tornavis/source/blender/bmesh/operators/bmo_bridge.cc

659 lines
20 KiB
C++

/* SPDX-FileCopyrightText: 2023 Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup bmesh
*
* Connect verts across faces (splits faces) and bridge tool.
*/
#include "BLI_listbase.h"
#include "BLI_math_vector.h"
#include "BLI_utildefines.h"
#include "bmesh.hh"
#include "intern/bmesh_operators_private.hh" /* own include */
/**
* TODO(@ideasman42): Many connected edge loops can cause an error attempting
* to create faces with duplicate vertices. While this needs to be investigated,
* it's simple enough to check for this case, see: #102232.
*/
#define USE_DUPLICATE_FACE_VERT_CHECK
#define EDGE_MARK 4
#define EDGE_OUT 8
#define FACE_OUT 16
/* el_a and el_b _must_ be same size */
static void bm_bridge_splice_loops(BMesh *bm,
LinkData *el_a,
LinkData *el_b,
const float merge_factor)
{
BMOperator op_weld;
BMOpSlot *slot_targetmap;
BMO_op_init(bm, &op_weld, 0, "weld_verts");
slot_targetmap = BMO_slot_get(op_weld.slots_in, "targetmap");
do {
BMVert *v_a = static_cast<BMVert *>(el_a->data), *v_b = static_cast<BMVert *>(el_b->data);
BM_data_interp_from_verts(bm, v_a, v_b, v_b, merge_factor);
interp_v3_v3v3(v_b->co, v_a->co, v_b->co, merge_factor);
BLI_assert(v_a != v_b);
BMO_slot_map_elem_insert(&op_weld, slot_targetmap, v_a, v_b);
} while ((void)(el_b = el_b->next), (el_a = el_a->next));
BMO_op_exec(bm, &op_weld);
BMO_op_finish(bm, &op_weld);
}
/* get the 2 loops matching 2 verts.
* first attempt to get the face corners that use the edge defined by v1 & v2,
* if that fails just get any loop that's on the vert (the first one) */
static void bm_vert_loop_pair(BMesh *bm, BMVert *v1, BMVert *v2, BMLoop **l1, BMLoop **l2)
{
BMEdge *e = BM_edge_exists(v1, v2);
BMLoop *l = e->l;
if (l) {
if (l->v == v1) {
*l1 = l;
*l2 = l->next;
}
else {
*l2 = l;
*l1 = l->next;
}
}
else {
/* fallback to _any_ loop */
*l1 = static_cast<BMLoop *>(BM_iter_at_index(bm, BM_LOOPS_OF_VERT, v1, 0));
*l2 = static_cast<BMLoop *>(BM_iter_at_index(bm, BM_LOOPS_OF_VERT, v2, 0));
}
}
/* el_b can have any offset */
static float bm_edgeloop_offset_length(LinkData *el_a,
LinkData *el_b,
LinkData *el_b_first,
const float len_max)
{
float len = 0.0f;
BLI_assert(el_a->prev == nullptr); /* must be first */
do {
len += len_v3v3(((BMVert *)el_a->data)->co, ((BMVert *)el_b->data)->co);
} while ((void)(el_b = el_b->next ? el_b->next : el_b_first),
(el_a = el_a->next) && (len < len_max));
return len;
}
static void bm_bridge_best_rotation(BMEdgeLoopStore *el_store_a, BMEdgeLoopStore *el_store_b)
{
ListBase *lb_a = BM_edgeloop_verts_get(el_store_a);
ListBase *lb_b = BM_edgeloop_verts_get(el_store_b);
LinkData *el_a = static_cast<LinkData *>(lb_a->first);
LinkData *el_b = static_cast<LinkData *>(lb_b->first);
LinkData *el_b_first = el_b;
