This repository has been archived on 2023-10-09. You can view files and clone it. You cannot open issues or pull requests or push a commit.
Files
blender-archive/source/blender/editors/transform/transform_constraints.c
jon denning 011327224e Transform Snap: nearest face snap mode, snapping options, refactoring.
This commit adds a new face nearest snapping mode, adds new snapping
options, and (lightly) refactors code around snapping.

The new face nearest snapping mode will snap transformed geometry to the
nearest surface in world space. In contrast, the original face snapping
mode uses projection (raycasting) to snap source to target geometry.
Face snapping therefore only works with what is visible, while nearest
face snapping can snap geometry to occluded parts of the scene. This new
mode is critical for retopology work, where some of the target mesh
might be occluded (ex: sliding an edge loop that wraps around the
backside of target mesh).

The nearest face snapping mode has two options: "Snap to Same Target"
and "Face Nearest Steps". When the Snap to Same Object option is
enabled, the selected source geometry will stay near the target that it
is nearest before editing started, which prevents the source geometry
from snapping to other targets. The Face Nearest Steps divides the
overall transformation for each vertex into n smaller transformations,
then applies those n transformations with surface snapping interlacing
each step. This steps option handles transformations that cross U-shaped
targets better.

The new snapping options allow the artist to better control which target
objects (objects to which the edited geometry is snapped) are considered
when snapping. In particular, the only option for filtering target
objects was a "Project onto Self", which allowed the currently edited
mesh to be considered as a target. Now, the artist can choose any
combination of the following to be considered as a target: the active
object, any edited object that isn't active (see note below), any non-
edited object. Additionally, the artist has another snapping option to
exclude objects that are not selectable as potential targets.

The Snapping Options dropdown has been lightly reorganized to allow for
the additional options.

Included in this patch:

- Snap target selection is more controllable for artist with additional
  snapping options.
- Renamed a few of the snap-related functions to better reflect what
  they actually do now. For example, `applySnapping` implies that this
  handles the snapping, while `applyProject` implies something entirely
  different is done there. However, better names would be
  `applySnappingAsGroup` and `applySnappingIndividual`, respectively,
  where `applySnappingIndividual` previously only does Face snapping.
- Added an initial coordinate parameter to snapping functions so that
  the nearest target before transforming can be determined(for "Snap to
  Same Object"), and so the transformation can be broken into smaller
  steps (for "Face Nearest Steps").
- Separated the BVH Tree getter code from mesh/edit mesh to its own
  function to reduce code duplication.
- Added icon for nearest face snapping.
- The original "Project onto Self" was actually not correct! This option
  should be called "Project onto Active" instead, but that only matters
  when editing multiple meshes at the same time. This patch makes this
  change in the UI.

Reviewed By: Campbell Barton, Germano Cavalcante

Differential Revision: https://developer.blender.org/D14591
2022-06-29 20:52:00 -04:00

1185 lines
33 KiB
C

/* SPDX-License-Identifier: GPL-2.0-or-later
* Copyright 2001-2002 NaN Holding BV. All rights reserved. */
/** \file
* \ingroup edtransform
*/
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "DNA_object_types.h"
#include "DNA_scene_types.h"
#include "DNA_screen_types.h"
#include "DNA_space_types.h"
#include "DNA_view3d_types.h"
#include "GPU_immediate.h"
#include "GPU_matrix.h"
#include "GPU_state.h"
#include "BLI_math.h"
#include "BLI_rect.h"
#include "BLI_string.h"
#include "BLI_utildefines.h"
#include "BKE_context.h"
#include "ED_view3d.h"
#include "BLT_translation.h"
#include "UI_resources.h"
#include "transform.h"
#include "transform_orientations.h"
#include "transform_snap.h"
/* Own include. */
#include "transform_constraints.h"
static void drawObjectConstraint(TransInfo *t);
/* -------------------------------------------------------------------- */
/** \name Internal Utilities
* \{ */
static void projection_matrix_calc(const TransInfo *t, float r_pmtx[3][3])
{
unit_m3(r_pmtx);
if (!(t->con.mode & CON_AXIS0)) {
