synfig-studio/trunk/src/synfigapp/blineconvert.cpp

   1 /* === S Y N F I G ========================================================= */
   2 /*!     \file blineconvert.cpp
   3 **      \brief Template File
   4 **
   5 **      $Id$
   6 **
   7 **      \legal
   8 **      Copyright (c) 2002-2005 Robert B. Quattlebaum Jr., Adrian Bentley
   9 **
  10 **      This package is free software; you can redistribute it and/or
  11 **      modify it under the terms of the GNU General Public License as
  12 **      published by the Free Software Foundation; either version 2 of
  13 **      the License, or (at your option) any later version.
  14 **
  15 **      This package is distributed in the hope that it will be useful,
  16 **      but WITHOUT ANY WARRANTY; without even the implied warranty of
  17 **      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  18 **      General Public License for more details.
  19 **      \endlegal
  20 */
  21 /* ========================================================================= */
  22
  23 /* === H E A D E R S ======================================================= */
  24
  25 #ifdef USING_PCH
  26 #       include "pch.h"
  27 #else
  28 #ifdef HAVE_CONFIG_H
  29 #       include <config.h>
  30 #endif
  31
  32 #include "blineconvert.h"
  33 #include <vector>
  34 #include <ETL/gaussian>
  35 #include <ETL/hermite>
  36 #include <ETL/clock>
  37 #include <float.h>
  38 #include <algorithm>
  39 #include <synfig/general.h>
  40 #include <cassert>
  41
  42 #endif
  43
  44 /* === U S I N G =========================================================== */
  45
  46 using namespace std;
  47 using namespace etl;
  48 using namespace synfig;
  49
  50 /* === M A C R O S ========================================================= */
  51
  52 #define EPSILON         (1e-10)
  53
  54 /* === G L O B A L S ======================================================= */
  55
  56 /* === P R O C E D U R E S ================================================= */
  57
  58 /* === M E T H O D S ======================================================= */
  59
  60
  61 //Derivative Functions for numerical approximation
  62
  63 //bias == 0 will get F' at f3, bias < 0 will get F' at f1, and bias > 0 will get F' at f5
  64 template < class T >
  65 inline void FivePointdt(T &df, const T &f1, const T &f2, const T &f3, const T &f4, const T &f5, int bias)
  66 {
  67         if (bias == 0)                          // middle
  68                 df = (f1 - f2*8 + f4*8 - f5)*(1/12.0f);
  69         else if (bias < 0)                      // left
  70                 df = (-f1*25 + f2*48 - f3*36 + f4*16 - f5*3)*(1/12.0f);
  71         else                                            // right
  72                 df = (f1*3 - f2*16 + f3*36 - f4*48 + f5*25)*(1/12.0f);
  73 }
  74
  75 template < class T >
  76 inline void ThreePointdt(T &df, const T &f1, const T &f2, const T &f3, int bias)
  77 {
  78         if (bias == 0)                          // middle
  79                 df = (-f1 + f3)*(1/2.0f);
  80         else if (bias < 0)                      // left
  81                 df = (-f1*3 + f2*4 - f3)*(1/2.0f);
  82         else                                            // right
  83                 df = (f1 - f2*4 + f3*3)*(1/2.0f);
  84 }
  85
  86 // template < class T >
  87 // inline void ThreePointddt(T &df, const T &f1, const T &f2, const T &f3, int bias)
  88 // {
  89 //      // a 3 point approximation pretends to have constant acceleration,
  90 //      // so only one algorithm needed for left, middle, or right
  91 //      df = (f1 -f2*2 + f3)*(1/2.0f);
  92 // }
  93 //
  94 // // WARNING -- totally broken
  95 // template < class T >
  96 // inline void FivePointddt(T &df, const T &f1, const T &f2, const T &f3, int bias)
  97 // {
  98 //      if(bias == 0)
  99 //      {
 100 //              assert(0); // !?