LinkData *el_b_best = nullptr;
float len_best = FLT_MAX;
for (; el_b; el_b = el_b->next) {
const float len = bm_edgeloop_offset_length(el_a, el_b, el_b_first, len_best);
if (len < len_best) {
el_b_best = el_b;
len_best = len;
}
}
if (el_b_best) {
BLI_listbase_rotate_first(lb_b, el_b_best);
}
}
static void bm_face_edges_tag_out(BMesh *bm, BMFace *f)
{
BMLoop *l_iter, *l_first;
l_iter = l_first = BM_FACE_FIRST_LOOP(f);
do {
BMO_edge_flag_enable(bm, l_iter->e, EDGE_OUT);
} while ((l_iter = l_iter->next) != l_first);
}
static bool bm_edge_test_cb(BMEdge *e, void *bm_v)
{
return BMO_edge_flag_test((BMesh *)bm_v, e, EDGE_MARK);
}
static void bridge_loop_pair(BMesh *bm,
BMEdgeLoopStore *el_store_a,
BMEdgeLoopStore *el_store_b,
const bool use_merge,
const float merge_factor,
const int twist_offset)
{
const float eps = 0.00001f;
LinkData *el_a_first, *el_b_first;
const bool is_closed = BM_edgeloop_is_closed(el_store_a) && BM_edgeloop_is_closed(el_store_b);
int el_store_a_len, el_store_b_len;
bool el_store_b_free = false;
float el_dir[3];
float dot_a, dot_b;
const bool use_edgeout = true;
el_store_a_len = BM_edgeloop_length_get((BMEdgeLoopStore *)el_store_a);
el_store_b_len = BM_edgeloop_length_get((BMEdgeLoopStore *)el_store_b);
if (el_store_a_len < el_store_b_len) {
std::swap(el_store_a_len, el_store_b_len);
std::swap(el_store_a, el_store_b);
}
if (use_merge) {
BLI_assert(el_store_a_len == el_store_b_len);
}
if (el_store_a_len != el_store_b_len) {
BM_mesh_elem_hflag_disable_all(bm, BM_FACE | BM_EDGE, BM_ELEM_TAG, false);
}
sub_v3_v3v3(el_dir, BM_edgeloop_center_get(el_store_a), BM_edgeloop_center_get(el_store_b));
if (is_closed) {
/* if all loops are closed this will calculate twice for all loops */
BM_edgeloop_calc_normal(bm, el_store_a);
BM_edgeloop_calc_normal(bm, el_store_b);
}
else {
ListBase *lb_a = BM_edgeloop_verts_get(el_store_a);
ListBase *lb_b = BM_edgeloop_verts_get(el_store_b);
/* normalizing isn't strictly needed but without we may get very large values */
float no[3];
float dir_a_orig[3], dir_b_orig[3];
float dir_a[3], dir_b[3];
const float *test_a, *test_b;
sub_v3_v3v3(dir_a_orig,
((BMVert *)(((LinkData *)lb_a->first)->data))->co,
((BMVert *)(((LinkData *)lb_a->last)->data))->co);
sub_v3_v3v3(dir_b_orig,
((BMVert *)(((LinkData *)lb_b->first)->data))->co,
((BMVert *)(((LinkData *)lb_b->last)->data))->co);
/* make the directions point out from the normals, 'no' is used as a temp var */
cross_v3_v3v3(no, dir_a_orig, el_dir);
cross_v3_v3v3(dir_a, no, el_dir);
cross_v3_v3v3(no, dir_b_orig, el_dir);
cross_v3_v3v3(dir_b, no, el_dir);
if (LIKELY(!is_zero_v3(dir_a) && !is_zero_v3(dir_b))) {
test_a = dir_a;
test_b = dir_b;
}
else {
/**
* This is a corner case:
*
* <pre>
* (loop a) (loop b)
* +--------+ +--------+
* </pre>
*
* When loops are aligned to the direction between