zero_v3(r_pmtx[0]);
}
if (!(t->con.mode & CON_AXIS1)) {
zero_v3(r_pmtx[1]);
}
if (!(t->con.mode & CON_AXIS2)) {
zero_v3(r_pmtx[2]);
}
float mat[3][3];
mul_m3_m3m3(mat, r_pmtx, t->spacemtx_inv);
mul_m3_m3m3(r_pmtx, t->spacemtx, mat);
}
static void view_vector_calc(const TransInfo *t, const float focus[3], float r_vec[3])
{
if (t->persp != RV3D_ORTHO) {
sub_v3_v3v3(r_vec, t->viewinv[3], focus);
}
else {
copy_v3_v3(r_vec, t->viewinv[2]);
}
normalize_v3(r_vec);
}
/* ************************** CONSTRAINTS ************************* */
#define CONSTRAIN_EPSILON 0.0001f
static void constraint_plane_calc(const TransInfo *t, float r_plane[4])
{
const float *constraint_vector[2];
int n = 0;
for (int i = 0; i < 3; i++) {
if (t->con.mode & (CON_AXIS0 << i)) {
constraint_vector[n++] = t->spacemtx[i];
if (n == 2) {
break;
}
}
}
BLI_assert(n == 2);
cross_v3_v3v3(r_plane, constraint_vector[0], constraint_vector[1]);
normalize_v3(r_plane);
r_plane[3] = -dot_v3v3(r_plane, t->center_global);
}
void constraintNumInput(TransInfo *t, float vec[3])
{
int mode = t->con.mode;
if (mode & CON_APPLY) {
float nval = (t->flag & T_NULL_ONE) ? 1.0f : 0.0f;
const int dims = getConstraintSpaceDimension(t);
if (dims == 2) {
int axis = mode & (CON_AXIS0 | CON_AXIS1 | CON_AXIS2);
if (axis == (CON_AXIS0 | CON_AXIS1)) {
/* vec[0] = vec[0]; */ /* same */
/* vec[1] = vec[1]; */ /* same */
vec[2] = nval;
}
else if (axis == (CON_AXIS1 | CON_AXIS2)) {
vec[2] = vec[1];
vec[1] = vec[0];
vec[0] = nval;
}
else if (axis == (CON_AXIS0 | CON_AXIS2)) {
/* vec[0] = vec[0]; */ /* same */
vec[2] = vec[1];
vec[1] = nval;
}
}
else if (dims == 1) {
if (mode & CON_AXIS0) {
/* vec[0] = vec[0]; */ /* same */
vec[1] = nval;
vec[2] = nval;
}
else if (mode & CON_AXIS1) {
vec[1] = vec[0];
vec[0] = nval;
vec[2] = nval;
}
else if (mode & CON_AXIS2) {
vec[2] = vec[0];
vec[0] = nval;
vec[1] = nval;
}
}
}
}
static void viewAxisCorrectCenter(const TransInfo *t, float t_con_center[3])
{
if (t->spacetype == SPACE_VIEW3D) {
// View3D *v3d = t->area->spacedata.first;
const float min_dist = 1.0f; /* v3d->clip_start; */
float dir[3];
float l;
sub_v3_v3v3(dir, t_con_center, t->viewinv[3]);
if (dot_v3v3(dir, t->viewinv[2]) < 0.0f) {
negate_v3(dir);
}
project_v3_v3v3(dir, dir, t->viewinv[2]);
l = len_v3(dir);
if (l < min_dist) {
float diff[3];
normalize_v3_v3_length(diff, t->viewinv[2], min_dist - l);
sub_v3_v3(t_con_center, diff);
}
}
}
/**
* Axis calculation taking the view into account, correcting view-aligned axis.
*/
static void axisProjection(const TransInfo *t,
const float axis[3],
const float in[3],
float out[3])
{
float norm[3], vec[3], factor, angle;
float t_con_center[3];
if (is_zero_v3(in)) {
return;
}
copy_v3_v3(t_con_center, t->center_global);
/* checks for center being too close to the view center */
viewAxisCorrectCenter(t, t_con_center);
angle = fabsf(angle_v3v3(axis, t->viewinv[2]));
if (angle > (float)M_PI_2) {
angle = (float)M_PI - angle;
}
/* For when view is parallel to constraint... will cause NaNs otherwise
* So we take vertical motion in 3D space and apply it to the
* constraint axis. Nice for camera grab + MMB */
if (angle < DEG2RADF(5.0f)) {
project_v3_v3v3(vec, in, t->viewinv[1]);
factor = dot_v3v3(t->viewinv[1], vec) * 2.0f;
/* Since camera distance is quite relative, use quadratic relationship.
* holding shift can compensate. */
if (factor < 0.0f) {
factor *= -factor;
}
else {
factor *= factor;
}
/* -factor makes move down going backwards */
normalize_v3_v3_length(out, axis, -factor);
}
else {
float v[3];
float norm_center[3];
float plane[3];
view_vector_calc(t, t_con_center, norm_center);
cross_v3_v3v3(plane, norm_center, axis);
project_v3_v3v3(vec, in, plane);
sub_v3_v3v3(vec, in, vec);
add_v3_v3v3(v, vec, t_con_center);
view_vector_calc(t, v, norm);
/* give arbitrary large value if projection is impossible */
factor = dot_v3v3(axis, norm);
if (1.0f - fabsf(factor) < 0.0002f) {
copy_v3_v3(out, axis);
if (factor > 0) {
mul_v3_fl(out, 1000000000.0f);
}
else {
mul_v3_fl(out, -1000000000.0f);
}
}
else {
/* Use ray-ray intersection instead of line-line because this gave
* precision issues adding small values to large numbers. */
float mul;
if (isect_ray_ray_v3(t_con_center, axis, v, norm, &mul, NULL)) {
mul_v3_v3fl(out, axis, mul);
}
else {
/* In practice this should never fail. */
BLI_assert(0);
}
/* possible some values become nan when
* viewpoint and object are both zero */
if (!isfinite(out[0])) {
out[0] = 0.0f;
}
if (!isfinite(out[1])) {
out[1] = 0.0f;
}
if (!isfinite(out[2])) {
out[2] = 0.0f;
}
}
}
}
/**
* Snap to the intersection between the edge direction and the constraint plane.