 101 //              //middle
 102 //              //df = (- f1 + f2*16 - f3*30 +  f4*16 - f5)*(1/12.0f);
 103 //      }/*else if(bias < 0)
 104 //      {
 105 //              //left
 106 //              df = (f1*7 - f2*26*4 + f3*19*6 - f4*14*4 + f5*11)*(1/12.0f);
 107 //      }else
 108 //      {
 109 //              //right
 110 //              df = (f1*3 - f2*16 + f3*36 - f4*48 + f5*25)*(1/12.0f);
 111 //      }*/
 112 //      //side ones don't work, use 3 point
 113 // }
 114 //
 115 // //implement an arbitrary derivative
 116 // //dumb algorithm
 117 // template < class T >
 118 // void DerivativeApprox(T &df, const T f[], const Real t[], int npoints, int indexval)
 119 // {
 120 //      /*
 121 //      Lj(x) = PI_i!=j (x - xi) / PI_i!=j (xj - xi)
 122 //
 123 //      so Lj'(x) = SUM_k PI_i!=j|k (x - xi) / PI_i!=j (xj - xi)
 124 //      */
 125 //
 126 //      unsigned int i,j,k,i0,i1;
 127 //
 128 //      Real Lpj,mult,div,tj;
 129 //      Real tval = t[indexval];
 130 //
 131 //      //sum k
 132 //      for(j=0;j<npoints;++j)
 133 //      {
 134 //              Lpj = 0;
 135 //              div = 1;
 136 //              tj = t[j];
 137 //
 138 //              for(k=0;k<npoints;++k)
 139 //              {
 140 //                      if(k != j) //because there is no summand for k == j, since that term is missing from the original equation
 141 //                      {
 142 //                              //summation for k
 143 //                              for(i=0;i<npoints;++i)
 144 //                              {
 145 //                                      if(i != k)
 146 //                                      {
 147 //                                              mult *= tval - t[i];
 148 //                                      }
 149 //                              }
 150 //
 151 //                              Lpj += mult; //add into the summation
 152 //
 153 //                              //since the ks follow the exact pattern we need for the divisor (use that too)
 154 //                              div *= tj - t[k];
 155 //                      }
 156 //              }
 157 //
 158 //              //get the actual coefficient
 159 //              Lpj /= div;
 160 //
 161 //              //add it in to the equation
 162 //              df += f[j]*Lpj;
 163 //      }
 164 // }
 165
 166 //END numerical derivatives
 167
 168 // template < class T >
 169 // inline int sign(T f, T tol)
 170 // {
 171 //      if(f < -tol) return -1;
 172 //      if(f > tol) return 1;
 173 //      return 0;
 174 // }
 175
 176 void GetFirstDerivatives(const std::vector<synfig::Point> &f, unsigned int left, unsigned int right, char *out, unsigned int dfstride)
 177 {
 178         unsigned int current = left;
 179
 180         if(right - left < 2)
 181                 return;
 182         else if(right - left == 2)
 183         {
 184                 synfig::Vector v = f[left+1] - f[left];
 185
 186                 //set both to the one we want
 187                 *(synfig::Vector*)out = v;
 188                 out += dfstride;
 189                 *(synfig::Vector*)out = v;
 190                 out += dfstride;
 191         }
 192         else if(right - left < 6/*5*/) //should use 3 point
 193         {
 194                 //left then middle then right
 195                 ThreePointdt(*(synfig::Vector*)out,f[left+0], f[left+1], f[left+2], -1);
 196                 current++;
 197                 out += dfstride;
 198
 199                 for(;current < right-1; current++, out += dfstride)
 200                         ThreePointdt(*(synfig::Vector*)out,f[current-1], f[current], f[current+1], 0);
 201
 202                 ThreePointdt(*(synfig::Vector*)out,f[right-3], f[right-2], f[right-1], 1);
 203                 current++;
 204                 out += dfstride;
 205
 206         }else //can use 5 point
 207         {
 208                 //left 2 then middle bunch then right two
 209                 //may want to use 3 point for inner edge ones
 210
 211                 FivePointdt(*(synfig::Vector*)out,f[left+0], f[left+1], f[left+2], f[left+3], f[left+4], -2);
 212                 out += dfstride;
 213                 FivePointdt(*(synfig::Vector*)out,f[left+1], f[left+2], f[left+3], f[left+4], f[left+5], -1);
 214                 out += dfstride;
 215                 current += 2;
 216
 217                 for(;current < right-2; current++, out += dfstride)
