egra: optimised SSE memfill and colorblend

[iv.d.git] / region.d

/*
 * Pixel Region Operations
 * coded by Ketmar // Invisible Vector <ketmar@ketmar.no-ip.org>
 * Understanding is not required. Only obedience.
 *
 * This program is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, version 3 of the License ONLY.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program. If not, see <http://www.gnu.org/licenses/>.
 */
module iv.region /*is aliced*/;
import iv.alice;


// ////////////////////////////////////////////////////////////////////////// //
/// regions are copy-on-write shared, yay
struct Region {
  alias SpanType = ushort; /// you probably will never need this, but...

  /// combine operation for region combiner %-)
  enum CombineOp {
    Or, /// logic or
    And, /// logic and
    Xor, /// logic exclusive or
    Cut, /// cut solid parts
    NCut, /// cut empty parts
  }

  @property pure const nothrow @safe @nogc {
    int width () { pragma(inline, true); return (rdatap ? rdata.rwdt : 0); } ///
    int height () { pragma(inline, true); return (rdatap ? rdata.rhgt : 0); } ///
    bool solid () { pragma(inline, true); return (rdatap ? rdata.simple && rdata.rwdt > 0 && rdata.rhgt > 0 && rdata.simpleSolid : false); } ///
    bool empty () { pragma(inline, true); return (rdatap ? rdata.rwdt < 1 || rdata.rhgt < 1 || (rdata.simple && !rdata.simpleSolid) : true); } ///
  }

  /// can be used to save region data
  @property uint[] getData () const nothrow @safe {
    uint[] res;
    if (rdata.simple) {
      res ~= 1|(rdata.simpleSolid ? 2 : 0);
    } else {
      res ~= 0;
    }
    res[0] |= cast(int)(SpanType.sizeof<<4);
    res ~= rdata.rwdt;
    res ~= rdata.rhgt;
    if (!rdata.simple) {
      res ~= rdata.lineofs;
      foreach (SpanType d; rdata.data) res ~= d;
    }
    return res;
  }

  /// can be used to restore region data
  void setData (const(int)[] data) nothrow @safe {
    cow!false();
    if (data.length < 3 || (data[0]>>4) != SpanType.sizeof) assert(0, "invalid region data");
    rdata.rwdt = data[1];
    rdata.rhgt = data[2];
    rdata.simple = ((data[0]&0x01) != 0);
    rdata.lineofs = null;
    rdata.data = null;
    if (rdata.simple) {
      rdata.simpleSolid = ((data[0]&0x02) != 0);
    } else {
      foreach (int d; data[3..3+rdata.rhgt]) rdata.lineofs ~= d;
      foreach (int d; data[3+rdata.rhgt..$]) rdata.data ~= cast(SpanType)d;
    }
  }

  /// this creates solid region (by default)
  this (int awidth, int aheight, bool solid=true) nothrow @safe @nogc { setSize(awidth, aheight, solid); }
  ~this () nothrow @safe @nogc { decRC(); } /// release this region data
  this (this) nothrow @safe @nogc { if (rdatap) ++rdata.rc; } /// share this region data

  ///
  void setSize (int awidth, int aheight, bool solid=true) nothrow @safe @nogc {
    if (awidth <= 0 || aheight <= 0) awidth = aheight = 0;
    if (awidth > SpanType.max-1 || aheight > SpanType.max-1) assert(0, "Region internal error: region dimensions are too big");
    cow!false();
    rdata.rwdt = awidth;
    rdata.rhgt = aheight;
    rdata.simpleSolid = solid;
    rdata.lineofs = null;
    rdata.data = null;
  }

  /// is given point visible?
  bool visible (int x, int y) const pure nothrow @safe @nogc {
    // easiest cases
    if (!rdatap) return false;
    if (rdata.rwdt < 1 || rdata.rhgt < 1) return false;
    if (x < 0 || y < 0 || x >= rdata.rwdt || y >= rdata.rhgt) return false;
    if (rdata.simple) return true; // ok, easy case here
    // now the hard one
    immutable ldofs = rdata.lineofs[y];
    immutable len = rdata.data[ldofs];
    debug assert(len > 1);
    auto line = rdata.data[ldofs+1..ldofs+len];
    int idx = void; // will be initied in mixin
    mixin(FindSpanMixinStr!"idx");
    debug assert(idx < line.length); // too far (the thing that should not be)
    return ((idx+(line[idx] == x))%2 == 0);
  }

  /// punch a hole
  void punch (int x, int y, int w=1, int h=1) nothrow @trusted { pragma(inline, true); doPunchPatch!"punch"(x, y, w, h); }

  /// patch a hole
  void patch (int x, int y, int w=1, int h=1) nothrow @trusted { pragma(inline, true); doPunchPatch!"patch"(x, y, w, h); }

  // ////////////////////////////////////////////////////////////////////////// //
  enum State { Mixed = -1, Empty, Solid } /// WARNING! don't change the order!

  /// return span state %-)
  State spanState (int y, int x0, int x1) const pure nothrow @safe @nogc {
    if (rdata is null) return State.Empty;
    if (y < 0 || y >= rdata.rhgt || x1 < 0 || x0 >= rdata.rwdt || x1 < x0) return State.Empty;
    if (rdata.simple) {
      // if our span is not fully inside, it can be either Empty or Mixed
      if (rdata.simpleSolid) {
        return (x0 >= 0 && x1 < rdata.rwdt ? State.Solid : State.Mixed);
      } else {
        return State.Empty;
      }
    }
    immutable ldofs = rdata.lineofs[y];
    immutable len = rdata.data[ldofs];
    debug assert(len > 1);
    auto line = rdata.data[ldofs+1..ldofs+len];
    int idx = void; // will be initied in mixin
    immutable x = (x0 >= 0 ? x0 : 0);
    mixin(FindSpanMixinStr!"idx");
    debug assert(idx < line.length); // too far (the thing that should not be)
    // move to "real" span
    if (line[idx] == x) ++idx;
    // now, sx is line[idx-1], ex is line[idx]
    if (x1 >= (idx < line.length ? line[idx] : rdata.rwdt)) return State.Mixed;
    idx = (idx^1)&1; // we are interested only in last bit, and we converted it to State here
    // if our span is not fully inside, it can be either Empty or Mixed
    if (idx == State.Solid && x0 < 0) return State.Mixed;
    return cast(State)idx;
  }

