]*vec->vec[0] -
|
|
quat->quat[2]*quat->quat[2]*vec->vec[2] + 2*quat->quat[0]*quat->quat[1]*vec->vec[1] -
|
|
quat->quat[1]*quat->quat[1]*vec->vec[2] + quat->quat[0]*quat->quat[0]*vec->vec[2];
|
|
return newVectorObject(rot, 3, Py_NEW, NULL);
|
|
}
|
|
}else if(VectorObject_Check(arg1)){
|
|
vec = (VectorObject*)arg1;
|
|
|
|
if(!BaseMath_ReadCallback(vec))
|
|
return NULL;
|
|
|
|
if(QuaternionObject_Check(arg2)){
|
|
quat = (QuaternionObject*)arg2;
|
|
if(!BaseMath_ReadCallback(quat))
|
|
return NULL;
|
|
|
|
rot[0] = quat->quat[0]*quat->quat[0]*vec->vec[0] + 2*quat->quat[2]*quat->quat[0]*vec->vec[2] -
|
|
2*quat->quat[3]*quat->quat[0]*vec->vec[1] + quat->quat[1]*quat->quat[1]*vec->vec[0] +
|
|
2*quat->quat[2]*quat->quat[1]*vec->vec[1] + 2*quat->quat[3]*quat->quat[1]*vec->vec[2] -
|
|
quat->quat[3]*quat->quat[3]*vec->vec[0] - quat->quat[2]*quat->quat[2]*vec->vec[0];
|
|
rot[1] = 2*quat->quat[1]*quat->quat[2]*vec->vec[0] + quat->quat[2]*quat->quat[2]*vec->vec[1] +
|
|
2*quat->quat[3]*quat->quat[2]*vec->vec[2] + 2*quat->quat[0]*quat->quat[3]*vec->vec[0] -
|
|
quat->quat[3]*quat->quat[3]*vec->vec[1] + quat->quat[0]*quat->quat[0]*vec->vec[1] -
|
|
2*quat->quat[1]*quat->quat[0]*vec->vec[2] - quat->quat[1]*quat->quat[1]*vec->vec[1];
|
|
rot[2] = 2*quat->quat[1]*quat->quat[3]*vec->vec[0] + 2*quat->quat[2]*quat->quat[3]*vec->vec[1] +
|
|
quat->quat[3]*quat->quat[3]*vec->vec[2] - 2*quat->quat[0]*quat->quat[2]*vec->vec[0] -
|
|
quat->quat[2]*quat->quat[2]*vec->vec[2] + 2*quat->quat[0]*quat->quat[1]*vec->vec[1] -
|
|
quat->quat[1]*quat->quat[1]*vec->vec[2] + quat->quat[0]*quat->quat[0]*vec->vec[2];
|
|
return newVectorObject(rot, 3, Py_NEW, NULL);
|
|
}
|
|
}
|
|
|
|
PyErr_SetString(PyExc_RuntimeError, "quat_rotation(internal): internal problem rotating vector/point\n");
|
|
return NULL;
|
|
|
|
}
|
|
|
|
//----------------------------------Mathutils.Rand() --------------------
|
|
//returns a random number between a high and low value
|
|
static PyObject *M_Mathutils_Rand(PyObject * self, PyObject * args)
|
|
{
|
|
float high, low, range;
|
|
double drand;
|
|
//initializers
|
|
high = 1.0;
|
|
low = 0.0;
|
|
|
|
if(!PyArg_ParseTuple(args, "|ff", &low, &high)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.Rand(): expected nothing or optional (float, float)\n");
|
|
return NULL;
|
|
}
|
|
|
|
if((high < low) || (high < 0 && low > 0)) {
|
|
PyErr_SetString(PyExc_ValueError, "Mathutils.Rand(): high value should be larger than low value\n");
|
|
return NULL;
|
|
}
|
|
//get the random number 0 - 1
|
|
drand = BLI_drand();
|
|
|
|
//set it to range
|
|
range = high - low;
|
|
drand = drand * range;
|
|
drand = drand + low;
|
|
|
|
return PyFloat_FromDouble(drand);
|
|
}
|
|
//----------------------------------VECTOR FUNCTIONS---------------------
|
|
//----------------------------------Mathutils.AngleBetweenVecs() ---------
|
|
//calculates the angle between 2 vectors
|
|
static PyObject *M_Mathutils_AngleBetweenVecs(PyObject * self, PyObject * args)
|
|
{
|
|
VectorObject *vec1 = NULL, *vec2 = NULL;
|
|
double dot = 0.0f, angleRads, test_v1 = 0.0f, test_v2 = 0.0f;
|
|
int x, size;
|
|
|
|
if(!PyArg_ParseTuple(args, "O!O!", &vector_Type, &vec1, &vector_Type, &vec2))
|
|
goto AttributeError1; //not vectors
|
|
if(vec1->size != vec2->size)
|
|
goto AttributeError1; //bad sizes
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2))
|
|
return NULL;
|
|
|
|
//since size is the same....
