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
  43
  44 #endif
  45
  46 /* === U S I N G =========================================================== */
  47
  48 using namespace std;
  49 using namespace etl;
  50 using namespace synfig;
  51
  52 /* === M A C R O S ========================================================= */
  53
  54 #define EPSILON         (1e-10)
  55
  56 /* === G L O B A L S ======================================================= */
  57
  58 /* === P R O C E D U R E S ================================================= */
  59
  60 /* === M E T H O D S ======================================================= */
  61
  62
  63 //Derivative Functions for numerical approximation
  64
  65 //bias == 0 will get F' at f3, bias < 0 will get F' at f1, and bias > 0 will get F' at f5
  66 template < class T >
  67 inline void FivePointdt(T &df, const T &f1, const T &f2, const T &f3, const T &f4, const T &f5, int bias)
  68 {
  69         if(bias == 0)
  70         {
  71                 //middle
  72                 df = (f1 - f2*8 + f4*8 - f5)*(1/12.0f);
  73         }else if(bias < 0)
  74         {
  75                 //left
  76                 df = (-f1*25 + f2*48 - f3*36 + f4*16 - f5*3)*(1/12.0f);
  77         }else
  78         {
  79                 //right
  80                 df = (f1*3 - f2*16 + f3*36 - f4*48 + f5*25)*(1/12.0f);
  81         }
  82 }
  83
  84 template < class T >
  85 inline void ThreePointdt(T &df, const T &f1, const T &f2, const T &f3, int bias)
  86 {
  87         if(bias == 0)
  88         {
  89                 //middle
  90                 df = (-f1 + f3)*(1/2.0f);
  91         }else if(bias < 0)
  92         {
  93                 //left
  94                 df = (-f1*3 + f2*4 - f3)*(1/2.0f);
  95         }else
  96         {
  97                 //right
  98                 df = (f1 - f2*4 + f3*3)*(1/2.0f);
  99         }
 100 }
 101
 102 template < class T >
 103 inline void ThreePointddt(T &df, const T &f1, const T &f2, const T &f3, int bias)
 104 {
 105         //a 3 point approximation pretends to have constant acceleration, so only one algorithm needed for left, middle, or right
 106         df = (f1 -f2*2 + f3)*(1/2.0f);
 107 }
 108
 109 // WARNING -- totaly broken
 110 template < class T >
 111 inline void FivePointddt(T &df, const T &f1, const T &f2, const T &f3, int bias)
 112 {
 113         if(bias == 0)
 114         {
 115                 assert(0); // !?
 116                 //middle
 117                 //df = (- f1 + f2*16 - f3*30 +  f4*16 - f5)*(1/12.0f);
 118         }/*else if(bias < 0)
 119         {
 120                 //left
 121                 df = (f1*7 - f2*26*4 + f3*19*6 - f4*14*4 + f5*11)*(1/12.0f);
 122         }else
 123         {
 124                 //right
 125                 df = (f1*3 - f2*16 + f3*36 - f4*48 + f5*25)*(1/12.0f);
 126         }*/
 127         //side ones don't work, use 3 point
 128 }
 129
 130 //implement an arbitrary derivative
 131 //dumb algorithm
 132 template < class T >
 133 void DerivativeApprox(T &df, const T f[], const Real t[], int npoints, int indexval)
 134 {
 135         /*
 136         Lj(x) = PI_i!=j (x - xi) / PI_i!=j (xj - xi)
 137
 138         so Lj'(x) = SUM_k PI_i!=j|k (x - xi) / PI_i!=j (xj - xi)
 139         */
 140
 141         unsigned int i,j,k,i0,i1;
 142
 143         Real Lpj,mult,div,tj;
 144         Real tval = t[indexval];
 145
 146         //sum k
 147         for(j=0;j<npoints;++j)
 148         {
 149                 Lpj = 0;
 150                 div = 1;
 151                 tj = t[j];
 152
 153                 for(k=0;k<npoints;++k)
 154                 {
 155                         if(k != j) //because there is no summand for k == j, since that term is missing from the original equation
 156                         {
 157                                 //summation for k
 158                                 for(i=0;i<npoints;++i)
 159                                 {
 160                                         if(i != k)
 161                                         {
 162                                                 mult *= tval - t[i];
 163                                         }
 164                                 }
 165
 166                                 Lpj += mult; //add into the summation
 167
 168                                 //since the ks follow the exact patern we need for the divisor (use that too)
 169                                 div *= tj - t[k];
 170                         }
 171                 }
 172
 173                 //get the actual coefficient
 174                 Lpj /= div;
 175
 176                 //add it in to the equation
 177                 df += f[j]*Lpj;
 178         }
 179 }
 180
 181 //END numerical derivatives
 182
 183 template < class T >