 218                         FivePointdt(*(synfig::Vector*)out,f[current-2], f[current-1], f[current], f[current+1], f[current+2], 0);
 219
 220                 FivePointdt(*(synfig::Vector*)out,f[right-6], f[right-5], f[right-4], f[right-3], f[right-2], 2);
 221                 out += dfstride;
 222                 FivePointdt(*(synfig::Vector*)out,f[right-5], f[right-4], f[right-3], f[right-2], f[right-1], 1);
 223                 out += dfstride;
 224                 current += 2;
 225         }
 226 }
 227
 228 void GetSimpleDerivatives(const std::vector<synfig::Point> &f, int left, int right,
 229                                                         std::vector<synfig::Point> &df, int outleft,
 230                                                         const std::vector<synfig::Real> &/*di*/)
 231 {
 232         int i1,i2,i;
 233         int offset = 2; //df = 1/2 (f[i+o]-f[i-o])
 234
 235         assert((int)df.size() >= right-left+outleft); //must be big enough
 236
 237         for(i = left; i < right; ++i)
 238         {
 239                 //right now indices (figure out distance later)
 240                 i1 = std::max(left,i-offset);
 241                 i2 = std::max(left,i+offset);
 242
 243                 df[outleft++] = (f[i2] - f[i1])*0.5f;
 244         }
 245 }
 246
 247 //get the curve error from the double sample list of work points (hopefully that's enough)
 248 Real CurveError(const synfig::Point *pts, unsigned int n, std::vector<synfig::Point> &work, int left, int right)
 249 {
 250         if(right-left < 2) return -1;
 251
 252         int i,j;
 253
 254         //get distances to each point
 255         Real d,dtemp,dsum;
 256         //synfig::Vector v,vt;
 257         //synfig::Point p1,p2;
 258         synfig::Point pi;
 259         std::vector<synfig::Point>::const_iterator it;//,end = work.begin()+right;
 260
 261         //unsigned int size = work.size();
 262
 263         //for each line, get distance
 264         d = 0; //starts at 0
 265         for(i = 0; i < (int)n; ++i)
 266         {
 267                 pi = pts[i];
 268
 269                 dsum = FLT_MAX;
 270
 271                 it = work.begin()+left;
 272                 //p2 = *it++; //put it at left+1
 273                 for(j = left/*+1*/; j < right; ++j,++it)
 274                 {
 275                         /*p1 = p2;
 276                         p2 = *it;
 277
 278                         v = p2 - p1;
 279                         vt = pi - p1;
 280
 281                         dtemp = v.mag_squared() > 1e-12 ? (vt*v)/v.mag_squared() : 0; //get the projected time value for the current line
 282
 283                         //get distance to line segment with the time value clamped 0-1
 284                         if(dtemp >= 1)  //use p+v
 285                         {
 286                                 vt += v; //makes it pp - (p+v)
 287                         }else if(dtemp > 0)     //use vt-proj
 288                         {
 289                                 vt -= v*dtemp; // vt - proj_v(vt)       //must normalize the projection vector to work
 290                         }
 291
 292                         //else use p
 293                         dtemp = vt.mag_squared();*/
 294
 295                         dtemp = (pi - *it).mag_squared();
 296                         if(dtemp < dsum)
 297                                 dsum = dtemp;
 298                 }
 299
 300                 //accumulate the points' min distance from the curve
 301                 d += sqrt(dsum);
 302         }
 303
 304         return d;
 305 }
 306
 307 typedef synfigapp::BLineConverter::cpindex cpindex;
 308
 309 //has the index data and the tangent scale data (relevant as it may be)
 310 int tessellate_curves(const std::vector<cpindex> &inds, const std::vector<Point> &f, const std::vector<synfig::Vector> &df, std::vector<Point> &work)
 311 {
 312         if(inds.size() < 2)
 313                 return 0;
 314
 315         etl::hermite<Point>     curve;
 316         int ntess = 0;
 317
 318         std::vector<cpindex>::const_iterator j = inds.begin(),j2, end = inds.end();
 319
 320         unsigned int ibase = inds[0].curind;
 321
 322         j2 = j++;
 323         for(; j != end; j2 = j++)
 324         {
 325                 //if this curve has invalid error (in j) then retessellate its work points (requires reparametrization, etc.)