  static private template IsGoodSDG(T) {
    private import std.traits;
    static private template IsGoodRT(T) { enum IsGoodRT = is(T == void) || is(T == bool) || is(T : int); }
    static private template IsGoodAT(T) { enum IsGoodAT = is(T == int) || is(T == long) || is(T == uint) || is(T == ulong); }
    enum IsGoodSDG = isCallable!T && IsGoodRT!(ReturnType!T) && (variadicFunctionStyle!T == Variadic.no) &&
      Parameters!T.length == 2 && IsGoodAT!(Parameters!T[0]) && IsGoodAT!(Parameters!T[1]);
  }

  /// call delegate for each solid or empty span
  /// for non-void returning delegates, return !0 to exit
  auto spans(bool solids=true, DG) (int y, int x0, int x1, scope DG dg) const if (IsGoodSDG!DG) { return spansEnumerator!(DG, solids)(y, 0, x0, x1, dg); }

  /// call delegate for each solid or empty span
  /// for non-void returning delegates, return !0 to exit
  /// `ofsx` will be automatically subtracted from `x0` and `x1` args, and added to `x0` and `x1` delegate args
  auto spans(bool solids=true, DG) (int y, int ofsx, int x0, int x1, scope DG dg) const if (IsGoodSDG!DG) { return spansEnumerator!(DG, solids)(y, ofsx, x0, x1, dg); }

  /// element of span range
  static struct XPair { int x0, x1; }

  /// get range of spans
  auto spanRange(bool solids=true) (int y, int x0, int x1) nothrow @safe @nogc { return spanRange!solids(y, 0, x0, x1); }

  /// ditto
  auto spanRange(bool solids=true) (int y, int ofsx, int x0, int x1) nothrow @safe @nogc {
    static struct SpanRange(bool solids) {
      int ofsx, x0, x1, rwdt, idx;
      ubyte eosNM; // empty(bit0), nomore(bit1) ;-)
      XPair fpair; // front
      const(SpanType)[] line;

    nothrow @trusted @nogc:
      this (ref Region reg, int y, int aofsx, int ax0, int ax1) {
        ofsx = aofsx;
        x0 = ax0;
        x1 = ax1;
        if (x0 > x1) { eosNM = 0x01; return; }
        if (reg.rdata is null) {
          static if (!solids) {
            fpair.x0 = x0;
            fpair.x1 = x1;
            eosNM = 0x02;
          } else {
            eosNM = 0x01;
          }
          return;
        }
        rwdt = reg.rdata.rwdt;
        x0 -= ofsx;
        x1 -= ofsx;
        if (y < 0 || y >= reg.rdata.rhgt || x1 < 0 || x0 >= rwdt || x1 < x0) {
          static if (!solids) {
            fpair.x0 = x0+ofsx;
            fpair.x1 = x1+ofsx;
            eosNM = 0x02;
          } else {
            eosNM = 0x01;
          }
          return;
        }
        if (reg.rdata.simple) {
          if (reg.rdata.simpleSolid) {
            static if (solids) {
              if (x0 < 0) x0 = 0;
              if (x1 >= rwdt) x1 = rwdt-1;
              if (x0 <= x1) {
                fpair.x0 = x0+ofsx;
                fpair.x1 = x1+ofsx;
                eosNM = 0x02;
              } else {
                eosNM = 0x01;
              }
            } else {
              if (x0 < 0) {
                fpair.x0 = x0+ofsx;
                fpair.x1 = -1+ofsx;
                eosNM = (x1 < rwdt ? 0x02 : 0x04);
              } else {
                if (x1 >= rwdt) {
                  fpair.x0 = rwdt+ofsx;
                  fpair.x1 = x1+ofsx;
                  eosNM = 0x02;
                } else {
                  eosNM = 0x01;
                }
              }
            }
          } else {
            static if (!solids) {
              fpair.x0 = x0+ofsx;
              fpair.x1 = x1+ofsx;
              eosNM = 0x02;
            } else {
              eosNM = 0x01;
            }
          }
          return;
        }
        // edge cases are checked
        immutable ldofs = reg.rdata.lineofs[y];
        immutable len = reg.rdata.data[ldofs];
        debug assert(len > 1);
        line = reg.rdata.data[ldofs+1..ldofs+len];
        // beyond left border? move to first solid span
        bool hasOne = false;
        if (x0 < 0) {
          int ex = (line[0] == 0 ? line[1]-1 : -1);
          // is first span empty too?
          if (ex >= x1) {
            static if (!solids) {
              fpair.x0 = x0+ofsx;
              fpair.x1 = x1+ofsx;
              eosNM = 0x02;
            } else {
              eosNM = 0x01;
            }
            return;
          }
          static if (!solids) {
            fpair.x0 = x0+ofsx;
            fpair.x1 = ex+ofsx;
            hasOne = true;
            if (x0 == -9) {
              //import iv.writer; writeln("*");
            }
          }
          x0 = ex+1;
        }
        static if (solids) { if (x1 >= rwdt) x1 = rwdt-1; }
        //int idx = void; // will be initied in mixin
        alias x = x0; // for mixin
        mixin(FindSpanMixinStr!"idx");
        debug assert(idx < line.length); // too far (the thing that should not be)
        // move to "real" span, so sx is line[idx-1], ex+1 is line[idx]
        if (line[idx] == x) ++idx;
        if (!hasOne) popFront();
      }

      @property auto save () pure {
        SpanRange!solids res;
        res.ofsx = ofsx;
        res.x0 = x0;
        res.x1 = x1;
        res.rwdt = rwdt;
        res.idx = idx;
        res.eosNM = eosNM;
        res.fpair = fpair;
        res.line = line;
        return res;
      }