*/
static void constraint_snap_plane_to_edge(const TransInfo *t, const float plane[4], float r_out[3])
{
float lambda;
const float *edge_snap_point = t->tsnap.snapPoint;
const float *edge_dir = t->tsnap.snapNormal;
bool is_aligned = fabsf(dot_v3v3(edge_dir, plane)) < CONSTRAIN_EPSILON;
if (!is_aligned && isect_ray_plane_v3(edge_snap_point, edge_dir, plane, &lambda, false)) {
madd_v3_v3v3fl(r_out, edge_snap_point, edge_dir, lambda);
sub_v3_v3(r_out, t->tsnap.snapTarget);
}
}
static void UNUSED_FUNCTION(constraint_snap_plane_to_face(const TransInfo *t,
const float plane[4],
float r_out[3]))
{
float face_plane[4], isect_orig[3], isect_dir[3];
const float *face_snap_point = t->tsnap.snapPoint;
const float *face_normal = t->tsnap.snapNormal;
plane_from_point_normal_v3(face_plane, face_snap_point, face_normal);
bool is_aligned = fabsf(dot_v3v3(plane, face_plane)) > (1.0f - CONSTRAIN_EPSILON);
if (!is_aligned && isect_plane_plane_v3(plane, face_plane, isect_orig, isect_dir)) {
closest_to_ray_v3(r_out, face_snap_point, isect_orig, isect_dir);
sub_v3_v3(r_out, t->tsnap.snapTarget);
}
}
void transform_constraint_snap_axis_to_edge(const TransInfo *t,
const float axis[3],
float r_out[3])
{
float lambda;
const float *edge_snap_point = t->tsnap.snapPoint;
const float *edge_dir = t->tsnap.snapNormal;
bool is_aligned = fabsf(dot_v3v3(axis, edge_dir)) > (1.0f - CONSTRAIN_EPSILON);
if (!is_aligned &&
isect_ray_ray_v3(t->tsnap.snapTarget, axis, edge_snap_point, edge_dir, &lambda, NULL)) {
mul_v3_v3fl(r_out, axis, lambda);
}
}
void transform_constraint_snap_axis_to_face(const TransInfo *t,
const float axis[3],
float r_out[3])
{
float lambda;
float face_plane[4];
const float *face_snap_point = t->tsnap.snapPoint;
const float *face_normal = t->tsnap.snapNormal;
plane_from_point_normal_v3(face_plane, face_snap_point, face_normal);
bool is_aligned = fabsf(dot_v3v3(axis, face_plane)) < CONSTRAIN_EPSILON;
if (!is_aligned && isect_ray_plane_v3(t->tsnap.snapTarget, axis, face_plane, &lambda, false)) {
mul_v3_v3fl(r_out, axis, lambda);
}
}
/**
* Return true if the 2x axis are both aligned when projected into the view.
* In this case, we can't usefully project the cursor onto the plane.
*/
static bool isPlaneProjectionViewAligned(const TransInfo *t, const float plane[4])
{
const float eps = 0.001f;
float view_to_plane[3];
view_vector_calc(t, t->center_global, view_to_plane);
float factor = dot_v3v3(plane, view_to_plane);
return fabsf(factor) < eps;
}
static void planeProjection(const TransInfo *t, const float in[3], float out[3])
{
float vec[3], factor, norm[3];
add_v3_v3v3(vec, in, t->center_global);
view_vector_calc(t, vec, norm);
sub_v3_v3v3(vec, out, in);
factor = dot_v3v3(vec, norm);
if (factor == 0.0f) {
return; /* prevent divide by zero */
}
factor = dot_v3v3(vec, vec) / factor;
copy_v3_v3(vec, norm);
mul_v3_fl(vec, factor);
add_v3_v3v3(out, in, vec);
}
static short transform_orientation_or_default(const TransInfo *t)
{
short orientation = t->orient[t->orient_curr].type;
if (orientation == V3D_ORIENT_CUSTOM_MATRIX) {
/* Use the real value of the "orient_type". */
orientation = t->orient[O_DEFAULT].type;
}
return orientation;
}
static const float (*transform_object_axismtx_get(const TransInfo *t,
const TransDataContainer *UNUSED(tc),
const TransData *td))[3]
{
if (transform_orientation_or_default(t) == V3D_ORIENT_GIMBAL) {
BLI_assert(t->orient_type_mask & (1 << V3D_ORIENT_GIMBAL));
if (t->options & (CTX_POSE_BONE | CTX_OBJECT)) {
return td->ext->axismtx_gimbal;
}
}
return td->axismtx;
}
/**
* Generic callback for constant spatial constraints applied to linear motion
*
* The `in` vector in projected into the constrained space and then further
* projected along the view vector.