 184 inline int sign(T f, T tol)
 185 {
 186         if(f < -tol) return -1;
 187         if(f > tol) return 1;
 188         return 0;
 189 }
 190
 191 void GetFirstDerivatives(const std::vector<synfig::Point> &f, unsigned int left, unsigned int right, char *out, unsigned int dfstride)
 192 {
 193         unsigned int current = left;
 194
 195         if(right - left < 2)
 196                 return;
 197         else if(right - left < 3)
 198         {
 199                 synfig::Vector v = f[left+1] - f[left];
 200
 201                 //set both to the one we want
 202                 *(synfig::Vector*)out = v;
 203                 out += dfstride;
 204                 *(synfig::Vector*)out = v;
 205                 out += dfstride;
 206         }
 207         else if(right - left < 6/*5*/) //should use 3 point
 208         {
 209                 //left then middle then right
 210                 ThreePointdt(*(synfig::Vector*)out,f[left+0], f[left+1], f[left+2], -1);
 211                 current += 1;
 212                 out += dfstride;
 213
 214                 for(;current < right-1; current++, out += dfstride)
 215                 {
 216                         ThreePointdt(*(synfig::Vector*)out,f[current-1], f[current], f[current+1], 0);
 217                 }
 218
 219                 ThreePointdt(*(synfig::Vector*)out,f[right-3], f[right-2], f[right-1], 1);
 220                 current++;
 221                 out += dfstride;
 222
 223         }else //can use 5 point
 224         {
 225                 //left 2 then middle bunch then right two
 226                 //may want to use 3 point for inner edge ones
 227
 228                 FivePointdt(*(synfig::Vector*)out,f[left+0], f[left+1], f[left+2], f[left+3], f[left+4], -2);
 229                 out += dfstride;
 230                 FivePointdt(*(synfig::Vector*)out,f[left+1], f[left+2], f[left+3], f[left+4], f[left+5], -1);
 231                 out += dfstride;
 232                 current += 2;
 233
 234                 for(;current < right-2; current++, out += dfstride)
 235                 {
 236                         FivePointdt(*(synfig::Vector*)out,f[current-2], f[current-1], f[current], f[current+1], f[current+2], 0);
 237                 }
 238
 239                 FivePointdt(*(synfig::Vector*)out,f[right-5], f[right-4], f[right-3], f[right-2], f[right-1], 1);
 240                 out += dfstride;
 241                 FivePointdt(*(synfig::Vector*)out,f[right-6], f[right-5], f[right-4], f[right-3], f[right-2], 2);
 242                 out += dfstride;
 243                 current += 2;
 244         }
 245 }
 246
 247 void GetSimpleDerivatives(const std::vector<synfig::Point> &f, int left, int right,
 248                                                         std::vector<synfig::Point> &df, int outleft,
 249                                                         const std::vector<synfig::Real> &di)
 250 {
 251         int i1,i2,i;
 252         int offset = 2; //df = 1/2 (f[i+o]-f[i-o])
 253
 254         assert((int)df.size() >= right-left+outleft); //must be big enough
 255
 256         for(i = left; i < right; ++i)
 257         {
 258                 //right now indices (figure out distance later)
 259                 i1 = std::max(left,i-offset);
 260                 i2 = std::max(left,i+offset);
 261
 262                 df[outleft++] = (f[i2] - f[i1])*0.5f;
 263         }
 264 }
 265
 266 //get the curve error from the double sample list of work points (hopefully that's enough)
 267 Real CurveError(const synfig::Point *pts, unsigned int n, std::vector<synfig::Point> &work, int left, int right)
 268 {
 269         if(right-left < 2) return -1;
 270
 271         int i,j;
 272
 273         //get distances to each point
 274         Real d,dtemp,dsum;
 275         //synfig::Vector v,vt;
 276         //synfig::Point p1,p2;
 277         synfig::Point pi;
 278         std::vector<synfig::Point>::const_iterator it;//,end = work.begin()+right;
 279
 280         //unsigned int size = work.size();
 281
 282         //for each line, get distance
 283         d = 0; //starts at 0
 284         for(i = 0; i < (int)n; ++i)
 285         {
 286                 pi = pts[i];
 287
 288                 dsum = FLT_MAX;
 289
 290                 it = work.begin()+left;
 291                 //p2 = *it++; //put it at left+1
 292                 for(j = left/*+1*/; j < right; ++j,++it)
 293                 {
 294                         /*p1 = p2;
 295                         p2 = *it;
 296
 297                         v = p2 - p1;
 298                         vt = pi - p1;
 299
 300                         dtemp = v.mag_squared() > 1e-12 ? (vt*v)/v.mag_squared() : 0; //get the projected time value for the current line
 301
 302                         //get distance to line segment with the time value clamped 0-1