|
|
size = vec1->size;
|
|
|
|
for(x = 0; x < size; x++) {
|
|
test_v1 += vec1->vec[x] * vec1->vec[x];
|
|
test_v2 += vec2->vec[x] * vec2->vec[x];
|
|
}
|
|
if (!test_v1 || !test_v2){
|
|
goto AttributeError2; //zero-length vector
|
|
}
|
|
|
|
//dot product
|
|
for(x = 0; x < size; x++) {
|
|
dot += vec1->vec[x] * vec2->vec[x];
|
|
}
|
|
dot /= (sqrt(test_v1) * sqrt(test_v2));
|
|
|
|
angleRads = (double)saacos(dot);
|
|
|
|
#ifdef USE_MATHUTILS_DEG
|
|
return PyFloat_FromDouble(angleRads * (180/ Py_PI));
|
|
#else
|
|
return PyFloat_FromDouble(angleRads);
|
|
#endif
|
|
AttributeError1:
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.AngleBetweenVecs(): expects (2) VECTOR objects of the same size\n");
|
|
return NULL;
|
|
|
|
AttributeError2:
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.AngleBetweenVecs(): zero length vectors are not acceptable arguments\n");
|
|
return NULL;
|
|
}
|
|
//----------------------------------Mathutils.MidpointVecs() -------------
|
|
//calculates the midpoint between 2 vectors
|
|
static PyObject *M_Mathutils_MidpointVecs(PyObject * self, PyObject * args)
|
|
{
|
|
VectorObject *vec1 = NULL, *vec2 = NULL;
|
|
float vec[4];
|
|
int x;
|
|
|
|
if(!PyArg_ParseTuple(args, "O!O!", &vector_Type, &vec1, &vector_Type, &vec2)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.MidpointVecs(): expects (2) vector objects of the same size\n");
|
|
return NULL;
|
|
}
|
|
if(vec1->size != vec2->size) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.MidpointVecs(): expects (2) vector objects of the same size\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2))
|
|
return NULL;
|
|
|
|
for(x = 0; x < vec1->size; x++) {
|
|
vec[x] = 0.5f * (vec1->vec[x] + vec2->vec[x]);
|
|
}
|
|
return newVectorObject(vec, vec1->size, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------Mathutils.ProjectVecs() -------------
|
|
//projects vector 1 onto vector 2
|
|
static PyObject *M_Mathutils_ProjectVecs(PyObject * self, PyObject * args)
|
|
{
|
|
VectorObject *vec1 = NULL, *vec2 = NULL;
|
|
float vec[4];
|
|
double dot = 0.0f, dot2 = 0.0f;
|
|
int x, size;
|
|
|
|
if(!PyArg_ParseTuple(args, "O!O!", &vector_Type, &vec1, &vector_Type, &vec2)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.ProjectVecs(): expects (2) vector objects of the same size\n");
|
|
return NULL;
|
|
}
|
|
if(vec1->size != vec2->size) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.ProjectVecs(): expects (2) vector objects of the same size\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2))
|
|
return NULL;
|
|
|
|
|
|
//since they are the same size...
|
|
size = vec1->size;
|
|
|
|
//get dot products
|
|
for(x = 0; x < size; x++) {
|
|
dot += vec1->vec[x] * vec2->vec[x];
|
|
dot2 += vec2->vec[x] * vec2->vec[x];
|
|
}
|
|
//projection
|
|
dot /= dot2;
|
|
for(x = 0; x < size; x++) {
|
|
vec[x] = (float)(dot * vec2->vec[x]);
|
|
}
|
|
return newVectorObject(vec, size, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------MATRIX FUNCTIONS--------------------
|
|
//----------------------------------Mathutils.RotationMatrix() ----------
|
|
//mat is a 1D array of floats - row[0][0],row[0][1], row[1][0], etc.
|
|
//creates a rotation matrix
|
|
static PyObject *M_Mathutils_RotationMatrix(PyObject * self, PyObject * args)
|
|
{
|
|
VectorObject *vec = NULL;
|
|
char *axis = NULL;
|
|
int matSize;
|
|
float angle = 0.0f, norm = 0.0f, cosAngle = 0.0f, sinAngle = 0.0f;
|
|
float mat[16] = {0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
|
|
0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f};
|
|
|
|
if(!PyArg_ParseTuple(args, "fi|sO!", &angle, &matSize, &axis, &vector_Type, &vec)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.RotationMatrix(): expected float int and optional string and vector\n");
|
|
return NULL;
|
|
}
|
|
|
|
#ifdef USE_MATHUTILS_DEG
|
|
/* Clamp to -360:360 */
|
|
while (angle<-360.0f)
|
|
angle+=360.0;
|
|
while (angle>360.0f)
|
|
angle-=360.0;
|
|
#else
|
|
while (angle<-(Py_PI*2))
|
|
angle+=(Py_PI*2);
|
|
while (angle>(Py_PI*2))
|
|
angle-=(Py_PI*2);
|
|
#endif
|
|
|
|
if(matSize != 2 && matSize != 3 && matSize != 4) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.RotationMatrix(): can only return a 2x2 3x3 or 4x4 matrix\n");
|
|
return NULL;
|
|
}
|
|
if(matSize == 2 && (axis != NULL || vec != NULL)) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.RotationMatrix(): cannot create a 2x2 rotation matrix around arbitrary axis\n");
|
|
return NULL;
|
|
}
|
|
if((matSize == 3 || matSize == 4) && axis == NULL) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.RotationMatrix(): please choose an axis of rotation for 3d and 4d matrices\n");
|
|
return NULL;
|
|
}
|
|
if(axis) {
|
|
if(((strcmp(axis, "r") == 0) || (strcmp(axis, "R") == 0)) && vec == NULL) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.RotationMatrix(): please define the arbitrary axis of rotation\n");
|
|
return NULL;
|
|
}
|
|
}
|
|
if(vec) {
|
|
if(vec->size != 3) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.RotationMatrix(): the arbitrary axis must be a 3D vector\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec))
|
|
return NULL;
|
|
|
|
}
|
|
#ifdef USE_MATHUTILS_DEG
|
|
//convert to radians
|
|
angle = angle * (float) (Py_PI / 180);
|
|
#endif
|
|
|
|
if(axis == NULL && matSize == 2) {
|
|
//2D rotation matrix
|
|
mat[0] = (float) cos (angle);
|
|
mat[1] = (float) sin (angle);
|
|
mat[2] = -((float) sin(angle));
|
|
mat[3] = (float) cos(angle);
|
|
} else if((strcmp(axis, "x") == 0) || (strcmp(axis, "X") == 0)) {
|
|
//rotation around X
|
|
mat[0] = 1.0f;
|
|
mat[4] = (float) cos(angle);
|
|
mat[5] = (float) sin(angle);
|
|
mat[7] = -((float) sin(angle));
|
|
mat[8] = (float) cos(angle);
|
|
} else if((strcmp(axis, "y") == 0) || (strcmp(axis, "Y") == 0)) {
|
|
//rotation around Y
|
|
mat[0] = (float) cos(angle);
|
|
mat[2] = -((float) sin(angle));
|
|
mat[4] = 1.0f;
|
|
mat[6] = (float) sin(angle);
|
|