* the loops values of 'dir_a/b' is degenerate,
* in this case compare the original directions
* (before they were corrected by 'el_dir'),
* see: #43013
*/
test_a = dir_a_orig;
test_b = dir_b_orig;
}
if (dot_v3v3(test_a, test_b) < 0.0f) {
BM_edgeloop_flip(bm, el_store_b);
}
normalize_v3_v3(no, el_dir);
BM_edgeloop_calc_normal_aligned(bm, el_store_a, no);
BM_edgeloop_calc_normal_aligned(bm, el_store_b, no);
}
dot_a = dot_v3v3(BM_edgeloop_normal_get(el_store_a), el_dir);
dot_b = dot_v3v3(BM_edgeloop_normal_get(el_store_b), el_dir);
if (UNLIKELY((len_squared_v3(el_dir) < eps) || ((fabsf(dot_a) < eps) && (fabsf(dot_b) < eps)))) {
/* in this case there is no depth between the two loops,
* eg: 2x 2d circles, one scaled smaller,
* in this case 'el_dir' can't be used, just ensure we have matching flipping. */
if (dot_v3v3(BM_edgeloop_normal_get(el_store_a), BM_edgeloop_normal_get(el_store_b)) < 0.0f) {
BM_edgeloop_flip(bm, el_store_b);
}
}
else if ((dot_a < 0.0f) != (dot_b < 0.0f)) {
BM_edgeloop_flip(bm, el_store_b);
}
/* we only care about flipping if we make faces */
if (use_merge == false) {
float no[3];
add_v3_v3v3(no, BM_edgeloop_normal_get(el_store_a), BM_edgeloop_normal_get(el_store_b));
if (dot_v3v3(no, el_dir) < 0.0f) {
BM_edgeloop_flip(bm, el_store_a);
BM_edgeloop_flip(bm, el_store_b);
}
/* vote on winding (so new face winding is based on existing connected faces) */
if (bm->totface) {
BMEdgeLoopStore *estore_pair[2] = {el_store_a, el_store_b};
int i;
int winding_votes[2] = {0, 0};
int winding_dir = 1;
for (i = 0; i < 2; i++, winding_dir = -winding_dir) {
LISTBASE_FOREACH (LinkData *, el, BM_edgeloop_verts_get(estore_pair[i])) {
LinkData *el_next = BM_EDGELINK_NEXT(estore_pair[i], el);
if (el_next) {
BMEdge *e = BM_edge_exists(static_cast<BMVert *>(el->data),
static_cast<BMVert *>(el_next->data));
if (e && BM_edge_is_boundary(e)) {
winding_votes[i] += ((e->l->v == el->data) ? winding_dir : -winding_dir);
}
}
}
}
if (winding_votes[0] || winding_votes[1]) {
bool flip[2] = {false, false};
/* for direction aligned loops we can't rely on the directly we have,
* use the winding defined by the connected faces (see #48356). */
if (fabsf(dot_a) < eps) {
if (winding_votes[0] < 0) {
flip[0] = !flip[0];
winding_votes[0] *= -1;
}
}
if (fabsf(dot_b) < eps) {
if (winding_votes[1] < 0) {
flip[1] = !flip[1];
winding_votes[1] *= -1;
}
}
/* when both loops contradict the winding, flip them so surrounding geometry matches */
if ((winding_votes[0] + winding_votes[1]) < 0) {
flip[0] = !flip[0];
flip[1] = !flip[1];
/* valid but unused */
#if 0
winding_votes[0] *= -1;
winding_votes[1] *= -1;
#endif
}
if (flip[0]) {
BM_edgeloop_flip(bm, el_store_a);
}
if (flip[1]) {
BM_edgeloop_flip(bm, el_store_b);
}
}
}
}
if (el_store_a_len > el_store_b_len) {