 303                         if(dtemp >= 1)  //use p+v
 304                         {
 305                                 vt += v; //makes it pp - (p+v)
 306                         }else if(dtemp > 0)     //use vt-proj
 307                         {
 308                                 vt -= v*dtemp; // vt - proj_v(vt)       //must normalize the projection vector to work
 309                         }
 310
 311                         //else use p
 312                         dtemp = vt.mag_squared();*/
 313
 314                         dtemp = (pi - *it).mag_squared();
 315                         if(dtemp < dsum)
 316                                 dsum = dtemp;
 317                 }
 318
 319                 //accumulate the points' min distance from the curve
 320                 d += sqrt(dsum);
 321         }
 322
 323         return d;
 324 }
 325
 326 typedef synfigapp::BLineConverter::cpindex cpindex;
 327
 328 //has the index data and the tangent scale data (relevant as it may be)
 329 int tesselate_curves(const std::vector<cpindex> &inds, const std::vector<Point> &f, const std::vector<synfig::Vector> &df, std::vector<Point> &work)
 330 {
 331         if(inds.size() < 2)
 332                 return 0;
 333
 334         etl::hermite<Point>     curve;
 335         int ntess = 0;
 336
 337         std::vector<cpindex>::const_iterator j = inds.begin(),j2, end = inds.end();
 338
 339         unsigned int ibase = inds[0].curind;
 340
 341         j2 = j++;
 342         for(; j != end; j2 = j++)
 343         {
 344                 //if this curve has invalid error (in j) then retesselate it's work points (requires reparametrization, etc.)
 345                 if(j->error < 0)
 346                 {
 347                         //get the stepsize etc. for the number of points in here
 348                         unsigned int n = j->curind - j2->curind + 1; //thats the number of points in the span
 349                         unsigned int k, kend, i0, i3;
 350                         //so reset the right chunk
 351
 352                         Real t, dt = 1/(Real)(n*2-2); //assuming that they own only n points
 353
 354                         //start at first intermediate
 355                         t = 0;
 356
 357                         i0 = j2->curind; i3 = j->curind;
 358                         k = (i0-ibase)*2; //start on first intermediary point (2x+1)
 359                         kend = (i3-ibase)*2; //last point to set (not intermediary)
 360
 361                         //build hermite curve, it's easier
 362                         curve.p1() = f[i0];
 363                         curve.p2() = f[i3];
 364                         curve.t1() = df[i0]*(df[i0].mag_squared() > 1e-4 ? j2->tangentscale/df[i0].mag() : j2->tangentscale);
 365                         curve.t2() = df[i3]*(df[i3].mag_squared() > 1e-4 ? j->tangentscale/df[i3].mag() : j->tangentscale);
 366                         curve.sync();
 367
 368                         //MUST include the end point (since we are ignoring left one)
 369                         for(; k < kend; ++k, t += dt)
 370                         {
 371                                 work[k] = curve(t);
 372                         }
 373
 374                         work[k] = curve(1); //k == kend, t == 1 -> c(t) == p2
 375                         ++ntess;
 376                 }
 377         }
 378
 379         return ntess;
 380 }
 381
 382 synfigapp::BLineConverter::BLineConverter()
 383 {
 384         pixelwidth = 1;
 385         smoothness = 0.70f;
 386         width = 0;
 387 };
 388
 389 void
 390 synfigapp::BLineConverter::clear()
 391 {
 392         f.clear();
 393         f_w.clear();
 394         ftemp.clear();
 395         df.clear();
 396         cvt.clear();
 397         brk.clear();
 398         di.clear();
 399         d_i.clear();
 400         work.clear();
 401         curind.clear();
 402 }
 403
 404 void
 405 synfigapp::BLineConverter::operator () (std::list<synfig::BLinePoint> &out, const std::list<synfig::Point> &in,const std::list<synfig::Real> &in_w)
 406 {
 407         //Profiling information
 408         /*etl::clock::value_type initialprocess=0, curveval=0, breakeval=0, disteval=0;
 409         etl::clock::value_type preproceval=0, tesseval=0, erroreval=0, spliteval=0;
 410         unsigned int                    numpre=0, numtess=0, numerror=0, numsplit=0;
 411         etl::clock_realtime timer,total;*/
 412
 413         //total.reset();
 414         if(in.size()<=1)
 415                 return;
 416
 417         clear();
 418
 419         //removing digitization error harder than expected
 420
 421         //intended to fix little pen errors caused by the way people draw (tiny juts in opposite direction)
 422         //Different solutions