mat[8] = (float) cos(angle);
|
|
} else if((strcmp(axis, "z") == 0) || (strcmp(axis, "Z") == 0)) {
|
|
//rotation around Z
|
|
mat[0] = (float) cos(angle);
|
|
mat[1] = (float) sin(angle);
|
|
mat[3] = -((float) sin(angle));
|
|
mat[4] = (float) cos(angle);
|
|
mat[8] = 1.0f;
|
|
} else if((strcmp(axis, "r") == 0) || (strcmp(axis, "R") == 0)) {
|
|
//arbitrary rotation
|
|
//normalize arbitrary axis
|
|
norm = (float) sqrt(vec->vec[0] * vec->vec[0] +
|
|
vec->vec[1] * vec->vec[1] +
|
|
vec->vec[2] * vec->vec[2]);
|
|
vec->vec[0] /= norm;
|
|
vec->vec[1] /= norm;
|
|
vec->vec[2] /= norm;
|
|
|
|
if (isnan(vec->vec[0]) || isnan(vec->vec[1]) || isnan(vec->vec[2])) {
|
|
/* zero length vector, return an identity matrix, could also return an error */
|
|
mat[0]= mat[4] = mat[8] = 1.0f;
|
|
} else {
|
|
/* create matrix */
|
|
cosAngle = (float) cos(angle);
|
|
sinAngle = (float) sin(angle);
|
|
mat[0] = ((vec->vec[0] * vec->vec[0]) * (1 - cosAngle)) +
|
|
cosAngle;
|
|
mat[1] = ((vec->vec[0] * vec->vec[1]) * (1 - cosAngle)) +
|
|
(vec->vec[2] * sinAngle);
|
|
mat[2] = ((vec->vec[0] * vec->vec[2]) * (1 - cosAngle)) -
|
|
(vec->vec[1] * sinAngle);
|
|
mat[3] = ((vec->vec[0] * vec->vec[1]) * (1 - cosAngle)) -
|
|
(vec->vec[2] * sinAngle);
|
|
mat[4] = ((vec->vec[1] * vec->vec[1]) * (1 - cosAngle)) +
|
|
cosAngle;
|
|
mat[5] = ((vec->vec[1] * vec->vec[2]) * (1 - cosAngle)) +
|
|
(vec->vec[0] * sinAngle);
|
|
mat[6] = ((vec->vec[0] * vec->vec[2]) * (1 - cosAngle)) +
|
|
(vec->vec[1] * sinAngle);
|
|
mat[7] = ((vec->vec[1] * vec->vec[2]) * (1 - cosAngle)) -
|
|
(vec->vec[0] * sinAngle);
|
|
mat[8] = ((vec->vec[2] * vec->vec[2]) * (1 - cosAngle)) +
|
|
cosAngle;
|
|
}
|
|
} else {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.RotationMatrix(): unrecognizable axis of rotation type - expected x,y,z or r\n");
|
|
return NULL;
|
|
}
|
|
if(matSize == 4) {
|
|
//resize matrix
|
|
mat[10] = mat[8];
|
|
mat[9] = mat[7];
|
|
mat[8] = mat[6];
|
|
mat[7] = 0.0f;
|
|
mat[6] = mat[5];
|
|
mat[5] = mat[4];
|
|
mat[4] = mat[3];
|
|
mat[3] = 0.0f;
|
|
}
|
|
//pass to matrix creation
|
|
return newMatrixObject(mat, matSize, matSize, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------Mathutils.TranslationMatrix() -------
|
|
//creates a translation matrix
|
|
static PyObject *M_Mathutils_TranslationMatrix(PyObject * self, VectorObject * vec)
|
|
{
|
|
float mat[16] = {0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
|
|
0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f};
|
|
|
|
if(!VectorObject_Check(vec)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.TranslationMatrix(): expected vector\n");
|
|
return NULL;
|
|
}
|
|
if(vec->size != 3 && vec->size != 4) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.TranslationMatrix(): vector must be 3D or 4D\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec))
|
|
return NULL;
|
|
|
|
//create a identity matrix and add translation
|
|
Mat4One((float(*)[4]) mat);
|
|
mat[12] = vec->vec[0];
|
|
mat[13] = vec->vec[1];
|
|
mat[14] = vec->vec[2];
|
|
|
|
return newMatrixObject(mat, 4, 4, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------Mathutils.ScaleMatrix() -------------
|
|
//mat is a 1D array of floats - row[0][0],row[0][1], row[1][0], etc.
|
|
//creates a scaling matrix
|
|
static PyObject *M_Mathutils_ScaleMatrix(PyObject * self, PyObject * args)
|
|
{
|
|
VectorObject *vec = NULL;
|
|
float norm = 0.0f, factor;
|
|
int matSize, x;
|
|
float mat[16] = {0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
|
|
0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f};
|
|
|
|
if(!PyArg_ParseTuple(args, "fi|O!", &factor, &matSize, &vector_Type, &vec)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.ScaleMatrix(): expected float int and optional vector\n");
|
|
return NULL;
|
|
}
|
|
if(matSize != 2 && matSize != 3 && matSize != 4) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.ScaleMatrix(): can only return a 2x2 3x3 or 4x4 matrix\n");
|
|
return NULL;
|
|
}
|
|
if(vec) {
|
|
if(vec->size > 2 && matSize == 2) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.ScaleMatrix(): please use 2D vectors when scaling in 2D\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec))
|
|
return NULL;
|
|
|
|
}
|
|
if(vec == NULL) { //scaling along axis
|
|
if(matSize == 2) {
|
|
mat[0] = factor;
|
|
mat[3] = factor;
|
|
} else {
|
|
mat[0] = factor;
|
|
mat[4] = factor;
|
|
mat[8] = factor;
|
|
}
|
|
} else { //scaling in arbitrary direction
|
|
//normalize arbitrary axis
|
|
for(x = 0; x < vec->size; x++) {
|
|
norm += vec->vec[x] * vec->vec[x];
|
|
}
|
|
norm = (float) sqrt(norm);
|
|
for(x = 0; x < vec->size; x++) {
|
|
vec->vec[x] /= norm;
|
|
}
|
|
if(matSize == 2) {
|
|
mat[0] = 1 +((factor - 1) *(vec->vec[0] * vec->vec[0]));
|
|
mat[1] =((factor - 1) *(vec->vec[0] * vec->vec[1]));
|
|
mat[2] =((factor - 1) *(vec->vec[0] * vec->vec[1]));
|
|
mat[3] = 1 + ((factor - 1) *(vec->vec[1] * vec->vec[1]));
|
|
} else {
|
|
mat[0] = 1 + ((factor - 1) *(vec->vec[0] * vec->vec[0]));
|
|
mat[1] =((factor - 1) *(vec->vec[0] * vec->vec[1]));
|
|
mat[2] =((factor - 1) *(vec->vec[0] * vec->vec[2]));
|
|
mat[3] =((factor - 1) *(vec->vec[0] * vec->vec[1]));
|
|
mat[4] = 1 + ((factor - 1) *(vec->vec[1] * vec->vec[1]));
|
|
mat[5] =((factor - 1) *(vec->vec[1] * vec->vec[2]));
|
|
mat[6] =((factor - 1) *(vec->vec[0] * vec->vec[2]));
|
|
mat[7] =((factor - 1) *(vec->vec[1] * vec->vec[2]));
|
|
mat[8] = 1 + ((factor - 1) *(vec->vec[2] * vec->vec[2]));
|
|
}
|
|
}
|
|
if(matSize == 4) {
|
|
//resize matrix
|
|
mat[10] = mat[8];
|
|
mat[9] = mat[7];
|
|
mat[8] = mat[6];
|
|
mat[7] = 0.0f;
|
|
mat[6] = mat[5];
|
|
mat[5] = mat[4];
|
|
mat[4] = mat[3];
|
|
mat[3] = 0.0f;
|
|
}
|
|
//pass to matrix creation
|
|
return newMatrixObject(mat, matSize, matSize, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------Mathutils.OrthoProjectionMatrix() ---
|
|
//mat is a 1D array of floats - row[0][0],row[0][1], row[1][0], etc.