el_store_b = BM_edgeloop_copy(el_store_b);
BM_edgeloop_expand(bm, el_store_b, el_store_a_len, false, nullptr);
el_store_b_free = true;
}
if (is_closed) {
bm_bridge_best_rotation(el_store_a, el_store_b);
/* add twist */
if (twist_offset != 0) {
const int len_b = BM_edgeloop_length_get(el_store_b);
ListBase *lb_b = BM_edgeloop_verts_get(el_store_b);
LinkData *el_b = static_cast<LinkData *>(BLI_rfindlink(lb_b, mod_i(twist_offset, len_b)));
BLI_listbase_rotate_first(lb_b, el_b);
}
}
/* Assign after flipping is finalized */
el_a_first = static_cast<LinkData *>(BM_edgeloop_verts_get(el_store_a)->first);
el_b_first = static_cast<LinkData *>(BM_edgeloop_verts_get(el_store_b)->first);
if (use_merge) {
bm_bridge_splice_loops(bm, el_a_first, el_b_first, merge_factor);
}
else {
LinkData *el_a = el_a_first;
LinkData *el_b = el_b_first;
LinkData *el_a_next;
LinkData *el_b_next;
while (true) {
BMFace *f, *f_example;
BMLoop *l_iter;
BMVert *v_a, *v_b, *v_a_next, *v_b_next;
BMLoop *l_a = nullptr;
BMLoop *l_b = nullptr;
BMLoop *l_a_next = nullptr;
BMLoop *l_b_next = nullptr;
if (is_closed) {
el_a_next = BM_EDGELINK_NEXT(el_store_a, el_a);
el_b_next = BM_EDGELINK_NEXT(el_store_b, el_b);
}
else {
el_a_next = el_a->next;
el_b_next = el_b->next;
if (ELEM(nullptr, el_a_next, el_b_next)) {
break;
}
}
v_a = static_cast<BMVert *>(el_a->data);
v_b = static_cast<BMVert *>(el_b->data);
v_a_next = static_cast<BMVert *>(el_a_next->data);
v_b_next = static_cast<BMVert *>(el_b_next->data);
/* get loop data - before making the face */
if (v_b != v_b_next) {
bm_vert_loop_pair(bm, v_a, v_a_next, &l_a, &l_a_next);
bm_vert_loop_pair(bm, v_b, v_b_next, &l_b, &l_b_next);
}
else {
/* lazy, could be more clever here */
bm_vert_loop_pair(bm, v_a, v_a_next, &l_a, &l_a_next);
l_b = l_b_next = static_cast<BMLoop *>(BM_iter_at_index(bm, BM_LOOPS_OF_VERT, v_b, 0));
}
if (l_a && l_a_next == nullptr) {
l_a_next = l_a;
}
if (l_a_next && l_a == nullptr) {
l_a = l_a_next;
}
if (l_b && l_b_next == nullptr) {
l_b_next = l_b;
}
if (l_b_next && l_b == nullptr) {
l_b = l_b_next;
}
f_example = l_a ? l_a->f : (l_b ? l_b->f : nullptr);
if (v_b != v_b_next) {
#ifdef USE_DUPLICATE_FACE_VERT_CHECK /* Only check for duplicates between loops. */
BLI_assert((v_b != v_b_next) && (v_a_next != v_a));
if (UNLIKELY(ELEM(v_b, v_a_next, v_a) || ELEM(v_b_next, v_a_next, v_a))) {
f = nullptr;
}
else
#endif
{
BMVert *v_arr[4] = {v_b, v_b_next, v_a_next, v_a};
f = BM_face_exists(v_arr, 4);
if (f == nullptr) {
/* copy if loop data if its is missing on one ring */
f = BM_face_create_verts(bm, v_arr, 4, nullptr, BM_CREATE_NOP, true);
l_iter = BM_FACE_FIRST_LOOP(f);
if (l_b) {
BM_elem_attrs_copy(bm, l_b, l_iter);
}
l_iter = l_iter->next;
if (l_b_next) {
BM_elem_attrs_copy(bm, l_b_next, l_iter);
}
l_iter = l_iter->next;
if (l_a_next) {