 326                 if(j->error < 0)
 327                 {
 328                         //get the stepsize etc. for the number of points in here
 329                         unsigned int n = j->curind - j2->curind + 1; //thats the number of points in the span
 330                         unsigned int k, kend, i0, i3;
 331                         //so reset the right chunk
 332
 333                         Real t, dt = 1/(Real)(n*2-2); //assuming that they own only n points
 334
 335                         //start at first intermediate
 336                         t = 0;
 337
 338                         i0 = j2->curind; i3 = j->curind;
 339                         k = (i0-ibase)*2; //start on first intermediary point (2x+1)
 340                         kend = (i3-ibase)*2; //last point to set (not intermediary)
 341
 342                         //build hermite curve, it's easier
 343                         curve.p1() = f[i0];
 344                         curve.p2() = f[i3];
 345                         curve.t1() = df[i0-ibase] * (df[i0-ibase].mag_squared() > 1e-4
 346                                                                                  ? j2->tangentscale/df[i0-ibase].mag()
 347                                                                                  : j2->tangentscale);
 348                         curve.t2() = df[i3-ibase] * (df[i3-ibase].mag_squared() > 1e-4
 349                                                                                  ? j->tangentscale/df[i3-ibase].mag()
 350                                                                                  : j->tangentscale);
 351                         curve.sync();
 352
 353                         //MUST include the end point (since we are ignoring left one)
 354                         for(; k < kend; ++k, t += dt)
 355                         {
 356                                 work[k] = curve(t);
 357                         }
 358
 359                         work[k] = curve(1); //k == kend, t == 1 -> c(t) == p2
 360                         ++ntess;
 361                 }
 362         }
 363
 364         return ntess;
 365 }
 366
 367 synfigapp::BLineConverter::BLineConverter()
 368 {
 369         pixelwidth = 1;
 370         smoothness = 0.70f;
 371         width = 0;
 372 };
 373
 374 void
 375 synfigapp::BLineConverter::clear()
 376 {
 377         f.clear();
 378         f_w.clear();
 379         ftemp.clear();
 380         df.clear();
 381         cvt.clear();
 382         brk.clear();
 383         di.clear();
 384         d_i.clear();
 385         work.clear();
 386         curind.clear();
 387 }
 388
 389 void
 390 synfigapp::BLineConverter::operator () (std::list<synfig::BLinePoint> &out, const std::list<synfig::Point> &in,const std::list<synfig::Real> &in_w)
 391 {
 392         //Profiling information
 393         /*etl::clock::value_type initialprocess=0, curveval=0, breakeval=0, disteval=0;
 394         etl::clock::value_type preproceval=0, tesseval=0, erroreval=0, spliteval=0;
 395         unsigned int                    numpre=0, numtess=0, numerror=0, numsplit=0;
 396         etl::clock_realtime timer,total;*/
 397
 398         //total.reset();
 399         if(in.size()<=1)
 400                 return;
 401
 402         clear();
 403
 404         //removing digitization error harder than expected
 405
 406         //intended to fix little pen errors caused by the way people draw (tiny juts in opposite direction)
 407         //Different solutions
 408         //      Average at both end points (will probably eliminate many points at each end of the samples)
 409         //      Average after the break points are found (weird points would still affect the curve)
 410         //      Just always get rid of breaks at the beginning and end if they are a certain distance apart
 411         //              This is will be current approach so all we do now is try to remove duplicate points
 412
 413         //remove duplicate points - very bad for fitting
 414
 415         //timer.reset();
 416
 417         {
 418                 std::list<synfig::Point>::const_iterator i = in.begin(), end = in.end();
 419                 std::list<synfig::Real>::const_iterator iw = in_w.begin();