|
|
//creates an ortho projection matrix
|
|
static PyObject *M_Mathutils_OrthoProjectionMatrix(PyObject * self, PyObject * args)
|
|
{
|
|
VectorObject *vec = NULL;
|
|
char *plane;
|
|
int matSize, x;
|
|
float norm = 0.0f;
|
|
float mat[16] = {0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
|
|
0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f};
|
|
|
|
if(!PyArg_ParseTuple(args, "si|O!", &plane, &matSize, &vector_Type, &vec)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.OrthoProjectionMatrix(): expected string and int and optional vector\n");
|
|
return NULL;
|
|
}
|
|
if(matSize != 2 && matSize != 3 && matSize != 4) {
|
|
PyErr_SetString(PyExc_AttributeError,"Mathutils.OrthoProjectionMatrix(): can only return a 2x2 3x3 or 4x4 matrix\n");
|
|
return NULL;
|
|
}
|
|
if(vec) {
|
|
if(vec->size > 2 && matSize == 2) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.OrthoProjectionMatrix(): please use 2D vectors when scaling in 2D\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec))
|
|
return NULL;
|
|
|
|
}
|
|
if(vec == NULL) { //ortho projection onto cardinal plane
|
|
if(((strcmp(plane, "x") == 0)
|
|
|| (strcmp(plane, "X") == 0)) && matSize == 2) {
|
|
mat[0] = 1.0f;
|
|
} else if(((strcmp(plane, "y") == 0)
|
|
|| (strcmp(plane, "Y") == 0))
|
|
&& matSize == 2) {
|
|
mat[3] = 1.0f;
|
|
} else if(((strcmp(plane, "xy") == 0)
|
|
|| (strcmp(plane, "XY") == 0))
|
|
&& matSize > 2) {
|
|
mat[0] = 1.0f;
|
|
mat[4] = 1.0f;
|
|
} else if(((strcmp(plane, "xz") == 0)
|
|
|| (strcmp(plane, "XZ") == 0))
|
|
&& matSize > 2) {
|
|
mat[0] = 1.0f;
|
|
mat[8] = 1.0f;
|
|
} else if(((strcmp(plane, "yz") == 0)
|
|
|| (strcmp(plane, "YZ") == 0))
|
|
&& matSize > 2) {
|
|
mat[4] = 1.0f;
|
|
mat[8] = 1.0f;
|
|
} else {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.OrthoProjectionMatrix(): unknown plane - expected: x, y, xy, xz, yz\n");
|
|
return NULL;
|
|
}
|
|
} else { //arbitrary plane
|
|
//normalize arbitrary axis
|
|
for(x = 0; x < vec->size; x++) {
|
|
norm += vec->vec[x] * vec->vec[x];
|
|
}
|
|
norm = (float) sqrt(norm);
|
|
for(x = 0; x < vec->size; x++) {
|
|
vec->vec[x] /= norm;
|
|
}
|
|
if(((strcmp(plane, "r") == 0)
|
|
|| (strcmp(plane, "R") == 0)) && matSize == 2) {
|
|
mat[0] = 1 - (vec->vec[0] * vec->vec[0]);
|
|
mat[1] = -(vec->vec[0] * vec->vec[1]);
|
|
mat[2] = -(vec->vec[0] * vec->vec[1]);
|
|
mat[3] = 1 - (vec->vec[1] * vec->vec[1]);
|
|
} else if(((strcmp(plane, "r") == 0)
|
|
|| (strcmp(plane, "R") == 0))
|
|
&& matSize > 2) {
|
|
mat[0] = 1 - (vec->vec[0] * vec->vec[0]);
|
|
mat[1] = -(vec->vec[0] * vec->vec[1]);
|
|
mat[2] = -(vec->vec[0] * vec->vec[2]);
|
|
mat[3] = -(vec->vec[0] * vec->vec[1]);
|
|
mat[4] = 1 - (vec->vec[1] * vec->vec[1]);
|
|
mat[5] = -(vec->vec[1] * vec->vec[2]);
|
|
mat[6] = -(vec->vec[0] * vec->vec[2]);
|
|
mat[7] = -(vec->vec[1] * vec->vec[2]);
|
|
mat[8] = 1 - (vec->vec[2] * vec->vec[2]);
|
|
} else {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.OrthoProjectionMatrix(): unknown plane - expected: 'r' expected for axis designation\n");
|
|
return NULL;
|
|
}
|
|
}
|
|
if(matSize == 4) {
|
|
//resize matrix
|
|
mat[10] = mat[8];
|
|
mat[9] = mat[7];
|
|
mat[8] = mat[6];
|
|
mat[7] = 0.0f;
|
|
mat[6] = mat[5];
|
|
mat[5] = mat[4];
|
|
mat[4] = mat[3];
|
|
mat[3] = 0.0f;
|
|
}
|
|
//pass to matrix creation
|
|
return newMatrixObject(mat, matSize, matSize, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------Mathutils.ShearMatrix() -------------
|
|
//creates a shear matrix
|
|
static PyObject *M_Mathutils_ShearMatrix(PyObject * self, PyObject * args)
|
|
{
|
|
int matSize;
|
|
char *plane;
|
|
float factor;
|
|
float mat[16] = {0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
|
|
0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f};
|
|
|
|
if(!PyArg_ParseTuple(args, "sfi", &plane, &factor, &matSize)) {
|
|
PyErr_SetString(PyExc_TypeError,"Mathutils.ShearMatrix(): expected string float and int\n");
|
|
return NULL;
|
|
}
|
|
if(matSize != 2 && matSize != 3 && matSize != 4) {
|
|
PyErr_SetString(PyExc_AttributeError,"Mathutils.ShearMatrix(): can only return a 2x2 3x3 or 4x4 matrix\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(((strcmp(plane, "x") == 0) || (strcmp(plane, "X") == 0))