      @property bool empty () const pure { return ((eosNM&0x01) != 0); }
      @property XPair front () const pure { return XPair(fpair.x0, fpair.x1); }

      void popFront () {
        if (eosNM&0x02) eosNM = 0x01;
        if (eosNM&0x01) return;
        // edge case
        if (eosNM&0x04) {
          if (x1 >= rwdt) {
            static if (!solids) {
              fpair.x0 = rwdt+ofsx;
              fpair.x1 = x1+ofsx;
              eosNM = 0x02;
            } else {
              eosNM = 0x01;
            }
          } else {
            eosNM = 0x01;
          }
          return;
        }
        bool hasOne = false;
        // process spans
        while (x0 <= x1) {
          int ex = line[idx]-1;
          int cex = (ex < x1 ? ex : x1); // clipped ex
          // emit part from x0 to ex if necessary
          static if (solids) {
            // current span is solid?
            if (idx%2 == 0) {
              fpair.x0 = x0+ofsx;
              fpair.x1 = cex+ofsx;
              hasOne = true;
            }
          } else {
            // current span is empty?
            if (idx%2 == 1) {
              fpair.x0 = x0+ofsx;
              fpair.x1 = (ex < rwdt-1 ? cex : x1)+ofsx;
              hasOne = true;
              if (ex == rwdt-1) { eosNM = 0x02; return; }
            } else {
              if (ex == rwdt-1) { x0 = rwdt; break; }
            }
          }
          x0 = ex+1;
          ++idx;
          if (hasOne) return;
        }
        if (hasOne) return;
        static if (!solids) {
          if (x0 <= x1) {
            fpair.x0 = x0+ofsx;
            fpair.x1 = x1+ofsx;
            eosNM = 0x02;
            return;
          }
        }
        eosNM = 0x01;
      }
    }

    return SpanRange!solids(this, y, ofsx, x0, x1);
  }

private:
  // ////////////////////////////////////////////////////////////////////////// //
  auto spansEnumerator(DG, bool solids) (int y, int ofsx, int x0, int x1, scope DG dg) const {
    import std.traits : ReturnType;
    static if (is(ReturnType!DG == void)) {
      enum ReturnFail = "return;";
      enum DgCall(string args) = "dg("~args~");";
    } else {
      static if (is(ReturnType!DG == bool)) enum ReturnFail = "return false;"; else enum ReturnFail = "return 0;";
      enum DgCall(string args) = "if (auto xres = dg("~args~")) return xres;";
    }
    if (x0 > x1 || dg is null) mixin(ReturnFail);
    if (rdata is null) {
      static if (!solids) dg(x0, x1);
      mixin(ReturnFail);
    }
    x0 -= ofsx;
    x1 -= ofsx;
    if (y < 0 || y >= rdata.rhgt || x1 < 0 || x0 >= rdata.rwdt || x1 < x0) {
      static if (!solids) { mixin(DgCall!"x0+ofsx, x1+ofsx"); }
      mixin(ReturnFail);
    }
    if (rdata.simple) {
      if (rdata.simpleSolid) {
        static if (solids) {
          if (x0 < 0) x0 = 0;
          if (x1 >= rdata.rwdt) x1 = rdata.rwdt-1;
          if (x0 <= x1) { mixin(DgCall!"x0+ofsx, x1+ofsx"); }
        } else {
          if (x0 < 0) { mixin(DgCall!"x0+ofsx, -1+ofsx"); }
          if (x1 >= rdata.rwdt) { mixin(DgCall!"rdata.rwdt+ofsx, x1+ofsx"); }
        }
      } else {
        static if (!solids) { mixin(DgCall!"x0+ofsx, x1+ofsx"); }
      }
      mixin(ReturnFail);
    }
    immutable ldofs = rdata.lineofs[y];
    immutable len = rdata.data[ldofs];
    debug assert(len > 1);
    auto line = rdata.data[ldofs+1..ldofs+len];
    // beyond left border? move to first solid span
    if (x0 < 0) {
      int ex = (rdata.data[ldofs+1] == 0 ? rdata.data[ldofs+2]-1 : -1);
      // is first span empty too?
      if (ex >= x1) { static if (!solids) { mixin(DgCall!"x0+ofsx, x1+ofsx"); } mixin(ReturnFail); }
      static if (!solids) { mixin(DgCall!"x0+ofsx, ex+ofsx"); }
      x0 = ex+1;
    }
    static if (solids) { if (x1 >= rdata.rwdt) x1 = rdata.rwdt-1; }
    int idx = void; // will be initied in mixin
    alias x = x0; // for mixin
    mixin(FindSpanMixinStr!"idx");
    debug assert(idx < line.length); // too far (the thing that should not be)
    // move to "real" span, so sx is line[idx-1], ex+1 is line[idx]
    if (line[idx] == x) ++idx;
    // process spans
    while (x0 <= x1) {
      int ex = line[idx]-1;
      int cex = (ex < x1 ? ex : x1); // clipped ex
      // emit part from x0 to ex if necessary
      static if (solids) {
        // current span is solid?
        if (idx%2 == 0) { mixin(DgCall!"x0+ofsx, cex+ofsx"); }
      } else {
        // current span is empty?
        if (idx%2 == 1) {
          { mixin(DgCall!"x0+ofsx, (ex < rdata.rwdt-1 ? cex : x1)+ofsx"); }
          if (ex == rdata.rwdt-1) mixin(ReturnFail);
        } else {
          if (ex == rdata.rwdt-1) { x0 = rdata.rwdt; break; }
        }
      }
      x0 = ex+1;
      ++idx;
      //static if (!solids) { if (x0 == rdata.rwdt) break; }
    }
    static if (!solids) { if (x0 <= x1) { mixin(DgCall!"x0+ofsx, x1+ofsx"); } }
    mixin(ReturnFail);
  }