 423         //      Average at both end points (will probably eliminate many points at each end of the samples)
 424         //      Average after the break points are found (weird points would still affect the curve)
 425         //      Just always get rid of breaks at the beginning and end if they are a certain distance apart
 426         //              This is will be current approach so all we do now is try to remove duplicate points
 427
 428         //remove duplicate points - very bad for fitting
 429
 430         //timer.reset();
 431
 432         {
 433                 std::list<synfig::Point>::const_iterator i = in.begin(), end = in.end();
 434                 std::list<synfig::Real>::const_iterator iw = in_w.begin();
 435                 synfig::Point   c;
 436
 437                 if(in.size() == in_w.size())
 438                 {
 439                         for(;i != end; ++i,++iw)
 440                         {
 441                                 //eliminate duplicate points
 442                                 if(*i != c)
 443                                 {
 444                                         f.push_back(c = *i);
 445                                         f_w.push_back(*iw);
 446                                 }
 447                         }
 448                 }else
 449                 {
 450                         for(;i != end; ++i)
 451                         {
 452                                 //eliminate duplicate points
 453                                 if(*i != c)
 454                                 {
 455                                         f.push_back(c = *i);
 456                                 }
 457                         }
 458                 }
 459         }
 460         //initialprocess = timer();
 461
 462         if(f.size()<=6)
 463                 return;
 464
 465         //get curvature information
 466         //timer.reset();
 467
 468         {
 469                 int i,i0,i1;
 470                 synfig::Vector v1,v2;
 471
 472                 cvt.resize(f.size());
 473
 474                 cvt.front() = 1;
 475                 cvt.back() = 1;
 476
 477                 for(i = 1; i < (int)f.size()-1; ++i)
 478                 {
 479                         i0 = std::max(0,i - 2);
 480                         i1 = std::min((int)(f.size()-1),i + 2);
 481
 482                         v1 = f[i] - f[i0];
 483                         v2 = f[i1] - f[i];
 484
 485                         cvt[i] = (v1*v2)/(v1.mag()*v2.mag());
 486                 }
 487         }
 488
 489         //curveval = timer();
 490         //synfig::info("calculated curvature");
 491
 492         //find corner points and interpolate inside those
 493         //timer.reset();
 494         {
 495                 //break at sharp derivative points
 496                 //TODO tolerance should be set based upon digitization resolution (length dependent index selection)
 497                 Real    tol = 0;                //break tolerance, for the cosine of the change in angle (really high curvature or something)
 498                 Real    fixdistsq = 4*width*width; //the distance to ignore breaks at the end points (for fixing stuff)
 499                 unsigned int i = 0;
 500
 501                 int             maxi = -1, last=0;
 502                 Real    minc = 1;
 503
 504                 brk.push_back(0);
 505
 506                 for(i = 1; i < cvt.size()-1; ++i)
 507                 {
 508                         //insert if too sharp (we need to break the tangents to insert onto the break list)
 509
 510                         if(cvt[i] < tol)
 511                         {
 512                                 if(cvt[i] < minc)
 513                                 {
 514                                         minc = cvt[i];
 515                                         maxi = i;
 516                                 }
 517                         }else if(maxi >= 0)
 518                         {
 519                                 if(maxi >= last + 8)
 520                                 {
 521                                         //synfig::info("break: %d-%d",maxi+1,cvt.size());
 522                                         brk.push_back(maxi);
 523                                         last = maxi;
 524                                 }
 525                                 maxi = -1;
 526                                 minc = 1;
 527                         }
 528                 }
 529
 530                 brk.push_back(i);
 531
 532                 //postprocess for breaks too close to eachother
 533                 Real d = 0;
 534                 Point p = f[brk.front()];
 535
 536                 //first set
 537                 for(i = 1; i < brk.size()-1; ++i) //do not want to include end point...