|
|
&& matSize == 2) {
|
|
mat[0] = 1.0f;
|
|
mat[2] = factor;
|
|
mat[3] = 1.0f;
|
|
} else if(((strcmp(plane, "y") == 0)
|
|
|| (strcmp(plane, "Y") == 0)) && matSize == 2) {
|
|
mat[0] = 1.0f;
|
|
mat[1] = factor;
|
|
mat[3] = 1.0f;
|
|
} else if(((strcmp(plane, "xy") == 0)
|
|
|| (strcmp(plane, "XY") == 0)) && matSize > 2) {
|
|
mat[0] = 1.0f;
|
|
mat[4] = 1.0f;
|
|
mat[6] = factor;
|
|
mat[7] = factor;
|
|
} else if(((strcmp(plane, "xz") == 0)
|
|
|| (strcmp(plane, "XZ") == 0)) && matSize > 2) {
|
|
mat[0] = 1.0f;
|
|
mat[3] = factor;
|
|
mat[4] = 1.0f;
|
|
mat[5] = factor;
|
|
mat[8] = 1.0f;
|
|
} else if(((strcmp(plane, "yz") == 0)
|
|
|| (strcmp(plane, "YZ") == 0)) && matSize > 2) {
|
|
mat[0] = 1.0f;
|
|
mat[1] = factor;
|
|
mat[2] = factor;
|
|
mat[4] = 1.0f;
|
|
mat[8] = 1.0f;
|
|
} else {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.ShearMatrix(): expected: x, y, xy, xz, yz or wrong matrix size for shearing plane\n");
|
|
return NULL;
|
|
}
|
|
if(matSize == 4) {
|
|
//resize matrix
|
|
mat[10] = mat[8];
|
|
mat[9] = mat[7];
|
|
mat[8] = mat[6];
|
|
mat[7] = 0.0f;
|
|
mat[6] = mat[5];
|
|
mat[5] = mat[4];
|
|
mat[4] = mat[3];
|
|
mat[3] = 0.0f;
|
|
}
|
|
//pass to matrix creation
|
|
return newMatrixObject(mat, matSize, matSize, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------QUATERNION FUNCTIONS-----------------
|
|
|
|
//----------------------------------Mathutils.DifferenceQuats() ---------
|
|
//returns the difference between 2 quaternions
|
|
static PyObject *M_Mathutils_DifferenceQuats(PyObject * self, PyObject * args)
|
|
{
|
|
QuaternionObject *quatU = NULL, *quatV = NULL;
|
|
float quat[4], tempQuat[4];
|
|
double dot = 0.0f;
|
|
int x;
|
|
|
|
if(!PyArg_ParseTuple(args, "O!O!", &quaternion_Type, &quatU, &quaternion_Type, &quatV)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.DifferenceQuats(): expected Quaternion types");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(quatU) || !BaseMath_ReadCallback(quatV))
|
|
return NULL;
|
|
|
|
tempQuat[0] = quatU->quat[0];
|
|
tempQuat[1] = -quatU->quat[1];
|
|
tempQuat[2] = -quatU->quat[2];
|
|
tempQuat[3] = -quatU->quat[3];
|
|
|
|
dot = sqrt(tempQuat[0] * tempQuat[0] + tempQuat[1] * tempQuat[1] +
|
|
tempQuat[2] * tempQuat[2] + tempQuat[3] * tempQuat[3]);
|
|
|
|
for(x = 0; x < 4; x++) {
|
|
tempQuat[x] /= (float)(dot * dot);
|
|
}
|
|
QuatMul(quat, tempQuat, quatV->quat);
|
|
return newQuaternionObject(quat, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------Mathutils.Slerp() ------------------
|
|
//attemps to interpolate 2 quaternions and return the result
|
|
static PyObject *M_Mathutils_Slerp(PyObject * self, PyObject * args)
|
|
{
|
|
QuaternionObject *quatU = NULL, *quatV = NULL;
|
|
float quat[4], quat_u[4], quat_v[4], param;
|
|
double x, y, dot, sinT, angle, IsinT;
|
|
int z;
|
|
|
|
if(!PyArg_ParseTuple(args, "O!O!f", &quaternion_Type, &quatU, &quaternion_Type, &quatV, ¶m)) {
|
|
PyErr_SetString(PyExc_TypeError, "Mathutils.Slerp(): expected Quaternion types and float");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(quatU) || !BaseMath_ReadCallback(quatV))
|
|
return NULL;
|
|
|
|
if(param > 1.0f || param < 0.0f) {
|
|
PyErr_SetString(PyExc_AttributeError, "Mathutils.Slerp(): interpolation factor must be between 0.0 and 1.0");
|
|
return NULL;
|
|
}
|
|
|
|
//copy quats
|
|
for(z = 0; z < 4; z++){
|
|
quat_u[z] = quatU->quat[z];
|
|
quat_v[z] = quatV->quat[z];
|
|
}
|
|
|
|
//dot product
|
|
dot = quat_u[0] * quat_v[0] + quat_u[1] * quat_v[1] +
|
|
quat_u[2] * quat_v[2] + quat_u[3] * quat_v[3];
|
|
|
|
//if negative negate a quat (shortest arc)
|
|
if(dot < 0.0f) {
|
|
quat_v[0] = -quat_v[0];
|
|
quat_v[1] = -quat_v[1];
|
|
quat_v[2] = -quat_v[2];
|
|
quat_v[3] = -quat_v[3];
|
|
dot = -dot;
|
|
}
|
|
if(dot > .99999f) { //very close
|
|
x = 1.0f - param;
|
|
y = param;
|
|
} else {
|
|
//calculate sin of angle
|
|
sinT = sqrt(1.0f - (dot * dot));
|
|
//calculate angle
|
|
angle = atan2(sinT, dot);
|
|
//caluculate inverse of sin(theta)
|
|
IsinT = 1.0f / sinT;