  // ////////////////////////////////////////////////////////////////////////// //
  // find element index
  // always returns index of key which is >= `x`
  private enum FindSpanMixinStr(string minAndRes) = "{
    ("~minAndRes~") = 0;
    int max = cast(int)line.length;
    while (("~minAndRes~") < max) {
      int mid = (("~minAndRes~")+max)/2; // ignore possible overflow, it can't happen here
      debug assert(mid < max);
      if (line[mid] < x) ("~minAndRes~") = mid+1; else max = mid;
    }
    //return ("~minAndRes~"); // actually, key is found if (max == min/*always*/ && min < line.length && line[min] == x)
  }";

  // ////////////////////////////////////////////////////////////////////////// //
  // punch a hole, patch a hole
  // mode: "punch", "patch"
  //FIXME: overflows
  void doPunchPatch(string mode) (int x, int y, int w=1, int h=1) nothrow @trusted {
    static assert(mode == "punch" || mode == "patch", "Region: invalid mode: "~mode);
    if (rdata is null) return;
    static if (mode == "punch") {
      if (empty) return;
    } else {
      if (solid) return;
    }
    if (w < 1 || h < 1) return;
    if (x >= rdata.rwdt || y >= rdata.rhgt) return;
    //TODO: overflow check
    if (x < 0) {
      if (x+w <= 0) return;
      w += x;
      x = 0;
    }
    if (y < 0) {
      if (y+h <= 0) return;
      h += y;
      y = 0;
    }
    int x1 = x+w-1;
    if (x1 >= rdata.rwdt) x1 = rdata.rwdt-1;
    debug assert(x <= x1);
    int y1 = y+h-1;
    if (y1 >= rdata.rhgt) y1 = rdata.rhgt-1;
    debug assert(y <= y1);
    foreach (int cy; y..y1+1) doPunchPatchLine!mode(cy, x, x1);
  }


  // ////////////////////////////////////////////////////////////////////////// //
  void makeRoom (usize ofs, ssize count) nothrow @trusted {
    import core.stdc.string : memmove;
    debug assert(ofs <= rdata.data.length);
    if (count > 0) {
      // make room
      // `assumeSafeAppend` was already called in caller
      //rdata.data.assumeSafeAppend.length += count;
      rdata.data.length += count;
      if (ofs+count < rdata.data.length) memmove(rdata.data.ptr+ofs+count, rdata.data.ptr+ofs, SpanType.sizeof*(rdata.data.length-ofs-count));
    } else if (count < 0) {
      // remove rdata.data
      count = -count;
      debug assert(ofs+count <= rdata.data.length);
      if (ofs+count == rdata.data.length) {
        rdata.data.length = ofs;
      } else {
        immutable auto left = rdata.data.length-ofs-count;
        memmove(rdata.data.ptr+ofs, rdata.data.ptr+ofs+count, SpanType.sizeof*(rdata.data.length-ofs-count));
        rdata.data.length -= count;
      }
      //rdata.data.assumeSafeAppend; // in case we will want to grow later
    }
  }

  // replace span data at plofs with another data from spofs, return # of bytes added (or removed, if negative)
  ssize replaceSpanData (usize plofs, usize spofs) nothrow @trusted {
    //import core.stdc.string : memcpy;
    debug assert(spofs < rdata.data.length && spofs+rdata.data[spofs] == rdata.data.length);
    debug assert(plofs <= spofs && plofs+rdata.data[plofs] <= spofs);
    if (plofs == spofs) return 0; // nothing to do; just in case
    auto oldlen = rdata.data[plofs];
    auto newlen = rdata.data[spofs];
    // same length?
    ssize ins = cast(ssize)newlen-cast(ssize)oldlen;
    if (ins) {
      makeRoom(plofs, ins);
      spofs += ins;
    }
    if (newlen > 0) rdata.data[plofs..plofs+newlen] = rdata.data[spofs..spofs+newlen]; //memcpy(rdata.data.ptr+plofs, rdata.data.ptr+spofs, SpanType.sizeof*newlen);
    return ins;
  }

  // insert span data from spofs at plofs
  void insertSpanData (usize plofs, usize spofs) nothrow @trusted {
    //import core.stdc.string : memcpy;
    debug assert(spofs < rdata.data.length && spofs+rdata.data[spofs] == rdata.data.length);
    debug assert(plofs <= spofs);
    if (plofs == spofs) return; // nothing to do; just in case
    auto newlen = rdata.data[spofs];
    makeRoom(plofs, newlen);
    spofs += newlen;
    rdata.data[plofs..plofs+newlen] = rdata.data[spofs..spofs+newlen];
    //memcpy(rdata.data.ptr+plofs, rdata.data.ptr+spofs, SpanType.sizeof*newlen);
  }

  bool isEqualLines (int y0, int y1) nothrow @trusted @nogc {
    import core.stdc.string : memcmp;
    if (y0 < 0 || y1 < 0 || y0 >= rdata.rhgt || y1 >= rdata.rhgt) return false;
    auto ofs0 = rdata.lineofs[y0];
    auto ofs1 = rdata.lineofs[y1];
    if (rdata.data[ofs0] != rdata.data[ofs1]) return false;
    return (memcmp(rdata.data.ptr+ofs0, rdata.data.ptr+ofs1, SpanType.sizeof*rdata.data[ofs0]) == 0);
  }

  // all args must be valid
  // [x0..x1]
  void doPunchPatchLine(string mode) (int y, int x0, int x1) nothrow @trusted {
    static if (mode == "patch") {
      if (rdata.simple && rdata.simpleSolid) return; // no need to patch completely solid region
    } else {
      if (rdata.simple && !rdata.simpleSolid) return; // no need to patch completely empty region
    }