BM_elem_attrs_copy(bm, l_a_next, l_iter);
}
l_iter = l_iter->next;
if (l_a) {
BM_elem_attrs_copy(bm, l_a, l_iter);
}
}
}
}
else {
#ifdef USE_DUPLICATE_FACE_VERT_CHECK /* Only check for duplicates between loops. */
BLI_assert(v_a_next != v_a);
if (UNLIKELY(ELEM(v_b, v_a_next, v_a))) {
f = nullptr;
}
else
#endif
{
BMVert *v_arr[3] = {v_b, v_a_next, v_a};
f = BM_face_exists(v_arr, 3);
if (f == nullptr) {
/* fan-fill a triangle */
f = BM_face_create_verts(bm, v_arr, 3, nullptr, BM_CREATE_NOP, true);
l_iter = BM_FACE_FIRST_LOOP(f);
if (l_b) {
BM_elem_attrs_copy(bm, l_b, l_iter);
}
l_iter = l_iter->next;
if (l_a_next) {
BM_elem_attrs_copy(bm, l_a_next, l_iter);
}
l_iter = l_iter->next;
if (l_a) {
BM_elem_attrs_copy(bm, l_a, l_iter);
}
}
}
}
#ifdef USE_DUPLICATE_FACE_VERT_CHECK
if (f != nullptr)
#endif
{
if (f_example && (f_example != f)) {
BM_elem_attrs_copy(bm, f_example, f);
}
BMO_face_flag_enable(bm, f, FACE_OUT);
BM_elem_flag_enable(f, BM_ELEM_TAG);
/* tag all edges of the face, untag the loop edges after */
if (use_edgeout) {
bm_face_edges_tag_out(bm, f);
}
}
if (el_a_next == el_a_first) {
break;
}
el_a = el_a_next;
el_b = el_b_next;
}
}
if (el_store_a_len != el_store_b_len) {
BMEdgeLoopStore *estore_pair[2] = {el_store_a, el_store_b};
int i;
BMOperator op_sub;
/* when we have to bridge between different sized edge-loops,
* be clever and post-process for best results */
/* triangulate inline */
BMO_op_initf(bm, &op_sub, 0, "triangulate faces=%hf", BM_ELEM_TAG, true);
/* calc normals for input faces before executing */
{
BMOIter siter;
BMFace *f;
BMO_ITER (f, &siter, op_sub.slots_in, "faces", BM_FACE) {
BM_face_normal_update(f);
}
}
BMO_op_exec(bm, &op_sub);
BMO_slot_buffer_flag_enable(bm, op_sub.slots_out, "faces.out", BM_FACE, FACE_OUT);
BMO_slot_buffer_hflag_enable(bm, op_sub.slots_out, "faces.out", BM_FACE, BM_ELEM_TAG, false);
BMO_op_finish(bm, &op_sub);
/* tag verts on each side so we can restrict rotation of edges to verts on the same side */
for (i = 0; i < 2; i++) {
LISTBASE_FOREACH (LinkData *, el, BM_edgeloop_verts_get(estore_pair[i])) {
BM_elem_flag_set((BMVert *)el->data, BM_ELEM_TAG, i);
}
}
BMO_op_initf(bm,
&op_sub,
0,
"beautify_fill faces=%hf edges=ae use_restrict_tag=%b method=%i",
BM_ELEM_TAG,
true,
1);
if (use_edgeout) {
BMOIter siter;
BMFace *f;
BMO_ITER (f, &siter, op_sub.slots_in, "faces", BM_FACE) {
BMO_face_flag_enable(bm, f, FACE_OUT);
bm_face_edges_tag_out(bm, f);
}
}
BMO_op_exec(bm, &op_sub);
/* there may also be tagged faces that didn't rotate, mark input */
if (use_edgeout) {
BMOIter siter;
BMFace *f;
BMO_ITER (f, &siter, op_sub.slots_out, "geom.out", BM_FACE) {
BMO_face_flag_enable(bm, f, FACE_OUT);
bm_face_edges_tag_out(bm, f);
}
}
else {
BMO_slot_buffer_flag_enable(bm, op_sub.slots_out, "geom.out", BM_FACE, FACE_OUT);
}