|
|
x = sin((1.0f - param) * angle) * IsinT;
|
|
y = sin(param * angle) * IsinT;
|
|
}
|
|
//interpolate
|
|
quat[0] = (float)(quat_u[0] * x + quat_v[0] * y);
|
|
quat[1] = (float)(quat_u[1] * x + quat_v[1] * y);
|
|
quat[2] = (float)(quat_u[2] * x + quat_v[2] * y);
|
|
quat[3] = (float)(quat_u[3] * x + quat_v[3] * y);
|
|
|
|
return newQuaternionObject(quat, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------EULER FUNCTIONS----------------------
|
|
//---------------------------------INTERSECTION FUNCTIONS--------------------
|
|
//----------------------------------Mathutils.Intersect() -------------------
|
|
static PyObject *M_Mathutils_Intersect( PyObject * self, PyObject * args )
|
|
{
|
|
VectorObject *ray, *ray_off, *vec1, *vec2, *vec3;
|
|
float dir[3], orig[3], v1[3], v2[3], v3[3], e1[3], e2[3], pvec[3], tvec[3], qvec[3];
|
|
float det, inv_det, u, v, t;
|
|
int clip = 1;
|
|
|
|
if(!PyArg_ParseTuple(args, "O!O!O!O!O!|i", &vector_Type, &vec1, &vector_Type, &vec2, &vector_Type, &vec3, &vector_Type, &ray, &vector_Type, &ray_off , &clip)) {
|
|
PyErr_SetString( PyExc_TypeError, "expected 5 vector types\n" );
|
|
return NULL;
|
|
}
|
|
if(vec1->size != 3 || vec2->size != 3 || vec3->size != 3 || ray->size != 3 || ray_off->size != 3) {
|
|
PyErr_SetString( PyExc_TypeError, "only 3D vectors for all parameters\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2) || !BaseMath_ReadCallback(vec3) || !BaseMath_ReadCallback(ray) || !BaseMath_ReadCallback(ray_off))
|
|
return NULL;
|
|
|
|
VECCOPY(v1, vec1->vec);
|
|
VECCOPY(v2, vec2->vec);
|
|
VECCOPY(v3, vec3->vec);
|
|
|
|
VECCOPY(dir, ray->vec);
|
|
Normalize(dir);
|
|
|
|
VECCOPY(orig, ray_off->vec);
|
|
|
|
/* find vectors for two edges sharing v1 */
|
|
VecSubf(e1, v2, v1);
|
|
VecSubf(e2, v3, v1);
|
|
|
|
/* begin calculating determinant - also used to calculated U parameter */
|
|
Crossf(pvec, dir, e2);
|
|
|
|
/* if determinant is near zero, ray lies in plane of triangle */
|
|
det = Inpf(e1, pvec);
|
|
|
|
if (det > -0.000001 && det < 0.000001) {
|
|
Py_RETURN_NONE;
|
|
}
|
|
|
|
inv_det = 1.0f / det;
|
|
|
|
/* calculate distance from v1 to ray origin */
|
|
VecSubf(tvec, orig, v1);
|
|
|
|
/* calculate U parameter and test bounds */
|
|
u = Inpf(tvec, pvec) * inv_det;
|
|
if (clip && (u < 0.0f || u > 1.0f)) {
|
|
Py_RETURN_NONE;
|
|
}
|
|
|
|
/* prepare to test the V parameter */
|
|
Crossf(qvec, tvec, e1);
|
|
|
|
/* calculate V parameter and test bounds */
|
|
v = Inpf(dir, qvec) * inv_det;
|
|
|
|
if (clip && (v < 0.0f || u + v > 1.0f)) {
|
|
Py_RETURN_NONE;
|
|
}
|
|
|
|
/* calculate t, ray intersects triangle */
|
|
t = Inpf(e2, qvec) * inv_det;
|
|
|
|
VecMulf(dir, t);
|
|
VecAddf(pvec, orig, dir);
|
|
|
|
return newVectorObject(pvec, 3, Py_NEW, NULL);
|
|
}
|
|
//----------------------------------Mathutils.LineIntersect() -------------------
|
|
/* Line-Line intersection using algorithm from mathworld.wolfram.com */
|
|
static PyObject *M_Mathutils_LineIntersect( PyObject * self, PyObject * args )
|
|
{
|
|
PyObject * tuple;
|
|
VectorObject *vec1, *vec2, *vec3, *vec4;
|
|
float v1[3], v2[3], v3[3], v4[3], i1[3], i2[3];
|
|
|
|
if( !PyArg_ParseTuple( args, "O!O!O!O!", &vector_Type, &vec1, &vector_Type, &vec2, &vector_Type, &vec3, &vector_Type, &vec4 ) ) {
|
|
PyErr_SetString( PyExc_TypeError, "expected 4 vector types\n" );
|
|
return NULL;
|
|
}
|
|
if( vec1->size != vec2->size || vec1->size != vec3->size || vec1->size != vec2->size) {
|
|
PyErr_SetString( PyExc_TypeError,"vectors must be of the same size\n" );
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2) || !BaseMath_ReadCallback(vec3) || !BaseMath_ReadCallback(vec4))
|
|
return NULL;
|
|
|
|
if( vec1->size == 3 || vec1->size == 2) {
|
|
int result;
|
|
|
|
if (vec1->size == 3) {
|
|
VECCOPY(v1, vec1->vec);
|
|
VECCOPY(v2, vec2->vec);
|
|
VECCOPY(v3, vec3->vec);
|
|
VECCOPY(v4, vec4->vec);
|
|
}
|
|
else {
|
|
v1[0] = vec1->vec[0];
|
|
v1[1] = vec1->vec[1];
|
|
v1[2] = 0.0f;
|
|
|
|
v2[0] = vec2->vec[0];
|
|
v2[1] = vec2->vec[1];
|
|
v2[2] = 0.0f;
|
|
|
|
v3[0] = vec3->vec[0];
|
|
v3[1] = vec3->vec[1];
|
|
v3[2] = 0.0f;
|
|
|
|
v4[0] = vec4->vec[0];