    // check if we really have to do anything here
    static if (mode == "patch") {
      if (spanState(y, x0, x1) == State.Solid) return;
      //enum psmode = true;
      enum op = CombineOp.Or;
    } else {
      if (spanState(y, x0, x1) == State.Empty) return;
      //enum psmode = false;
      enum op = CombineOp.Cut;
    }

    doCombine(y, (uint lofs, ref SpanType[] dptr) nothrow @trusted {
      // note that after `assumeSafeAppend` we can increase length without further `assumeSafeAppend` calls
      debug(region_more_prints) { import core.stdc.stdio : printf; printf("op=%d; x0=%d; x1=%d; rwdt=%d\n", cast(int)op, x0, x1, rdata.rwdt); }
      SpanType dsp = cast(SpanType)(x1-x0+1);
      debug(region_more_prints) {
        import core.stdc.stdio : printf;
        auto cspd = CSPD(op, &dsp, x1-x0+1, x0);
        while (!cspd.empty) {
          printf(" (%d,%d,%d)", cspd.sx, cspd.ex, (cspd.solid ? 1 : 0));
          cspd.popFront();
        }
        printf("\n");
      }
      combineSpans(
        (int x) nothrow @trusted { dptr ~= cast(SpanType)x; },
        CSPD(CombineOp.Or, rdata.data.ptr+lofs+1, rdata.rwdt), // base span
        CSPD(op, &dsp, x1-x0+1, x0),
      );
    });
  }

  // `combine`: `lofs` is starting index in `rdata.data` for base line (i.e. length element)
  //            it should  build a new valid line data, starting from `rdata.data.length`, not including line length tho
  // all args must be valid
  void doCombine() (int y, scope void delegate (uint lofs, ref SpanType[] dptr) nothrow @trusted combine) nothrow @trusted {
    // bad luck, build new line
    cow!true();
    if (rdata.simple) {
      // build complex region rdata.data
      if (rdata.lineofs is null) {
        rdata.lineofs.length = rdata.rhgt; // allocate and clear
      } else {
        if (rdata.lineofs.length < rdata.rhgt) rdata.lineofs.assumeSafeAppend;
        rdata.lineofs.length = rdata.rhgt;
        rdata.lineofs[] = 0; // clear
      }
      rdata.data.length = 0;
      if (rdata.simpleSolid) {
        rdata.data.assumeSafeAppend ~= 2; // length
      } else {
        rdata.data.assumeSafeAppend ~= 3; // length
        rdata.data ~= 0; // dummy solid
      }
      rdata.data ~= cast(SpanType)rdata.rwdt; // the only span
      rdata.simple = false;
    }

    auto lofs = rdata.lineofs[y]; // current line offset
    int lsize = rdata.data.ptr[lofs]; // current line size
    auto tmppos = cast(uint)rdata.data.length; // starting position of the new line data

    //patchSpan!psmode(lofs+1, x0, x1);
    rdata.data.assumeSafeAppend ~= 0; // length
    combine(lofs, rdata.data);
    debug(region_more_prints) { import core.stdc.stdio : printf; printf("LEN=%d\n", cast(int)(rdata.data.length-tmppos)); }
    if (rdata.data.length-tmppos > SpanType.max) assert(0, "region internal error: line data too big");
    rdata.data.ptr[tmppos] = cast(SpanType)(rdata.data.length-tmppos);

    debug(region_more_prints) {
      import core.stdc.stdio : printf;
      foreach (SpanType t; rdata.data[tmppos..$]) printf(" %u", cast(uint)t);
      printf("\n");
    }

    int newsize = rdata.data[tmppos]; // size of the new line

    // was this line first in slab?
    auto prevofs = (y > 0 ? rdata.lineofs[y-1] : -1);
    auto nextofs = (y+1 < rdata.rhgt ? rdata.lineofs[y+1] : -2);

    // place new line data, breaking span if necessary
    if (prevofs != lofs && nextofs != lofs) {
      // we were a slab on our own?
      // replace line
      auto delta = replaceSpanData(lofs, tmppos);
      tmppos += delta;
      if (delta) foreach (ref ofs; rdata.lineofs[y+1..$]) ofs += delta;
    } else if (prevofs != lofs && nextofs == lofs) {
      // we were a slab start
      // insert at lofs
      insertSpanData(lofs, tmppos);
      tmppos += newsize;
      foreach (ref ofs; rdata.lineofs[y+1..$]) ofs += newsize;
    } else if (prevofs == lofs && nextofs != lofs) {
      // we were a slab end
      // insert after lofs
      lofs += lsize;
      insertSpanData(lofs, tmppos);
      tmppos += newsize;
      rdata.lineofs[y] = lofs;
      foreach (ref ofs; rdata.lineofs[y+1..$]) ofs += newsize;
    } else {
      //import core.stdc.string : memcpy;
      // we were a slab brick
      debug assert(prevofs == lofs && nextofs == lofs);
      // the most complex case
      // insert us after lofs, insert slab start after us, fix slab and offsets
      // insert us
      lofs += lsize;
      insertSpanData(lofs, tmppos);
      tmppos += newsize;
      rdata.lineofs[y] = lofs;
      // insert old slab start
      lofs += newsize;
      lsize = rdata.data[prevofs];
      makeRoom(lofs, lsize);
      //memcpy(rdata.data.ptr+lofs, rdata.data.ptr+prevofs, SpanType.sizeof*lsize);
      rdata.data[lofs..lofs+lsize] = rdata.data[prevofs..prevofs+lsize];
      // fix current slab
      int ny = y+1;
      while (ny < rdata.rhgt && rdata.lineofs[ny] == prevofs) rdata.lineofs[ny++] = lofs;
      // fix offsets
      newsize += lsize; // simple optimization
      while (ny < rdata.rhgt) rdata.lineofs[ny++] += newsize;
      newsize -= lsize;
    }

    // remove extra data
    lofs = rdata.lineofs[$-1];
    rdata.data.length = lofs+rdata.data[lofs];