* (in perspective mode, the view vector is relative to the position on screen)
*/
static void applyAxisConstraintVec(const TransInfo *t,
const TransDataContainer *UNUSED(tc),
const TransData *td,
const float in[3],
float out[3])
{
copy_v3_v3(out, in);
if (!td && t->con.mode & CON_APPLY) {
bool is_snap_to_point = false, is_snap_to_edge = false, is_snap_to_face = false;
mul_m3_v3(t->con.pmtx, out);
if (activeSnap(t)) {
if (validSnap(t)) {
is_snap_to_edge = (t->tsnap.snapElem & SCE_SNAP_MODE_EDGE) != 0;
is_snap_to_face = (t->tsnap.snapElem & SCE_SNAP_MODE_FACE_RAYCAST) != 0;
is_snap_to_point = !is_snap_to_edge && !is_snap_to_face;
}
else if (t->tsnap.snapElem & SCE_SNAP_MODE_GRID) {
is_snap_to_point = true;
}
}
/* With snap points, a projection is alright, no adjustments needed. */
if (!is_snap_to_point || is_snap_to_edge || is_snap_to_face) {
const int dims = getConstraintSpaceDimension(t);
if (dims == 2) {
if (!is_zero_v3(out)) {
float plane[4];
constraint_plane_calc(t, plane);
if (is_snap_to_edge) {
constraint_snap_plane_to_edge(t, plane, out);
}
else if (is_snap_to_face) {
/* Disabled, as it has not proven to be really useful. (See T82386). */
// constraint_snap_plane_to_face(t, plane, out);
}
else {
/* View alignment correction. */
if (!isPlaneProjectionViewAligned(t, plane)) {
planeProjection(t, in, out);
}
}
}
}
else if (dims == 1) {
float c[3];
if (t->con.mode & CON_AXIS0) {
copy_v3_v3(c, t->spacemtx[0]);
}
else if (t->con.mode & CON_AXIS1) {
copy_v3_v3(c, t->spacemtx[1]);
}
else {
BLI_assert(t->con.mode & CON_AXIS2);
copy_v3_v3(c, t->spacemtx[2]);
}
if (is_snap_to_edge) {
transform_constraint_snap_axis_to_edge(t, c, out);
}
else if (is_snap_to_face) {
transform_constraint_snap_axis_to_face(t, c, out);
}
else {
/* View alignment correction. */
axisProjection(t, c, in, out);
}
}
}
}
}
/**
* Generic callback for object based spatial constraints applied to linear motion
*
* At first, the following is applied without orientation
* The IN vector in projected into the constrained space and then further
* projected along the view vector.
* (in perspective mode, the view vector is relative to the position on screen).
*
* Further down, that vector is mapped to each data's space.
*/
static void applyObjectConstraintVec(const TransInfo *t,
const TransDataContainer *tc,
const TransData *td,
const float in[3],
float out[3])
{
if (!td) {
applyAxisConstraintVec(t, tc, td, in, out);
}
else {
/* Specific TransData's space. */
copy_v3_v3(out, in);
if (t->con.mode & CON_APPLY) {
mul_m3_v3(t->spacemtx_inv, out);
const float(*axismtx)[3] = transform_object_axismtx_get(t, tc, td);
mul_m3_v3(axismtx, out);
if (t->flag & T_EDIT) {
mul_m3_v3(tc->mat3_unit, out);
}
}
}
}
/**
* Generic callback for constant spatial constraints applied to resize motion.
*/
static void applyAxisConstraintSize(const TransInfo *t,
const TransDataContainer *UNUSED(tc),
const TransData *td,
float r_smat[3][3])
{
if (!td && t->con.mode & CON_APPLY) {
float tmat[3][3];
if (!(t->con.mode & CON_AXIS0)) {
r_smat[0][0] = 1.0f;
}
if (!(t->con.mode & CON_AXIS1)) {
r_smat[1][1] = 1.0f;
}
if (!(t->con.mode & CON_AXIS2)) {
r_smat[2][2] = 1.0f;
}
mul_m3_m3m3(tmat, r_smat, t->spacemtx_inv);
mul_m3_m3m3(r_smat, t->spacemtx, tmat);
}
}
/**
* Callback for object based spatial constraints applied to resize motion.