|
|
v4[1] = vec4->vec[1];
|
|
v4[2] = 0.0f;
|
|
}
|
|
|
|
result = LineIntersectLine(v1, v2, v3, v4, i1, i2);
|
|
|
|
if (result == 0) {
|
|
/* colinear */
|
|
Py_RETURN_NONE;
|
|
}
|
|
else {
|
|
tuple = PyTuple_New( 2 );
|
|
PyTuple_SetItem( tuple, 0, newVectorObject(i1, vec1->size, Py_NEW, NULL) );
|
|
PyTuple_SetItem( tuple, 1, newVectorObject(i2, vec1->size, Py_NEW, NULL) );
|
|
return tuple;
|
|
}
|
|
}
|
|
else {
|
|
PyErr_SetString( PyExc_TypeError, "2D/3D vectors only\n" );
|
|
return NULL;
|
|
}
|
|
}
|
|
|
|
|
|
|
|
//---------------------------------NORMALS FUNCTIONS--------------------
|
|
//----------------------------------Mathutils.QuadNormal() -------------------
|
|
static PyObject *M_Mathutils_QuadNormal( PyObject * self, PyObject * args )
|
|
{
|
|
VectorObject *vec1;
|
|
VectorObject *vec2;
|
|
VectorObject *vec3;
|
|
VectorObject *vec4;
|
|
float v1[3], v2[3], v3[3], v4[3], e1[3], e2[3], n1[3], n2[3];
|
|
|
|
if( !PyArg_ParseTuple( args, "O!O!O!O!", &vector_Type, &vec1, &vector_Type, &vec2, &vector_Type, &vec3, &vector_Type, &vec4 ) ) {
|
|
PyErr_SetString( PyExc_TypeError, "expected 4 vector types\n" );
|
|
return NULL;
|
|
}
|
|
if( vec1->size != vec2->size || vec1->size != vec3->size || vec1->size != vec4->size) {
|
|
PyErr_SetString( PyExc_TypeError,"vectors must be of the same size\n" );
|
|
return NULL;
|
|
}
|
|
if( vec1->size != 3 ) {
|
|
PyErr_SetString( PyExc_TypeError, "only 3D vectors\n" );
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2) || !BaseMath_ReadCallback(vec3) || !BaseMath_ReadCallback(vec4))
|
|
return NULL;
|
|
|
|
VECCOPY(v1, vec1->vec);
|
|
VECCOPY(v2, vec2->vec);
|
|
VECCOPY(v3, vec3->vec);
|
|
VECCOPY(v4, vec4->vec);
|
|
|
|
/* find vectors for two edges sharing v2 */
|
|
VecSubf(e1, v1, v2);
|
|
VecSubf(e2, v3, v2);
|
|
|
|
Crossf(n1, e2, e1);
|
|
Normalize(n1);
|
|
|
|
/* find vectors for two edges sharing v4 */
|
|
VecSubf(e1, v3, v4);
|
|
VecSubf(e2, v1, v4);
|
|
|
|
Crossf(n2, e2, e1);
|
|
Normalize(n2);
|
|
|
|
/* adding and averaging the normals of both triangles */
|
|
VecAddf(n1, n2, n1);
|
|
Normalize(n1);
|
|
|
|
return newVectorObject(n1, 3, Py_NEW, NULL);
|
|
}
|
|
|
|
//----------------------------Mathutils.TriangleNormal() -------------------
|
|
static PyObject *M_Mathutils_TriangleNormal( PyObject * self, PyObject * args )
|
|
{
|
|
VectorObject *vec1, *vec2, *vec3;
|
|
float v1[3], v2[3], v3[3], e1[3], e2[3], n[3];
|
|
|
|
if( !PyArg_ParseTuple( args, "O!O!O!", &vector_Type, &vec1, &vector_Type, &vec2, &vector_Type, &vec3 ) ) {
|
|
PyErr_SetString( PyExc_TypeError, "expected 3 vector types\n" );
|
|
return NULL;
|
|
}
|
|
if( vec1->size != vec2->size || vec1->size != vec3->size ) {
|
|
PyErr_SetString( PyExc_TypeError, "vectors must be of the same size\n" );
|
|
return NULL;
|
|
}
|
|
if( vec1->size != 3 ) {
|
|
PyErr_SetString( PyExc_TypeError, "only 3D vectors\n" );
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2) || !BaseMath_ReadCallback(vec3))
|
|
return NULL;
|
|
|
|
VECCOPY(v1, vec1->vec);
|
|
VECCOPY(v2, vec2->vec);
|
|
VECCOPY(v3, vec3->vec);
|
|
|
|
/* find vectors for two edges sharing v2 */
|
|
VecSubf(e1, v1, v2);
|
|
VecSubf(e2, v3, v2);
|
|
|
|
Crossf(n, e2, e1);
|
|
Normalize(n);
|
|
|
|
return newVectorObject(n, 3, Py_NEW, NULL);
|
|
}
|
|
|
|
//--------------------------------- AREA FUNCTIONS--------------------
|
|
//----------------------------------Mathutils.TriangleArea() -------------------
|
|
static PyObject *M_Mathutils_TriangleArea( PyObject * self, PyObject * args )
|
|
{
|
|
VectorObject *vec1, *vec2, *vec3;
|
|
float v1[3], v2[3], v3[3];
|
|
|
|
if( !PyArg_ParseTuple
|
|
( args, "O!O!O!", &vector_Type, &vec1, &vector_Type, &vec2
|
|
, &vector_Type, &vec3 ) ) {
|
|
PyErr_SetString( PyExc_TypeError, "expected 3 vector types\n");
|
|
return NULL;
|
|
}
|
|
if( vec1->size != vec2->size || vec1->size != vec3->size ) {
|
|
PyErr_SetString( PyExc_TypeError, "vectors must be of the same size\n" );
|
|
return NULL;
|
|
}
|
|
|
|
if(!BaseMath_ReadCallback(vec1) || !BaseMath_ReadCallback(vec2) || !BaseMath_ReadCallback(vec3))
|
|
return NULL;
|
|
|
|
if (vec1->size == 3) {