    // now check if we can join slabs
    // of course, this is somewhat wasteful, but if we'll combine this
    // check with previous code, the whole thing will explode to an
    // unmaintainable mess; anyway, we aren't on ZX Spectrum
    {
      bool upequ = isEqualLines(y-1, y);
      bool dnequ = isEqualLines(y+1, y);
      if (upequ || dnequ) {
        rdata.data.assumeSafeAppend; // we have to call it after shrinking
        lofs = rdata.lineofs[y];
        debug assert(rdata.data[lofs] == newsize);
        makeRoom(lofs, -newsize); // drop us
        if (upequ && dnequ) {
          // join prev and next slabs by removing two lines...
          auto pofs = rdata.lineofs[y-1];
          makeRoom(lofs, -newsize); // drop next line
          newsize *= 2;
          // and fixing offsets
          rdata.lineofs[y++] = pofs;
          auto sofs = rdata.lineofs[y];
          while (y < rdata.rhgt && rdata.lineofs[y] == sofs) rdata.lineofs[y++] = pofs;
        } else if (upequ) {
          // join prev slab
          rdata.lineofs[y] = rdata.lineofs[y-1];
          ++y;
        } else if (dnequ) {
          // lead next slab
          auto sofs = rdata.lineofs[++y];
          while (y < rdata.rhgt && rdata.lineofs[y] == sofs) rdata.lineofs[y++] = lofs;
        }
        // fix offsets
        foreach (ref ofs; rdata.lineofs[y..$]) ofs -= newsize;
      }
    }

    // check if we can collapse this region
    if (rdata.data.length == 2) {
      if (rdata.data.ptr[0] != 2 || rdata.data.ptr[1] != rdata.rwdt) return;
    } else if (rdata.data.length == 3) {
      if (rdata.data.ptr[0] != 3 || rdata.data.ptr[1] != 0 || rdata.data.ptr[2] != rdata.rwdt) return;
    } else {
      return;
    }
    foreach (immutable ofs; rdata.lineofs[1..$]) if (ofs != 0) return;

    rdata.simple = true;
    //static if (mode == "patch") rdata.simpleSolid = true; else rdata.simpleSolid = false;
    rdata.simpleSolid = (rdata.data.length == 2);
    rdata.lineofs.length = 0; // we may need it later, so keep it
  }

  static struct CSPD {
  nothrow @trusted @nogc:
    const(SpanType)* data;
    int width; // to detect span end
    int xofs;
    CombineOp op; // operation
    bool dsolid; // current span
    int csx;

    this (CombineOp aop, const(SpanType)* adata, int awdt, int axofs=0) {
      // if first span is zero-sized, this region starts with empty span
      op = aop;
      width = awdt;
      xofs = axofs;
      if (*adata == 0) {
        dsolid = false;
        ++adata;
      } else {
        dsolid = true;
      }
      data = adata;
    }

    this() (CombineOp aop, auto ref Region rg, int axofs=0) {
      this(aop, rg.ldata.ptr, rg.width, axofs);
    }

    @disable this (this); // no copies

    @property bool empty () const pure { pragma(inline, true); return (data is null); }
    @property bool solid () const pure { pragma(inline, true); return dsolid; }
    @property int sx () const pure { pragma(inline, true); return xofs+csx; }
    @property int ex () const pure { pragma(inline, true); return xofs+(*data)-1; }
    void popFront () {
      pragma(inline, true);
      csx = *data++;
      if (csx >= width) data = null;
      dsolid = !dsolid;
    }
  }

  // spans[0] should have `int .width`, `empty`, `popFront`, `sx`, `ex`, `solid`
  // others sould have: `empty`, `popFront`, `sx`, `ex`, `solid`, `op`
  // spans[0] should always start at 0 (i.e. it is alpha and omega)
  static void combineSpans(SPR...) (scope void delegate (int x) nothrow @safe putX, auto ref SPR spans) if (SPR.length > 1) {
    bool lastsolid = true; // it's ok
    int lastsx = 0; // it's ok

    void pushSpan() (int ex, bool solid) {
      debug(region_more_prints) {} else pragma(inline, true);
      //debug(region_more_prints) { import core.stdc.stdio : printf; printf("  ex=%d; solid=%d; lastsx=%d; lastsolid=%d\n", ex, (solid ? 1 : 0), lastsx, (lastsolid ? 1 : 0)); }
      //debug if (ex <= lastsx) { import core.stdc.stdio : printf; printf("ex=%d; lastsx=%d\n", ex, lastsx); }
      debug assert(ex >= lastsx);
      if (solid != lastsolid) {
        lastsolid = solid;
        putX(lastsx); // new span starts here
        debug(region_more_prints) { import core.stdc.stdio : printf; printf("   EMIT: %d\n", lastsx); }
      }
      lastsx = ex+1;
    }

    debug assert(!spans[0].empty);
    debug assert(spans[0].sx == 0);
    immutable sp0w = spans[0].width;
    int cursx = 0;
    while (!spans[0].empty) {
      // process other spans
      bool seenAliveSpan = false;
      bool nsolid = spans[0].solid;
      int nex = spans[0].ex;
      foreach (ref sp; spans[1..$]) {
        while (!sp.empty && sp.ex < cursx) sp.popFront();
        if (sp.empty) continue;
        seenAliveSpan = true;
        debug(region_more_prints) { import core.stdc.stdio : printf; printf(" cursx=%d; nex=%d; nsolid=%d; sp.sx=%d; sp.ex=%d; sp.solid=%d\n", cursx, nex, (nsolid ? 1 : 0), sp.sx, sp.ex, (sp.solid ? 1 : 0)); }
        //debug if (sp.sx > cursx) { import core.stdc.stdio : printf; printf("cursx=%d; sp.sx=%d; sp.ex=%d; sp.solid=%d\n", cursx, sp.sx, sp.ex, (sp.solid ? 1 : 0)); }
        //debug assert(sp.sx <= cursx);
        if (sp.sx > nex) continue; // too far
        if (sp.sx > cursx) { nex = sp.sx-1; continue; } // partial
        // do logic op
        final switch (sp.op) {
          case CombineOp.Or: nsolid = nsolid || sp.solid; break;
          case CombineOp.And: nsolid = nsolid && sp.solid; break;
          case CombineOp.Xor: if (sp.solid) nsolid = !nsolid; break;
          case CombineOp.Cut: if (sp.solid) nsolid = false; break;
          case CombineOp.NCut: if (!sp.solid) nsolid = false; break;
        }
        if (sp.ex < nex) nex = sp.ex;
      }
      pushSpan(nex, nsolid);
      if (!seenAliveSpan) {
        // no more alive spans, process span0 till the end
        debug(region_more_prints) { import core.stdc.stdio : printf; printf(" NM!\n"); }
        if (nex < spans[0].ex) pushSpan(spans[0].ex, spans[0].solid); // finish current span
        for (;;) {
          spans[0].popFront();
          if (spans[0].empty) break;
          pushSpan(spans[0].ex, spans[0].solid);
        }
        // put sentinel
        debug assert(lastsx <= sp0w);
        putX(sp0w);
        return;
      }
      if (nex < spans[0].ex) {
        // something was done, and first slab of span0 is not completely eaten
        cursx = nex+1;
      } else {
        // either no alive spans, or first slab of span0 is completely eaten
        spans[0].popFront();
        if (spans[0].empty) { putX(sp0w); return; } // done
        cursx = spans[0].sx;
      }
    }
    // put sentinel
    debug assert(lastsx <= sp0w);
    putX(sp0w);
  }