 420                 synfig::Point   c;
 421
 422                 if(in.size() == in_w.size())
 423                 {
 424                         for(;i != end; ++i,++iw)
 425                                 if(*i != c)             // eliminate duplicate points
 426                                 {
 427                                         f.push_back(c = *i);
 428                                         f_w.push_back(*iw);
 429                                 }
 430                 }
 431                 else
 432                 {
 433                         for(;i != end; ++i)
 434                                 if(*i != c)             // eliminate duplicate points
 435                                         f.push_back(c = *i);
 436                 }
 437         }
 438         //initialprocess = timer();
 439
 440         if(f.size()<=6)
 441                 return;
 442
 443         //get curvature information
 444         //timer.reset();
 445
 446         {
 447                 int i,i0,i1;
 448                 synfig::Vector v1,v2;
 449
 450                 cvt.resize(f.size());
 451
 452                 cvt.front() = 1;
 453                 cvt.back() = 1;
 454
 455                 for(i = 1; i < (int)f.size()-1; ++i)
 456                 {
 457                         i0 = std::max(0,i - 2);
 458                         i1 = std::min((int)(f.size()-1),i + 2);
 459
 460                         v1 = f[i] - f[i0];
 461                         v2 = f[i1] - f[i];
 462
 463                         cvt[i] = (v1*v2)/(v1.mag()*v2.mag());
 464                 }
 465         }
 466
 467         //curveval = timer();
 468         //synfig::info("calculated curvature");
 469
 470         //find corner points and interpolate inside those
 471         //timer.reset();
 472         {
 473                 //break at sharp derivative points
 474                 //TODO tolerance should be set based upon digitization resolution (length dependent index selection)
 475                 Real    tol = 0;                //break tolerance, for the cosine of the change in angle (really high curvature or something)
 476                 Real    fixdistsq = 4*width*width; //the distance to ignore breaks at the end points (for fixing stuff)
 477                 unsigned int i = 0;
 478
 479                 int             maxi = -1, last=0;
 480                 Real    minc = 1;
 481
 482                 brk.push_back(0);
 483
 484                 for(i = 1; i < cvt.size()-1; ++i)
 485                 {
 486                         //insert if too sharp (we need to break the tangents to insert onto the break list)
 487
 488                         if(cvt[i] < tol)
 489                         {
 490                                 if(cvt[i] < minc)
 491                                 {
 492                                         minc = cvt[i];
 493                                         maxi = i;
 494                                 }
 495                         }
 496                         else if(maxi >= 0)
 497                         {
 498                                 if(maxi >= last + 8)
 499                                 {
 500                                         //synfig::info("break: %d-%d",maxi+1,cvt.size());
 501                                         brk.push_back(maxi);
 502                                         last = maxi;
 503                                 }
 504                                 maxi = -1;
 505                                 minc = 1;
 506                         }
 507                 }
 508
 509                 brk.push_back(i);
 510
 511                 //postprocess for breaks too close to each other
 512                 Real d = 0;
 513                 Point p = f[brk.front()];
 514
 515                 //first set
 516                 for(i = 1; i < brk.size()-1; ++i) //do not want to include end point...
 517                 {
 518                         d = (f[brk[i]] - p).mag_squared();
 519                         if(d > fixdistsq) break; //don't want to group breaks if we ever get over the dist...
 520                 }
 521                 //want to erase all points before...