*/
static void applyObjectConstraintSize(const TransInfo *t,
const TransDataContainer *tc,
const TransData *td,
float r_smat[3][3])
{
if (td && t->con.mode & CON_APPLY) {
float tmat[3][3];
float imat[3][3];
const float(*axismtx)[3] = transform_object_axismtx_get(t, tc, td);
invert_m3_m3(imat, axismtx);
if (!(t->con.mode & CON_AXIS0)) {
r_smat[0][0] = 1.0f;
}
if (!(t->con.mode & CON_AXIS1)) {
r_smat[1][1] = 1.0f;
}
if (!(t->con.mode & CON_AXIS2)) {
r_smat[2][2] = 1.0f;
}
mul_m3_m3m3(tmat, r_smat, imat);
if (t->flag & T_EDIT) {
mul_m3_m3m3(r_smat, tc->mat3_unit, r_smat);
}
mul_m3_m3m3(r_smat, axismtx, tmat);
}
}
static void constraints_rotation_impl(const TransInfo *t,
const float axismtx[3][3],
float r_axis[3],
float *r_angle)
{
BLI_assert(t->con.mode & CON_APPLY);
int mode = t->con.mode & (CON_AXIS0 | CON_AXIS1 | CON_AXIS2);
switch (mode) {
case CON_AXIS0:
case (CON_AXIS1 | CON_AXIS2):
copy_v3_v3(r_axis, axismtx[0]);
break;
case CON_AXIS1:
case (CON_AXIS0 | CON_AXIS2):
copy_v3_v3(r_axis, axismtx[1]);
break;
case CON_AXIS2:
case (CON_AXIS0 | CON_AXIS1):
copy_v3_v3(r_axis, axismtx[2]);
break;
}
/* don't flip axis if asked to or if num input */
if (r_angle &&
!((mode & CON_NOFLIP) || hasNumInput(&t->num) || (t->flag & T_INPUT_IS_VALUES_FINAL))) {
float view_vector[3];
view_vector_calc(t, t->center_global, view_vector);
if (dot_v3v3(r_axis, view_vector) > 0.0f) {
*r_angle = -(*r_angle);
}
}
}
/**
* Generic callback for constant spatial constraints applied to rotations
*
* The rotation axis is copied into `vec`.
*
* In the case of single axis constraints, the rotation axis is directly the one constrained to.
* For planar constraints (2 axis), the rotation axis is the normal of the plane.
*
* The following only applies when #CON_NOFLIP is not set.
* The vector is then modified to always point away from the screen (in global space)
* This insures that the rotation is always logically following the mouse.
* (ie: not doing counterclockwise rotations when the mouse moves clockwise).
*/
static void applyAxisConstraintRot(const TransInfo *t,
const TransDataContainer *UNUSED(tc),
const TransData *td,
float r_axis[3],
float *r_angle)
{
if (!td && t->con.mode & CON_APPLY) {
constraints_rotation_impl(t, t->spacemtx, r_axis, r_angle);
}
}
/**
* Callback for object based spatial constraints applied to rotations
*
* The rotation axis is copied into `vec`.
*
* In the case of single axis constraints, the rotation axis is directly the one constrained to.
* For planar constraints (2 axis), the rotation axis is the normal of the plane.
*
* The following only applies when #CON_NOFLIP is not set.
* The vector is then modified to always point away from the screen (in global space)
* This insures that the rotation is always logically following the mouse.
* (ie: not doing counterclockwise rotations when the mouse moves clockwise).
*/
static void applyObjectConstraintRot(const TransInfo *t,
const TransDataContainer *tc,
const TransData *td,
float r_axis[3],
float *r_angle)
{
if (t->con.mode & CON_APPLY) {
float tmp_axismtx[3][3];
const float(*axismtx)[3];
/* on setup call, use first object */
if (td == NULL) {
BLI_assert(tc == NULL);
tc = TRANS_DATA_CONTAINER_FIRST_OK(t);
td = tc->data;
}
if (t->flag & T_EDIT) {
mul_m3_m3m3(tmp_axismtx, tc->mat3_unit, td->axismtx);
axismtx = tmp_axismtx;
}
else {
axismtx = transform_object_axismtx_get(t, tc, td);
}
constraints_rotation_impl(t, axismtx, r_axis, r_angle);
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Internal Setup Calls
* \{ */
void setConstraint(TransInfo *t, int mode, const char text[])
{
BLI_strncpy(t->con.text + 1, text, sizeof(t->con.text) - 1);
t->con.mode = mode;
projection_matrix_calc(t, t->con.pmtx);
startConstraint(t);
t->con.drawExtra = NULL;
t->con.applyVec = applyAxisConstraintVec;
t->con.applySize = applyAxisConstraintSize;
t->con.applyRot = applyAxisConstraintRot;
t->redraw = TREDRAW_HARD;
}
void setAxisMatrixConstraint(TransInfo *t, int mode, const char text[])
{
BLI_strncpy(t->con.text + 1, text, sizeof(t->con.text) - 1);
t->con.mode = mode;
projection_matrix_calc(t, t->con.pmtx);
startConstraint(t);
t->con.drawExtra = drawObjectConstraint;
t->con.applyVec = applyObjectConstraintVec;
t->con.applySize = applyObjectConstraintSize;
t->con.applyRot = applyObjectConstraintRot;
t->redraw = TREDRAW_HARD;
}
void setLocalConstraint(TransInfo *t, int mode, const char text[])
{
if ((t->flag & T_EDIT) || t->data_len_all == 1) {