 538                 {
 539                         d = (f[brk[i]] - p).mag_squared();
 540                         if(d > fixdistsq) break; //don't want to group breaks if we ever get over the dist...
 541                 }
 542                 //want to erase all points before...
 543                 if(i != 1)
 544                         brk.erase(brk.begin(),brk.begin()+i-1);
 545
 546                 //end set
 547                 p = f[brk.back()];
 548                 for(i = brk.size()-2; i > 0; --i) //start at one in from the end
 549                 {
 550                         d = (f[brk[i]] - p).mag_squared();
 551                         if(d > fixdistsq) break; //don't want to group breaks if we ever get over the dist
 552                 }
 553                 if(i != brk.size()-2)
 554                         brk.erase(brk.begin()+i+2,brk.end()); //erase all points that we found... found none if i has not advanced
 555                 //must not include the one we ended up on
 556         }
 557         //breakeval = timer();
 558         //synfig::info("found break points: %d",brk.size());
 559
 560         //get the distance calculation of the entire curve (for tangent scaling)
 561
 562         //timer.reset();
 563         {
 564                 synfig::Point p1,p2;
 565
 566                 p1=p2=f[0];
 567
 568                 di.resize(f.size()); d_i.resize(f.size());
 569                 Real d = 0;
 570                 for(unsigned int i = 0; i < f.size();)
 571                 {
 572                         d += (d_i[i] = (p2-p1).mag());
 573                         di[i] = d;
 574
 575                         p1=p2;
 576                         p2=f[++i];
 577                 }
 578         }
 579         //disteval = timer();
 580         //synfig::info("calculated distance");
 581
 582         //now break at every point - calculate new derivatives each time
 583
 584         //TODO
 585         //must be sure that the break points are 3 or more apart
 586         //then must also store the breaks which are not smooth, etc.
 587         //and figure out tangents between there
 588
 589         //for each pair of break points (as long as they are far enough apart) recursively subdivide stuff
 590         //ignore the detected intermediate points
 591         {
 592                 unsigned int i0=0,i3=0,is=0;
 593                 int i=0,j=0;
 594
 595                 bool done = false;
 596
 597                 Real errortol = smoothness*pixelwidth; //???? what the hell should this value be
 598
 599                 BLinePoint a;
 600                 synfig::Vector v;
 601
 602                 //intemp = f; //don't want to smooth out the corners
 603
 604                 bool breaktan = false, setwidth;
 605                 a.set_split_tangent_flag(false);
 606                 //a.set_width(width);
 607                 a.set_width(1.0f);
 608
 609                 setwidth = (f.size() == f_w.size());
 610
 611                 for(j = 0; j < (int)brk.size() - 1; ++j)
 612                 {
 613                         //for b[j] to b[j+1] subdivide and stuff
 614                         i0 = brk[j];
 615                         i3 = brk[j+1];
 616
 617                         unsigned int size = i3-i0+1; //must include the end points
 618
 619                         //new derivatives
 620                         //timer.reset();
 621                         ftemp.assign(f.begin()+i0, f.begin()+i3+1);
 622                         for(i=0;i<20;++i)
 623                                 gaussian_blur_3(ftemp.begin(),ftemp.end(),false);
 624
 625                         df.resize(size);
 626                         GetFirstDerivatives(ftemp,0,size,(char*)&df[0],sizeof(df[0]));
 627                         //GetSimpleDerivatives(ftemp,0,size,df,0,di);
 628                         //< don't have to worry about indexing stuff as it is all being taken car of right now
 629                         //preproceval += timer();
 630                         //numpre++;
 631
 632                         work.resize(size*2-1); //guarantee that all points will be tesselated correctly (one point inbetween every 2 adjacent points)
 633
 634                         //if size of work is size*2-1, the step size should be 1/(size*2 - 2)
 635                         //Real step = 1/(Real)(size*2 - 1);
 636
 637                         //start off with break points as indices
 638                         curind.clear();
 639                         curind.push_back(cpindex(i0,di[i3]-di[i0],0)); //0 error because no curve on the left
 640                         curind.push_back(cpindex(i3,di[i3]-di[i0],-1)); //error needs to be reevaluated
 641                         done = false; //we want to loop
 642
 643                         unsigned int dcount = 0;
 644
 645                         //while there are still enough points between us, and the error is too high subdivide (and invalidate neighbors that share tangents)
 646                         while(!done)
 647                         {
 648                                 //tesselate all curves with invalid error values
 649                                 work[0] = f[i0];
 650
 651                                 //timer.reset();