 522                 if(i != 1)
 523                         brk.erase(brk.begin(),brk.begin()+i-1);
 524
 525                 //end set
 526                 p = f[brk.back()];
 527                 for(i = brk.size()-2; i > 0; --i) //start at one in from the end
 528                 {
 529                         d = (f[brk[i]] - p).mag_squared();
 530                         if(d > fixdistsq) break; //don't want to group breaks if we ever get over the dist
 531                 }
 532                 if(i != brk.size()-2)
 533                         brk.erase(brk.begin()+i+2,brk.end()); //erase all points that we found... found none if i has not advanced
 534                 //must not include the one we ended up on
 535         }
 536         //breakeval = timer();
 537         //synfig::info("found break points: %d",brk.size());
 538
 539         //get the distance calculation of the entire curve (for tangent scaling)
 540
 541         //timer.reset();
 542         {
 543                 synfig::Point p1,p2;
 544
 545                 p1=p2=f[0];
 546
 547                 di.resize(f.size()); d_i.resize(f.size());
 548                 Real d = 0;
 549                 for(unsigned int i = 0; i < f.size();)
 550                 {
 551                         d += (d_i[i] = (p2-p1).mag());
 552                         di[i] = d;
 553
 554                         p1=p2;
 555                         p2=f[++i];
 556                 }
 557         }
 558         //disteval = timer();
 559         //synfig::info("calculated distance");
 560
 561         //now break at every point - calculate new derivatives each time
 562
 563         //TODO
 564         //must be sure that the break points are 3 or more apart
 565         //then must also store the breaks which are not smooth, etc.
 566         //and figure out tangents between there
 567
 568         //for each pair of break points (as long as they are far enough apart) recursively subdivide stuff
 569         //ignore the detected intermediate points
 570         {
 571                 unsigned int i0=0,i3=0,is=0;
 572                 int i=0,j=0;
 573
 574                 bool done = false;
 575
 576                 Real errortol = smoothness*pixelwidth; //???? what the hell should this value be
 577
 578                 BLinePoint a;
 579                 synfig::Vector v;
 580
 581                 //intemp = f; //don't want to smooth out the corners
 582
 583                 bool breaktan = false, setwidth;
 584                 a.set_split_tangent_flag(false);
 585                 //a.set_width(width);
 586                 a.set_width(1.0f);
 587
 588                 setwidth = (f.size() == f_w.size());
 589
 590                 for(j = 0; j < (int)brk.size() - 1; ++j)
 591                 {
 592                         //for b[j] to b[j+1] subdivide and stuff
 593                         i0 = brk[j];
 594                         i3 = brk[j+1];
 595
 596                         unsigned int size = i3-i0+1; //must include the end points
 597
 598                         //new derivatives
 599                         //timer.reset();
 600                         ftemp.assign(f.begin()+i0, f.begin()+i3+1);
 601                         for(i=0;i<20;++i)
 602                                 gaussian_blur_3(ftemp.begin(),ftemp.end(),false);
 603
 604                         df.resize(size);
 605
 606                         // Wondering whether the modification of the df vector
 607                         // using a char* pointer and pointer arithmetric was safe,
 608                         // I looked it up...
 609                         //
 610                         // http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2369.pdf tells me:
 611                         //
 612                         //      23.2.5  Class template vector [vector]
 613                         //
 614                         //      [...] The elements of a vector are stored contiguously,
 615                         //      meaning that if v is a vector<T,Allocator> where T is
 616                         //      some type other than bool, then it obeys the identity
 617                         //      &v[n] == &v[0] + n for all 0 <= n < v.size().