/* Although in edit-mode each object has its local space, use the
* orientation of the active object. */
setConstraint(t, mode, text);
}
else {
setAxisMatrixConstraint(t, mode, text);
}
}
void setUserConstraint(TransInfo *t, int mode, const char ftext[])
{
char text[256];
const short orientation = transform_orientation_or_default(t);
const char *spacename = transform_orientations_spacename_get(t, orientation);
BLI_snprintf(text, sizeof(text), ftext, spacename);
switch (orientation) {
case V3D_ORIENT_LOCAL:
case V3D_ORIENT_GIMBAL:
setLocalConstraint(t, mode, text);
break;
case V3D_ORIENT_NORMAL:
if (checkUseAxisMatrix(t)) {
setAxisMatrixConstraint(t, mode, text);
break;
}
ATTR_FALLTHROUGH;
case V3D_ORIENT_GLOBAL:
case V3D_ORIENT_VIEW:
case V3D_ORIENT_CURSOR:
case V3D_ORIENT_CUSTOM_MATRIX:
case V3D_ORIENT_CUSTOM:
default: {
setConstraint(t, mode, text);
break;
}
}
t->con.mode |= CON_USER;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Drawing Constraints
* \{ */
void drawConstraint(TransInfo *t)
{
TransCon *tc = &(t->con);
if (!ELEM(t->spacetype, SPACE_VIEW3D, SPACE_IMAGE, SPACE_NODE, SPACE_SEQ)) {
return;
}
if (!(tc->mode & CON_APPLY)) {
return;
}
if (t->flag & T_NO_CONSTRAINT) {
return;
}
if (tc->drawExtra) {
tc->drawExtra(t);
}
else {
if (tc->mode & CON_SELECT) {
float vec[3];
convertViewVec(t, vec, (t->mval[0] - t->con.imval[0]), (t->mval[1] - t->con.imval[1]));
add_v3_v3(vec, t->center_global);
drawLine(t, t->center_global, t->spacemtx[0], 'X', 0);
drawLine(t, t->center_global, t->spacemtx[1], 'Y', 0);
drawLine(t, t->center_global, t->spacemtx[2], 'Z', 0);
eGPUDepthTest depth_test_enabled = GPU_depth_test_get();
if (depth_test_enabled) {
GPU_depth_test(GPU_DEPTH_NONE);
}
const uint shdr_pos = GPU_vertformat_attr_add(
immVertexFormat(), "pos", GPU_COMP_F32, 3, GPU_FETCH_FLOAT);
immBindBuiltinProgram(GPU_SHADER_3D_LINE_DASHED_UNIFORM_COLOR);
float viewport_size[4];
GPU_viewport_size_get_f(viewport_size);
immUniform2f("viewport_size", viewport_size[2], viewport_size[3]);
immUniform1i("colors_len", 0); /* "simple" mode */
immUniformColor4f(1.0f, 1.0f, 1.0f, 1.0f);
immUniform1f("dash_width", 2.0f);
immUniform1f("dash_factor", 0.5f);
immBegin(GPU_PRIM_LINES, 2);
immVertex3fv(shdr_pos, t->center_global);
immVertex3fv(shdr_pos, vec);
immEnd();
immUnbindProgram();
if (depth_test_enabled) {
GPU_depth_test(GPU_DEPTH_LESS_EQUAL);
}
}
if (tc->mode & CON_AXIS0) {
drawLine(t, t->center_global, t->spacemtx[0], 'X', DRAWLIGHT);
}
if (tc->mode & CON_AXIS1) {
drawLine(t, t->center_global, t->spacemtx[1], 'Y', DRAWLIGHT);
}
if (tc->mode & CON_AXIS2) {
drawLine(t, t->center_global, t->spacemtx[2], 'Z', DRAWLIGHT);
}
}
}
void drawPropCircle(const struct bContext *C, TransInfo *t)
{
if (t->flag & T_PROP_EDIT) {
RegionView3D *rv3d = CTX_wm_region_view3d(C);
float tmat[4][4], imat[4][4];
if (t->spacetype == SPACE_VIEW3D && rv3d != NULL) {
copy_m4_m4(tmat, rv3d->viewmat);
invert_m4_m4(imat, tmat);
}
else {
unit_m4(tmat);
unit_m4(imat);
}
GPU_matrix_push();
if (t->spacetype == SPACE_VIEW3D) {
/* pass */
}
else if (t->spacetype == SPACE_IMAGE) {
GPU_matrix_scale_2f(1.0f / t->aspect[0], 1.0f / t->aspect[1]);
}
else if (ELEM(t->spacetype, SPACE_GRAPH, SPACE_ACTION)) {
/* only scale y */
rcti *mask = &t->region->v2d.mask;
rctf *datamask = &t->region->v2d.cur;
float xsize = BLI_rctf_size_x(datamask);
float ysize = BLI_rctf_size_y(datamask);
float xmask = BLI_rcti_size_x(mask);
float ymask = BLI_rcti_size_y(mask);
GPU_matrix_scale_2f(1.0f, (ysize / xsize) * (xmask / ymask));
}
eGPUDepthTest depth_test_enabled = GPU_depth_test_get();
if (depth_test_enabled) {
GPU_depth_test(GPU_DEPTH_NONE);
}
uint pos = GPU_vertformat_attr_add(immVertexFormat(), "pos", GPU_COMP_F32, 3, GPU_FETCH_FLOAT);
immBindBuiltinProgram(GPU_SHADER_3D_POLYLINE_UNIFORM_COLOR);
float viewport[4];
GPU_viewport_size_get_f(viewport);
GPU_blend(GPU_BLEND_ALPHA);
immUniform2fv("viewportSize", &viewport[2]);
immUniform1f("lineWidth", 3.0f * U.pixelsize);
immUniformThemeColorShadeAlpha(TH_GRID, -20, 255);
imm_drawcircball(t->center_global, t->prop_size, imat, pos);
immUniform1f("lineWidth", 1.0f * U.pixelsize);
immUniformThemeColorShadeAlpha(TH_GRID, 20, 255);
imm_drawcircball(t->center_global, t->prop_size, imat, pos);
immUnbindProgram();
if (depth_test_enabled) {
GPU_depth_test(GPU_DEPTH_LESS_EQUAL);
}
GPU_matrix_pop();
}
}
static void drawObjectConstraint(TransInfo *t)
{
/* Draw the first one lighter because that's the one who controls the others.