 652                                 /*numtess += */tesselate_curves(curind,f,df,work);
 653                                 //tesseval += timer();
 654
 655                                 //now get all error values
 656                                 //timer.reset();
 657                                 for(i = 1; i < (int)curind.size(); ++i)
 658                                 {
 659                                         if(curind[i].error < 0) //must have been retesselated, so now recalculate error value
 660                                         {
 661                                                 //evaluate error from points (starting at current index)
 662                                                 int size = curind[i].curind - curind[i-1].curind + 1;
 663                                                 curind[i].error = CurveError(&f[curind[i-1].curind], size,
 664                                                                                                          work,(curind[i-1].curind - i0)*2,(curind[i].curind - i0)*2+1);
 665
 666                                                 /*if(curind[i].error > 1.0e5)
 667                                                 {
 668                                                         synfig::info("Holy crap %d-%d error %f",curind[i-1].curind,curind[i].curind,curind[i].error);
 669                                                         curind[i].error = -1;
 670                                                         numtess += tesselate_curves(curind,f,df,work);
 671                                                         curind[i].error = CurveError(&f[curind[i-1].curind], size,
 672                                                                                                          work,0,work.size());//(curind[i-1].curind - i0)*2,(curind[i].curind - i0)*2+1);
 673                                                 }*/
 674                                                 //numerror++;
 675                                         }
 676                                 }
 677                                 //erroreval += timer();
 678
 679                                 //assume we're done
 680                                 done = true;
 681
 682                                 //check each error to see if it's too big, if so, then subdivide etc.
 683                                 int indsize = (int)curind.size();
 684                                 Real maxrelerror = 0;
 685                                 int maxi = -1;//, numpoints;
 686
 687                                 //timer.reset();
 688                                 //get the maximum error and split there
 689                                 for(i = 1; i < indsize; ++i)
 690                                 {
 691                                         //numpoints = curind[i].curind - curind[i-1].curind + 1;
 692
 693                                         if(curind[i].error > maxrelerror && curind[i].curind - curind[i-1].curind > 2) //only accept if it's valid
 694                                         {
 695                                                 maxrelerror = curind[i].error;
 696                                                 maxi = i;
 697                                         }
 698                                 }
 699
 700                                 //split if error is too great
 701                                 if(maxrelerror > errortol)
 702                                 {
 703                                         //add one to the left etc
 704                                         unsigned int    ibase = curind[maxi-1].curind, itop = curind[maxi].curind,
 705                                                                         ibreak = (ibase + itop)/2;
 706                                         Real scale, scale2;
 707
 708                                         assert(ibreak < f.size());
 709
 710                                         //synfig::info("Split %d -%d- %d, error: %f", ibase,ibreak,itop,maxrelerror);
 711
 712                                         if(ibase != itop)
 713                                         {
 714                                                 //invalidate current error of the changed tangents and add an extra segment
 715                                                 //enforce minimum tangents property
 716                                                 curind[maxi].error = -1;
 717                                                 curind[maxi-1].error = -1;
 718                                                 if(maxi+1 < indsize) curind[maxi+1].error = -1; //if there is a curve segment beyond this it will be effected as well
 719
 720                                                 scale = di[itop] - di[ibreak];
 721                                                 scale2 = maxi+1 < indsize ? di[curind[maxi+1].curind] - di[itop] : scale; //to the right valid?
 722                                                 curind[maxi].tangentscale = std::min(scale, scale2);
 723
 724                                                 scale = di[ibreak] - di[ibase];
 725                                                 scale2 = maxi >= 2 ? di[ibase] - di[curind[maxi-2].curind] : scale; // to the left valid -2 ?