 618                         //
 619                         GetFirstDerivatives(ftemp,0,size,(char*)&df[0],sizeof(df[0]));
 620
 621                         //GetSimpleDerivatives(ftemp,0,size,df,0,di);
 622                         //< don't have to worry about indexing stuff as it is all being taken care of right now
 623                         //preproceval += timer();
 624                         //numpre++;
 625
 626                         work.resize(size*2-1); //guarantee that all points will be tessellated correctly (one point in between every 2 adjacent points)
 627
 628                         //if size of work is size*2-1, the step size should be 1/(size*2 - 2)
 629                         //Real step = 1/(Real)(size*2 - 1);
 630
 631                         //start off with break points as indices
 632                         curind.clear();
 633                         curind.push_back(cpindex(i0,di[i3]-di[i0],0)); //0 error because no curve on the left
 634                         curind.push_back(cpindex(i3,di[i3]-di[i0],-1)); //error needs to be reevaluated
 635                         done = false; //we want to loop
 636
 637                         unsigned int dcount = 0;
 638
 639                         //while there are still enough points between us, and the error is too high subdivide (and invalidate neighbors that share tangents)
 640                         while(!done)
 641                         {
 642                                 //tessellate all curves with invalid error values
 643                                 work[0] = f[i0];
 644
 645                                 //timer.reset();
 646                                 /*numtess += */tessellate_curves(curind,f,df,work);
 647                                 //tesseval += timer();
 648
 649                                 //now get all error values
 650                                 //timer.reset();
 651                                 for(i = 1; i < (int)curind.size(); ++i)
 652                                 {
 653                                         if(curind[i].error < 0) //must have been retessellated, so now recalculate error value
 654                                         {
 655                                                 //evaluate error from points (starting at current index)
 656                                                 int size = curind[i].curind - curind[i-1].curind + 1;
 657                                                 curind[i].error = CurveError(&f[curind[i-1].curind], size,
 658                                                                                                          work,(curind[i-1].curind - i0)*2,(curind[i].curind - i0)*2+1);
 659
 660                                                 /*if(curind[i].error > 1.0e5)
 661                                                 {
 662                                                         synfig::info("Holy crap %d-%d error %f",curind[i-1].curind,curind[i].curind,curind[i].error);
 663                                                         curind[i].error = -1;
 664                                                         numtess += tessellate_curves(curind,f,df,work);
 665                                                         curind[i].error = CurveError(&f[curind[i-1].curind], size,
 666                                                                                                          work,0,work.size());//(curind[i-1].curind - i0)*2,(curind[i].curind - i0)*2+1);
 667                                                 }*/
 668                                                 //numerror++;
 669                                         }
 670                                 }
 671                                 //erroreval += timer();
 672
 673                                 //assume we're done
 674                                 done = true;
 675
 676                                 //check each error to see if it's too big, if so, then subdivide etc.
 677                                 int indsize = (int)curind.size();
 678                                 Real maxrelerror = 0;
 679                                 int maxi = -1;//, numpoints;
 680
 681                                 //timer.reset();
 682                                 //get the maximum error and split there
 683                                 for(i = 1; i < indsize; ++i)
 684                                 {
 685                                         //numpoints = curind[i].curind - curind[i-1].curind + 1;
 686
 687                                         if(curind[i].error > maxrelerror && curind[i].curind - curind[i-1].curind > 2) //only accept if it's valid
 688                                         {
 689                                                 maxrelerror = curind[i].error;
 690                                                 maxi = i;
 691                                         }
 692                                 }
 693
 694                                 //split if error is too great
 695                                 if(maxrelerror > errortol)
 696                                 {
 697                                         //add one to the left etc
 698                                         unsigned int    ibase = curind[maxi-1].curind, itop = curind[maxi].curind,
 699                                                                         ibreak = (ibase + itop)/2;
 700                                         Real scale, scale2;
 701
 702                                         assert(ibreak < f.size());
 703
 704                                         //synfig::info("Split %d -%d- %d, error: %f", ibase,ibreak,itop,maxrelerror);
 705
 706                                         if(ibase != itop)
 707                                         {
 708                                                 //invalidate current error of the changed tangents and add an extra segment
 709                                                 //enforce minimum tangents property
 710                                                 curind[maxi].error = -1;
 711                                                 curind[maxi-1].error = -1;
 712                                                 if(maxi+1 < indsize) curind[maxi+1].error = -1; //if there is a curve segment beyond this it will be effected as well
 713
 714                                                 scale = di[itop] - di[ibreak];
 715                                                 scale2 = maxi+1 < indsize ? di[curind[maxi+1].curind] - di[itop] : scale; //to the right valid?
 716                                                 curind[maxi].tangentscale = std::min(scale, scale2);
 717
 718                                                 scale = di[ibreak] - di[ibase];
 719                                                 scale2 = maxi >= 2 ? di[ibase] - di[curind[maxi-2].curind] : scale; // to the left valid -2 ?