* Meaning the transformation is projected on that one and just copied on the others
* constraint space.
* In a nutshell, the object with light axis is controlled by the user and the others follow.
* Without drawing the first light, users have little clue what they are doing.
*/
short options = DRAWLIGHT;
float tmp_axismtx[3][3];
FOREACH_TRANS_DATA_CONTAINER (t, tc) {
TransData *td = tc->data;
for (int i = 0; i < tc->data_len; i++, td++) {
float co[3];
const float(*axismtx)[3];
if (t->flag & T_PROP_EDIT) {
/* we're sorted, so skip the rest */
if (td->factor == 0.0f) {
break;
}
}
if (t->options & CTX_GPENCIL_STROKES) {
/* only draw a constraint line for one point, otherwise we can't see anything */
if ((options & DRAWLIGHT) == 0) {
break;
}
}
if (t->options & CTX_SEQUENCER_IMAGE) {
/* Because we construct an "L" shape to deform the sequence, we should skip
* all points except the first vertex. Otherwise we will draw the same axis constraint line
* 3 times for each strip.
*/
if (i % 3 != 0) {
continue;
}
}
if (t->flag & T_EDIT) {
mul_v3_m4v3(co, tc->mat, td->center);
mul_m3_m3m3(tmp_axismtx, tc->mat3_unit, td->axismtx);
axismtx = tmp_axismtx;
}
else {
if (t->options & CTX_POSE_BONE) {
mul_v3_m4v3(co, tc->mat, td->center);
}
else {
copy_v3_v3(co, td->center);
}
axismtx = transform_object_axismtx_get(t, tc, td);
}
if (t->con.mode & CON_AXIS0) {
drawLine(t, co, axismtx[0], 'X', options);
}
if (t->con.mode & CON_AXIS1) {
drawLine(t, co, axismtx[1], 'Y', options);
}
if (t->con.mode & CON_AXIS2) {
drawLine(t, co, axismtx[2], 'Z', options);
}
options &= ~DRAWLIGHT;
}
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Start / Stop Constraints
* \{ */
void startConstraint(TransInfo *t)
{
t->con.mode |= CON_APPLY;
*t->con.text = ' ';
t->num.idx_max = min_ii(getConstraintSpaceDimension(t) - 1, t->idx_max);
}
void stopConstraint(TransInfo *t)
{
if (t->orient_curr != O_DEFAULT) {
transform_orientations_current_set(t, O_DEFAULT);
}
t->con.mode &= ~(CON_APPLY | CON_SELECT);
*t->con.text = '\0';
t->num.idx_max = t->idx_max;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Middle Mouse Button Select
* \{ */
void initSelectConstraint(TransInfo *t)
{
if (t->orient_curr == O_DEFAULT) {
transform_orientations_current_set(t, O_SCENE);
}
setUserConstraint(t, CON_APPLY | CON_SELECT, "%s");
}
void selectConstraint(TransInfo *t)
{
if (t->con.mode & CON_SELECT) {
setNearestAxis(t);
startConstraint(t);
}
}
void postSelectConstraint(TransInfo *t)
{
t->con.mode &= ~CON_SELECT;
if (!(t->con.mode & (CON_AXIS0 | CON_AXIS1 | CON_AXIS2))) {
t->con.mode &= ~CON_APPLY;
}
}
static void setNearestAxis2d(TransInfo *t)
{
/* no correction needed... just use whichever one is lower */
if (abs(t->mval[0] - t->con.imval[0]) < abs(t->mval[1] - t->con.imval[1])) {
t->con.mode |= CON_AXIS1;
BLI_strncpy(t->con.text, TIP_(" along Y axis"), sizeof(t->con.text));
}
else {
t->con.mode |= CON_AXIS0;
BLI_strncpy(t->con.text, TIP_(" along X axis"), sizeof(t->con.text));
}
}
static void setNearestAxis3d(TransInfo *t)
{
float zfac;
float mvec[3], proj[3];
float len[3];
int i;
/* calculate mouse movement */
mvec[0] = (float)(t->mval[0] - t->con.imval[0]);
mvec[1] = (float)(t->mval[1] - t->con.imval[1]);
mvec[2] = 0.0f;
/* We need to correct axis length for the current zoom-level of view,
* this to prevent projected values to be clipped behind the camera
* and to overflow the short integers.