|
|
VECCOPY(v1, vec1->vec);
|
|
VECCOPY(v2, vec2->vec);
|
|
VECCOPY(v3, vec3->vec);
|
|
|
|
return PyFloat_FromDouble( AreaT3Dfl(v1, v2, v3) );
|
|
}
|
|
else if (vec1->size == 2) {
|
|
v1[0] = vec1->vec[0];
|
|
v1[1] = vec1->vec[1];
|
|
|
|
v2[0] = vec2->vec[0];
|
|
v2[1] = vec2->vec[1];
|
|
|
|
v3[0] = vec3->vec[0];
|
|
v3[1] = vec3->vec[1];
|
|
|
|
return PyFloat_FromDouble( AreaF2Dfl(v1, v2, v3) );
|
|
}
|
|
else {
|
|
PyErr_SetString( PyExc_TypeError, "only 2D,3D vectors are supported\n" );
|
|
return NULL;
|
|
}
|
|
}
|
|
|
|
/* Utility functions */
|
|
|
|
/*---------------------- EXPP_FloatsAreEqual -------------------------
|
|
Floating point comparisons
|
|
floatStep = number of representable floats allowable in between
|
|
float A and float B to be considered equal. */
|
|
int EXPP_FloatsAreEqual(float A, float B, int floatSteps)
|
|
{
|
|
int a, b, delta;
|
|
assert(floatSteps > 0 && floatSteps < (4 * 1024 * 1024));
|
|
a = *(int*)&A;
|
|
if (a < 0)
|
|
a = 0x80000000 - a;
|
|
b = *(int*)&B;
|
|
if (b < 0)
|
|
b = 0x80000000 - b;
|
|
delta = abs(a - b);
|
|
if (delta <= floatSteps)
|
|
return 1;
|
|
return 0;
|
|
}
|
|
/*---------------------- EXPP_VectorsAreEqual -------------------------
|
|
Builds on EXPP_FloatsAreEqual to test vectors */
|
|
int EXPP_VectorsAreEqual(float *vecA, float *vecB, int size, int floatSteps)
|
|
{
|
|
int x;
|
|
for (x=0; x< size; x++){
|
|
if (EXPP_FloatsAreEqual(vecA[x], vecB[x], floatSteps) == 0)
|
|
return 0;
|
|
}
|
|
return 1;
|
|
}
|
|
|
|
|
|
/* Mathutils Callbacks */
|
|
|
|
/* for mathutils internal use only, eventually should re-alloc but to start with we only have a few users */
|
|
Mathutils_Callback *mathutils_callbacks[8] = {NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL};
|
|
|
|
int Mathutils_RegisterCallback(Mathutils_Callback *cb)
|
|
{
|
|
int i;
|
|
|
|
/* find the first free slot */
|
|
for(i= 0; mathutils_callbacks[i]; i++) {
|
|
if(mathutils_callbacks[i]==cb) /* alredy registered? */
|
|
return i;
|
|
}
|
|
|
|
mathutils_callbacks[i] = cb;
|
|
return i;
|
|
}
|
|
|
|
/* use macros to check for NULL */
|
|
int _BaseMathObject_ReadCallback(BaseMathObject *self)
|
|
{
|
|
Mathutils_Callback *cb= mathutils_callbacks[self->cb_type];
|
|
if(cb->get(self->cb_user, self->cb_subtype, self->data))
|
|
return 1;
|
|
|
|
PyErr_Format(PyExc_SystemError, "%s user has become invalid", Py_TYPE(self)->tp_name);
|
|
return 0;
|
|
}
|
|
|
|
int _BaseMathObject_WriteCallback(BaseMathObject *self)
|
|
{
|
|
Mathutils_Callback *cb= mathutils_callbacks[self->cb_type];
|
|
if(cb->set(self->cb_user, self->cb_subtype, self->data))
|
|
return 1;
|
|
|
|
PyErr_Format(PyExc_SystemError, "%s user has become invalid", Py_TYPE(self)->tp_name);
|
|
return 0;
|
|
}
|
|
|
|
int _BaseMathObject_ReadIndexCallback(BaseMathObject *self, int index)
|
|
{
|
|
Mathutils_Callback *cb= mathutils_callbacks[self->cb_type];
|
|
if(cb->get_index(self->cb_user, self->cb_subtype, self->data, index))
|
|
return 1;
|
|
|
|
PyErr_Format(PyExc_SystemError, "%s user has become invalid", Py_TYPE(self)->tp_name);
|
|
return 0;
|
|
}
|
|
|
|
int _BaseMathObject_WriteIndexCallback(BaseMathObject *self, int index)
|
|
{
|
|
Mathutils_Callback *cb= mathutils_callbacks[self->cb_type];
|
|
if(cb->set_index(self->cb_user, self->cb_subtype, self->data, index))
|
|
return 1;
|
|
|
|
PyErr_Format(PyExc_SystemError, "%s user has become invalid", Py_TYPE(self)->tp_name);
|
|
return 0;
|
|
}
|
|
|
|
/* BaseMathObject generic functions for all mathutils types */
|
|
PyObject *BaseMathObject_getOwner( BaseMathObject * self, void *type )
|
|
{
|
|
PyObject *ret= self->cb_user ? self->cb_user : Py_None;
|
|
Py_INCREF(ret);
|
|
return ret;
|
|
}
|
|
|
|
PyObject *BaseMathObject_getWrapped( BaseMathObject *self, void *type )
|
|
{
|
|
return PyBool_FromLong((self->wrapped == Py_WRAP) ? 1:0);
|
|
}
|
|
|
|
void BaseMathObject_dealloc(BaseMathObject * self)
|
|
{
|
|
/* only free non wrapped */
|
|
if(self->wrapped != Py_WRAP)
|
|
PyMem_Free(self->data);
|
|
|
|
Py_XDECREF(self->cb_user);
|
|
Py_TYPE(self)->tp_free(self); // PyObject_DEL(self); // breaks subtypes
|
|
}
|
|
|