private:
  usize rdatap = 0; // hide from GC

  @property inout(RData)* rdata () inout pure const nothrow @trusted @nogc { static if (__VERSION__ > 2067) pragma(inline, true); return cast(RData*)rdatap; }

  void decRC () nothrow @trusted @nogc {
    if (rdatap != 0) {
      if (--rdata.rc == 0) {
        import core.memory : GC;
        import core.stdc.stdlib : free;
        GC.removeRange(rdata);
        free(rdata);
      }
      rdatap = 0; // just in case
    }
  }

  // copy-on-write mechanics
  void cow(bool doCopyData) () nothrow @trusted {
    auto srcd = rdata;
    if (srcd is null || srcd.rc != 1) {
      import core.memory : GC;
      import core.stdc.stdlib : malloc, free;
      import core.stdc.string : memcpy;
      auto dstd = cast(RData*)malloc(RData.sizeof);
      if (dstd is null) assert(0, "Region: out of memory"); // this is unlikely, and hey, just crash
      // init with default values
      //*dstd = RData.init;
      static immutable RData initr = RData.init;
      memcpy(dstd, &initr, RData.sizeof);
      //(*dstd).__ctor();
      if (srcd !is null) {
        // copy
        dstd.rwdt = srcd.rwdt;
        dstd.rhgt = srcd.rhgt;
        dstd.simple = srcd.simple;
        dstd.simpleSolid = srcd.simpleSolid;
        dstd.lineofs = null;
        dstd.data = null;
        dstd.rc = 1;
        static if (doCopyData) {
          if (!dstd.simple) {
            // copy complex region
            if (srcd.lineofs.length) dstd.lineofs = srcd.lineofs.dup;
            if (srcd.data.length) dstd.data = srcd.data.dup;
          }
        }
        --srcd.rc;
        assert(srcd.rc > 0);
      }
      rdatap = cast(usize)dstd;
      GC.addRange(rdata, RData.sizeof, typeid(RData));
    }
  }

  // region data
  // all data is here, so passing region struct around is painless
  static struct RData {
    int rwdt, rhgt; // width and height
    bool simple = true; // is this region a simple one (i.e. rectangular, without holes)?
    bool simpleSolid = true; // if it is simple, is it solid or empty?
    //WARNING! the following arrays should NEVER be shared!
    uint[] lineofs; // line data offset in data[]
    SpanType[] data;
      // data format for each line:
      //   len, data[len-1]
      // line items: list of increasing x coords; each coord marks start of next region
      // all even regions are solid, i.e.
      //   0..line[0]-1: solid region
      //   line[0]..line[1]-1: transparent (empty) region
      //   etc.
      // note that line[$-1] is always rwdt; it's used as sentinel too
      // `line[0] == 0` means that first span is transparent (empty)
      // (i.e. region starts from transparent span)
    int rc = 1; // refcount
  }
}


// ////////////////////////////////////////////////////////////////////////// //
version(sdpy_region_test) {
//static assert(0);
import iv.writer;


void dumpData (ref Region reg) {
  import iv.writer;
  if (reg.rdata.simple) { writeln("simple ", (reg.rdata.simpleSolid ? "solid" : "empty"), " region"); return; }
  foreach (immutable y, uint ofs; reg.rdata.lineofs) {
    if (y > 0 && reg.rdata.lineofs[y-1] == ofs) {
      writefln!"%5s:%3s: ditto"(ofs, y);
    } else {
      writef!"%5s:%3s: len="(ofs, y);
      write(reg.rdata.data[ofs]);
      auto end = ofs+reg.rdata.data[ofs];
      ++ofs;
      while (ofs < end) write("; ", reg.rdata.data[ofs++]);
      writeln;
    }
  }
}


void checkLineOffsets (ref Region reg) {
  if (reg.rdata.simple) return;
  foreach (immutable idx; 1..reg.rdata.lineofs.length) {
    if (reg.rdata.lineofs[idx-1] > reg.rdata.lineofs[idx]) assert(0, "invalid line offset data");
    // check for two equal, but unmerged lines
    if (reg.rdata.lineofs[idx-1] != reg.rdata.lineofs[idx]) {
      import core.stdc.string : memcmp;
      if (reg.rdata.data[reg.rdata.lineofs[idx-1]] == reg.rdata.data[reg.rdata.lineofs[idx]] &&
          memcmp(reg.rdata.data.ptr+reg.rdata.lineofs[idx-1], reg.rdata.data.ptr+reg.rdata.lineofs[idx], reg.SpanType.sizeof*reg.rdata.data[reg.rdata.lineofs[idx]]) == 0)
      {
        dumpData(reg);
        assert(0, "found two identical, but not merged lines");
      }
      if (reg.rdata.data[reg.rdata.lineofs[idx-1]] < 2) assert(0, "bad data (0)");
      if (reg.rdata.data[reg.rdata.lineofs[idx]] < 2) assert(0, "bad data (1)");
    }
  }
}