 720                                                 curind[maxi-1].tangentscale = std::min(scale, scale2);
 721
 722                                                 scale = std::min(di[ibreak] - di[ibase], di[itop] - di[ibreak]);
 723
 724                                                 curind.insert(curind.begin()+maxi,cpindex(ibreak, scale, -1));
 725                                                 //curind.push_back(cpindex(ibreak, scale, -1));
 726                                                 //std::sort(curind.begin(), curind.end());
 727
 728                                                 done = false;
 729                                                 //numsplit++;
 730                                         }
 731                                 }
 732                                 //spliteval += timer();
 733
 734                                 dcount++;
 735                         }
 736
 737                         //insert the last point too (just set tangent for now
 738                         is = curind[0].curind;
 739
 740                         //first point inherits current tangent status
 741                         v = df[is - i0];
 742                         if(v.mag_squared() > EPSILON)
 743                                 v *= (curind[0].tangentscale/v.mag());
 744
 745                         if(!breaktan)
 746                                 a.set_tangent(v);
 747                         else a.set_tangent2(v);
 748
 749                         a.set_vertex(f[is]);
 750                         if(setwidth)a.set_width(f_w[is]);
 751
 752                         out.push_back(a);
 753                         a.set_split_tangent_flag(false); //won't need to break anymore
 754                         breaktan = false;
 755
 756                         for(i = 1; i < (int)curind.size()-1; ++i)
 757                         {
 758                                 is = curind[i].curind;
 759
 760                                 //first point inherits current tangent status
 761                                 v = df[is-i0];
 762                                 if(v.mag_squared() > EPSILON)
 763                                         v *= (curind[i].tangentscale/v.mag());
 764
 765                                 a.set_tangent(v); // always inside, so guaranteed to be smooth
 766                                 a.set_vertex(f[is]);
 767                                 if(setwidth)a.set_width(f_w[is]);
 768
 769                                 out.push_back(a);
 770                         }
 771
 772                         //set the last point's data
 773                         is = curind.back().curind; //should already be this
 774
 775                         v = df[is-i0];
 776                         if(v.mag_squared() > EPSILON)
 777                                 v *= (curind.back().tangentscale/v.mag());
 778
 779                         a.set_tangent1(v);
 780                         a.set_split_tangent_flag(true);
 781                         breaktan = true;
 782
 783                         //will get the vertex and tangent 2 from next round
 784                 }
 785
 786                 a.set_vertex(f[i3]);
 787                 a.set_split_tangent_flag(false);
 788                 if(setwidth)
 789                         a.set_width(f_w[i3]);
 790                 out.push_back(a);
 791
 792                 /*etl::clock::value_type totaltime = total(),
 793                                                            misctime = totaltime - initialprocess - curveval - breakeval - disteval
 794                                                                           - preproceval - tesseval - erroreval - spliteval;
 795
 796                 synfig::info(
 797                         "Curve Convert Profile:\n"
 798                         "\tInitial Preprocess:    %f\n"
 799                         "\tCurvature Calculation: %f\n"
 800                         "\tBreak Calculation:     %f\n"
 801                         "\tDistance Calculation:  %f\n"
 802                         "  Algorithm: (numtimes,totaltime)\n"
 803                         "\tPreprocess step:      (%d,%f)\n"
 804                         "\tTessellation step:    (%d,%f)\n"
 805                         "\tError step:           (%d,%f)\n"
 806                         "\tSplit step:           (%d,%f)\n"
 807                         "  Num Input: %d, Num Output: %d\n"
 808                         "  Total time: %f, Misc time: %f\n",
 809                         initialprocess, curveval,breakeval,disteval,
 810                         numpre,preproceval,numtess,tesseval,numerror,erroreval,numsplit,spliteval,
 811                         in.size(),out.size(),
 812                         totaltime,misctime);*/
 813
 814                 return;
 815         }
 816 }
 817
 818 void synfigapp::BLineConverter::EnforceMinWidth(std::list<synfig::BLinePoint> &bline, synfig::Real min_pressure)
 819 {
 820         std::list<synfig::BLinePoint>::iterator i = bline.begin(),
 821                                                                                         end = bline.end();
 822
 823         for(i = bline.begin(); i != end; ++i)
 824                 if(i->get_width() < min_pressure)
 825                         i->set_width(min_pressure);
 826 }