BMO_op_finish(bm, &op_sub);
}
if (use_edgeout && use_merge == false) {
/* we've enabled all face edges above, now disable all loop edges */
BMEdgeLoopStore *estore_pair[2] = {el_store_a, el_store_b};
int i;
for (i = 0; i < 2; i++) {
LISTBASE_FOREACH (LinkData *, el, BM_edgeloop_verts_get(estore_pair[i])) {
LinkData *el_next = BM_EDGELINK_NEXT(estore_pair[i], el);
if (el_next) {
if (el->data != el_next->data) {
BMEdge *e = BM_edge_exists(static_cast<BMVert *>(el->data),
static_cast<BMVert *>(el_next->data));
BMO_edge_flag_disable(bm, e, EDGE_OUT);
}
}
}
}
}
if (el_store_b_free) {
BM_edgeloop_free(el_store_b);
}
}
void bmo_bridge_loops_exec(BMesh *bm, BMOperator *op)
{
ListBase eloops = {nullptr};
/* merge-bridge support */
const bool use_pairs = BMO_slot_bool_get(op->slots_in, "use_pairs");
const bool use_merge = BMO_slot_bool_get(op->slots_in, "use_merge");
const float merge_factor = BMO_slot_float_get(op->slots_in, "merge_factor");
const bool use_cyclic = BMO_slot_bool_get(op->slots_in, "use_cyclic") && (use_merge == false);
const int twist_offset = BMO_slot_int_get(op->slots_in, "twist_offset");
int count;
bool changed = false;
BMO_slot_buffer_flag_enable(bm, op->slots_in, "edges", BM_EDGE, EDGE_MARK);
count = BM_mesh_edgeloops_find(bm, &eloops, bm_edge_test_cb, bm);
BM_mesh_edgeloops_calc_center(bm, &eloops);
if (count < 2) {
BMO_error_raise(bm, op, BMO_ERROR_CANCEL, "Select at least two edge loops");
goto cleanup;
}
if (use_pairs && (count % 2)) {
BMO_error_raise(bm, op, BMO_ERROR_CANCEL, "Select an even number of loops to bridge pairs");
goto cleanup;
}
if (use_merge) {
bool match = true;
const int eloop_len = BM_edgeloop_length_get(static_cast<BMEdgeLoopStore *>(eloops.first));
LISTBASE_FOREACH (LinkData *, el_store, &eloops) {
if (eloop_len != BM_edgeloop_length_get((BMEdgeLoopStore *)el_store)) {
match = false;
break;
}
}
if (!match) {
BMO_error_raise(bm, op, BMO_ERROR_CANCEL, "Selected loops must have equal edge counts");
goto cleanup;
}
}
if (count > 2) {
if (use_pairs) {
BM_mesh_edgeloops_calc_normal(bm, &eloops);
}
BM_mesh_edgeloops_calc_order(bm, &eloops, use_pairs);
}
LISTBASE_FOREACH (LinkData *, el_store, &eloops) {
LinkData *el_store_next = el_store->next;
if (el_store_next == nullptr) {
if (use_cyclic && (count > 2)) {
el_store_next = static_cast<LinkData *>(eloops.first);
}
else {
break;
}
}
bridge_loop_pair(bm,
(BMEdgeLoopStore *)el_store,
(BMEdgeLoopStore *)el_store_next,
use_merge,
merge_factor,
twist_offset);
if (use_pairs) {
el_store = el_store->next;
}
changed = true;
}
cleanup:
BM_mesh_edgeloops_free(&eloops);
if (changed) {
if (use_merge == false) {
BMO_slot_buffer_from_enabled_flag(bm, op, op->slots_out, "faces.out", BM_FACE, FACE_OUT);
BMO_slot_buffer_from_enabled_flag(bm, op, op->slots_out, "edges.out", BM_EDGE, EDGE_OUT);
}
}
}