 726                                                 curind[maxi-1].tangentscale = std::min(scale, scale2);
 727
 728                                                 scale = std::min(di[ibreak] - di[ibase], di[itop] - di[ibreak]);
 729
 730                                                 curind.insert(curind.begin()+maxi,cpindex(ibreak, scale, -1));
 731                                                 //curind.push_back(cpindex(ibreak, scale, -1));
 732                                                 //std::sort(curind.begin(), curind.end());
 733
 734                                                 done = false;
 735                                                 //numsplit++;
 736                                         }
 737                                 }
 738                                 //spliteval += timer();
 739
 740                                 dcount++;
 741                         }
 742
 743                         //insert the last point too (just set tangent for now
 744                         is = curind[0].curind;
 745
 746                         //first point inherits current tangent status
 747                         v = df[is - i0];
 748                         if(v.mag_squared() > EPSILON)
 749                                 v *= (curind[0].tangentscale/v.mag());
 750
 751                         if(!breaktan)
 752                                 a.set_tangent(v);
 753                         else a.set_tangent2(v);
 754
 755                         a.set_vertex(f[is]);
 756                         if(setwidth)a.set_width(f_w[is]);
 757
 758                         out.push_back(a);
 759                         a.set_split_tangent_flag(false); //won't need to break anymore
 760                         breaktan = false;
 761
 762                         for(i = 1; i < (int)curind.size()-1; ++i)
 763                         {
 764                                 is = curind[i].curind;
 765
 766                                 //first point inherits current tangent status
 767                                 v = df[is-i0];
 768                                 if(v.mag_squared() > EPSILON)
 769                                         v *= (curind[i].tangentscale/v.mag());
 770
 771                                 a.set_tangent(v); // always inside, so guaranteed to be smooth
 772                                 a.set_vertex(f[is]);
 773                                 if(setwidth)a.set_width(f_w[is]);
 774
 775                                 out.push_back(a);
 776                         }
 777
 778                         //set the last point's data
 779                         is = curind.back().curind; //should already be this
 780
 781                         v = df[is-i0];
 782                         if(v.mag_squared() > EPSILON)
 783                                 v *= (curind.back().tangentscale/v.mag());
 784
 785                         a.set_tangent1(v);
 786                         a.set_split_tangent_flag(true);
 787                         breaktan = true;
 788
 789                         //will get the vertex and tangent 2 from next round
 790                 }
 791
 792                 a.set_vertex(f[i3]);
 793                 a.set_split_tangent_flag(false);
 794                 if(setwidth)
 795                         a.set_width(f_w[i3]);
 796                 out.push_back(a);
 797
 798                 /*etl::clock::value_type totaltime = total(),
 799                                                            misctime = totaltime - initialprocess - curveval - breakeval - disteval
 800                                                                           - preproceval - tesseval - erroreval - spliteval;
 801
 802                 synfig::info(
 803                         "Curve Convert Profile:\n"
 804                         "\tInitial Preprocess:    %f\n"
 805                         "\tCurvature Calculation: %f\n"
 806                         "\tBreak Calculation:     %f\n"
 807                         "\tDistance Calculation:  %f\n"
 808                         "  Algorithm: (numtimes,totaltime)\n"
 809                         "\tPreprocess step:      (%d,%f)\n"
 810                         "\tTesselation step:     (%d,%f)\n"
 811                         "\tError step:           (%d,%f)\n"
 812                         "\tSplit step:           (%d,%f)\n"
 813                         "  Num Input: %d, Num Output: %d\n"
 814                         "  Total time: %f, Misc time: %f\n",
 815                         initialprocess, curveval,breakeval,disteval,
 816                         numpre,preproceval,numtess,tesseval,numerror,erroreval,numsplit,spliteval,
 817                         in.size(),out.size(),
 818                         totaltime,misctime);*/
 819
 820                 return;
 821         }
 822 }
 823
 824 void synfigapp::BLineConverter::EnforceMinWidth(std::list<synfig::BLinePoint> &bline, synfig::Real min_pressure)
 825 {
 826         std::list<synfig::BLinePoint>::iterator i = bline.begin(),
 827                                                                                         end = bline.end();
 828
 829         for(i = bline.begin(); i != end; ++i)
 830         {
 831                 if(i->get_width() < min_pressure)
 832                 {
 833                         i->set_width(min_pressure);
 834                 }
 835         }
 836 }