* The formula used is a bit stupid, just a simplification of the subtraction
* of two 2D points 30 pixels apart (that's the last factor in the formula) after
* projecting them with #ED_view3d_win_to_delta and then get the length of that vector. */
zfac = mul_project_m4_v3_zfac(t->persmat, t->center_global);
zfac = len_v3(t->persinv[0]) * 2.0f / t->region->winx * zfac * 30.0f;
for (i = 0; i < 3; i++) {
float axis[3], axis_2d[2];
copy_v3_v3(axis, t->spacemtx[i]);
mul_v3_fl(axis, zfac);
/* now we can project to get window coordinate */
add_v3_v3(axis, t->center_global);
projectFloatView(t, axis, axis_2d);
sub_v2_v2v2(axis, axis_2d, t->center2d);
axis[2] = 0.0f;
if (normalize_v3(axis) > 1e-3f) {
project_v3_v3v3(proj, mvec, axis);
sub_v3_v3v3(axis, mvec, proj);
len[i] = normalize_v3(axis);
}
else {
len[i] = 1e10f;
}
}
if (len[0] <= len[1] && len[0] <= len[2]) {
if (t->modifiers & MOD_CONSTRAINT_SELECT_PLANE) {
t->con.mode |= (CON_AXIS1 | CON_AXIS2);
BLI_snprintf(t->con.text, sizeof(t->con.text), TIP_(" locking %s X axis"), t->spacename);
}
else {
t->con.mode |= CON_AXIS0;
BLI_snprintf(t->con.text, sizeof(t->con.text), TIP_(" along %s X axis"), t->spacename);
}
}
else if (len[1] <= len[0] && len[1] <= len[2]) {
if (t->modifiers & MOD_CONSTRAINT_SELECT_PLANE) {
t->con.mode |= (CON_AXIS0 | CON_AXIS2);
BLI_snprintf(t->con.text, sizeof(t->con.text), TIP_(" locking %s Y axis"), t->spacename);
}
else {
t->con.mode |= CON_AXIS1;
BLI_snprintf(t->con.text, sizeof(t->con.text), TIP_(" along %s Y axis"), t->spacename);
}
}
else if (len[2] <= len[1] && len[2] <= len[0]) {
if (t->modifiers & MOD_CONSTRAINT_SELECT_PLANE) {
t->con.mode |= (CON_AXIS0 | CON_AXIS1);
BLI_snprintf(t->con.text, sizeof(t->con.text), TIP_(" locking %s Z axis"), t->spacename);
}
else {
t->con.mode |= CON_AXIS2;
BLI_snprintf(t->con.text, sizeof(t->con.text), TIP_(" along %s Z axis"), t->spacename);
}
}
}
void setNearestAxis(TransInfo *t)
{
/* clear any prior constraint flags */
t->con.mode &= ~CON_AXIS0;
t->con.mode &= ~CON_AXIS1;
t->con.mode &= ~CON_AXIS2;
/* constraint setting - depends on spacetype */
if (t->spacetype == SPACE_VIEW3D) {
/* 3d-view */
setNearestAxis3d(t);
}
else {
/* assume that this means a 2D-Editor */
setNearestAxis2d(t);
}
projection_matrix_calc(t, t->con.pmtx);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Helper Functions
* \{ */
int constraintModeToIndex(const TransInfo *t)
{
if ((t->con.mode & CON_APPLY) == 0) {
return -1;
}
switch (t->con.mode & (CON_AXIS0 | CON_AXIS1 | CON_AXIS2)) {
case (CON_AXIS0):
case (CON_AXIS1 | CON_AXIS2):
return 0;
case (CON_AXIS1):
case (CON_AXIS0 | CON_AXIS2):
return 1;
case (CON_AXIS2):
case (CON_AXIS0 | CON_AXIS1):
return 2;
default:
return -1;
}
}
bool isLockConstraint(const TransInfo *t)
{
int mode = t->con.mode;
if ((mode & (CON_AXIS0 | CON_AXIS1)) == (CON_AXIS0 | CON_AXIS1)) {
return true;
}
if ((mode & (CON_AXIS1 | CON_AXIS2)) == (CON_AXIS1 | CON_AXIS2)) {
return true;
}
if ((mode & (CON_AXIS0 | CON_AXIS2)) == (CON_AXIS0 | CON_AXIS2)) {
return true;
}
return false;
}
int getConstraintSpaceDimension(const TransInfo *t)
{
int n = 0;
if (t->con.mode & CON_AXIS0) {
n++;
}
if (t->con.mode & CON_AXIS1) {
n++;
}
if (t->con.mode & CON_AXIS2) {
n++;
}
return n;
/* Someone willing to do it cryptically could do the following instead:
*
* `return t->con & (CON_AXIS0|CON_AXIS1|CON_AXIS2);`
*
* Based on the assumptions that the axis flags are one after the other and start at 1
*/
}
/** \} */