void buildBitmap (ref Region reg, int[] bmp) {
  if (reg.rdata.simple) {
    bmp[0..reg.width*reg.height] = (reg.rdata.simpleSolid ? 1 : 0);
    return;
  }
  bmp[0..reg.width*reg.height] = 42;
  foreach (immutable y, uint ofs; reg.rdata.lineofs) {
    usize a = y*reg.width;
    usize len = reg.rdata.data[ofs++];
    if (len < 1) assert(0, "invalid span");
    int sx = 0;
    bool solid = true;
    if (reg.rdata.data[ofs] == 0) { solid = false; ++ofs; }
    while (sx != reg.width) {
      // we should not have two consecutive zero-width spans
      if (reg.rdata.data[ofs] == 0 || reg.rdata.data[ofs] <= sx) {
        //foreach (immutable idx; 0..reg.rdata.data.length) if (reg.rdata.data[idx] >= 0) writeln(idx, ": ", reg.rdata.data[idx]); else break;
        //assert(reg.rdata.data[ofs+1] != 0);
        assert(0, "invalid span");
      }
      int ex = reg.rdata.data[ofs++];
      bmp[a+sx..a+ex] = (solid ? 1 : 0);
      solid = !solid;
      sx = ex;
    }
    debug assert(sx == reg.width);
  }
  foreach (immutable v; bmp[0..reg.width*reg.height]) if (v == 42) assert(0, "invalid region data");
}


int[] buildCoords (int[] bmp, int type, int x0, int x1) {
  bool isSolid (int x) { return (x >= 0 && x < bmp.length && bmp[x] != 0); }
  int[] res;
  while (x0 <= x1) {
    while (x0 <= x1 && type != isSolid(x0)) ++x0;
    if (x0 > x1) break;
    res ~= x0; // start
    while (x0 <= x1 && type == isSolid(x0)) ++x0;
    res ~= x0-1;
  }
  return res;
}


void fuzzyEnumerator () {
  import std.random;
  auto reg = Region(uniform!"[]"(1, 128), 1);
  int[] bmp, ebmp;
  bmp.length = reg.width*reg.height;
  ebmp.length = reg.width*reg.height;
  if (uniform!"[]"(0, 1)) {
    reg.rdata.simpleSolid = false;
    ebmp[] = 0; // default is empty
  } else {
    ebmp[] = 1; // default is solid
  }
  foreach (immutable tx0; 0..1000) {
    checkLineOffsets(reg);
    buildBitmap(reg, bmp[]);
    version(sdpy_aggressive_gc) { import core.memory : GC; GC.collect(); }
    debug(region_more_prints) {
      if (1/*bmp[] != ebmp[]*/) {
        assert(bmp.length == ebmp.length);
        writeln;
        foreach (immutable idx; 0..bmp.length) write(bmp[idx]); writeln;
        foreach (immutable idx; 0..ebmp.length) write(ebmp[idx]); writeln;
      }
    }
    assert(bmp[] == ebmp[]);
    foreach (immutable trx; 0..200) {
      //writeln("*");
      int x0 = uniform!"[)"(-10, reg.width+10);
      int x1 = uniform!"[)"(-10, reg.width+10);
      if (x0 > x1) { auto t = x0; x0 = x1; x1 = t; }
      int[] coords;
      int type = uniform!"[]"(0, 1);
      if (type == 0) {
        reg.spans!false(0, x0, x1, (int x0, int x1) { coords ~= x0; coords ~= x1; });
      } else {
        reg.spans!true(0, x0, x1, (int x0, int x1) { coords ~= x0; coords ~= x1; return 0; });
      }
      auto ecr = buildCoords(bmp[], type, x0, x1);
      assert(ecr[] == coords[]);
      // now check enumerator range
      coords.length = 0;
      if (type == 0) {
        foreach (ref pair; reg.spanRange!false(0, x0, x1)) { coords ~= pair.x0; coords ~= pair.x1; }
      } else {
        foreach (ref pair; reg.spanRange!true(0, x0, x1)) { coords ~= pair.x0; coords ~= pair.x1; }
      }
      if (ecr[] != coords[]) {
        import std.stdio : writeln;
        writeln("\ntype=", type);
        writeln("ecr=", ecr);
        writeln("crd=", coords);
      }
      assert(ecr[] == coords[]);
      version(sdpy_aggressive_gc) { import core.memory : GC; GC.collect(); }
    }
    // now do random punch/patch
    {
      int x = uniform!"[)"(0, reg.width);
      int y = uniform!"[)"(0, reg.height);
      int w = uniform!"[]"(0, reg.width);
      int h = uniform!"[]"(0, reg.height);
      int patch = uniform!"[]"(0, 1);
      debug(region_more_prints) { import core.stdc.stdio : printf; printf(":x0=%d; x1=%d; w=%d; solid=%d\n", x, x+w-1, w, patch); }
      if (patch) reg.patch(x, y, w, h); else reg.punch(x, y, w, h);
      version(sdpy_aggressive_gc) { import core.memory : GC; GC.collect(); }
      // fix ebmp
      foreach (int dy; y..y+h) {
        if (dy < 0) continue;
        if (dy >= reg.height) break;
        foreach (int dx; x..x+w) {
          if (dx < 0) continue;
          if (dx >= reg.width) break;
          ebmp[dy*reg.width+dx] = patch;
        }
      }
    }
  }
}


//enum OneSeed = 1586553857;

void main () {
  import iv.writer;
  import std.random;
  foreach (immutable trycount; 0..1000) {
    {
      auto seed = unpredictableSeed;
      static if (is(typeof(OneSeed))) seed = OneSeed;
      rndGen.seed(seed);
      write("try: ", trycount, "; seed = ", seed, " ... ");
    }
    fuzzyEnumerator();
    writeln("OK");
    static if (is(typeof(OneSeed))) break;
  }
}
}