harfbuzz/src/graphite2/src/Collider.cpp

1085 lines
44 KiB
C++

// SPDX-License-Identifier: MIT
// Copyright 2010, SIL International, All rights reserved.
#include <algorithm>
#include <limits>
#include <cmath>
#include <string>
#include <functional>
#include "inc/Collider.h"
#include "inc/Segment.h"
#include "inc/Slot.h"
#include "inc/GlyphCache.h"
#include "inc/Sparse.h"
#define ISQRT2 0.707106781f
// Possible rounding error for subbox boundaries: 0.016 = 1/64 = 1/256 * 4
// (values in font range from 0..256)
// #define SUBBOX_RND_ERR 0.016
using namespace graphite2;
//// SHIFT-COLLIDER ////
// Initialize the Collider to hold the basic movement limits for the
// target slot, the one we are focusing on fixing.
bool ShiftCollider::initSlot(Segment & seg, Slot & aSlot, const Rect &limit, float margin, float marginWeight,
const Position &currShift, const Position &currOffset, int dir, GR_MAYBE_UNUSED json * const dbgout)
{
int i;
float mx, mn;
float a, shift;
const GlyphCache &gc = seg.getFace()->glyphs();
unsigned short gid = aSlot.gid();
if (!gc.check(gid))
return false;
const BBox &bb = gc.getBoundingBBox(gid);
const SlantBox &sb = gc.getBoundingSlantBox(gid);
//float sx = aSlot.origin().x + currShift.x;
//float sy = aSlot.origin().y + currShift.y;
if (currOffset.x != 0.f || currOffset.y != 0.f)
_limit = Rect(limit.bl - currOffset, limit.tr - currOffset);
else
_limit = limit;
// For a ShiftCollider, these indices indicate which vector we are moving by:
// each _ranges represents absolute space with respect to the origin of the slot. Thus take into account true origins but subtract the vmin for the slot
for (i = 0; i < 4; ++i)
{
switch (i) {
case 0 : // x direction
mn = _limit.bl.x + currOffset.x;
mx = _limit.tr.x + currOffset.x;
_len[i] = bb.xa - bb.xi;
a = currOffset.y + currShift.y;
_ranges[i].initialise<XY>(mn, mx, margin, marginWeight, a);
break;
case 1 : // y direction
mn = _limit.bl.y + currOffset.y;
mx = _limit.tr.y + currOffset.y;
_len[i] = bb.ya - bb.yi;
a = currOffset.x + currShift.x;
_ranges[i].initialise<XY>(mn, mx, margin, marginWeight, a);
break;
case 2 : // sum (negatively sloped diagonal boundaries)
// pick closest x,y limit boundaries in s direction
shift = currOffset.x + currOffset.y + currShift.x + currShift.y;
mn = -2 * min(currShift.x - _limit.bl.x, currShift.y - _limit.bl.y) + shift;
mx = 2 * min(_limit.tr.x - currShift.x, _limit.tr.y - currShift.y) + shift;
_len[i] = sb.sa - sb.si;
a = currOffset.x - currOffset.y + currShift.x - currShift.y;
_ranges[i].initialise<SD>(mn, mx, margin / ISQRT2, marginWeight, a);
break;
case 3 : // diff (positively sloped diagonal boundaries)
// pick closest x,y limit boundaries in d direction
shift = currOffset.x - currOffset.y + currShift.x - currShift.y;
mn = -2 * min(currShift.x - _limit.bl.x, _limit.tr.y - currShift.y) + shift;
mx = 2 * min(_limit.tr.x - currShift.x, currShift.y - _limit.bl.y) + shift;
_len[i] = sb.da - sb.di;
a = currOffset.x + currOffset.y + currShift.x + currShift.y;
_ranges[i].initialise<SD>(mn, mx, margin / ISQRT2, marginWeight, a);
break;
}
}
_target = &aSlot;
if ((dir & 1) == 0)
{
// For LTR, switch and negate x limits.
_limit.bl.x = -1 * limit.tr.x;
//_limit.tr.x = -1 * limit.bl.x;
}
_currOffset = currOffset;
_currShift = currShift;
_origin = aSlot.origin() - currOffset; // the original anchor position of the glyph
_margin = margin;
_marginWt = marginWeight;
SlotCollision *c = seg.collisionInfo(aSlot);
_seqClass = c->seqClass();
_seqProxClass = c->seqProxClass();
_seqOrder = c->seqOrder();
return true;
}
template <class O>
float sdm(float vi, float va, float mx, float my, O op)
{
float res = 2 * mx - vi;
if (op(res, vi + 2 * my))
{
res = va + 2 * my;
if (op(res, 2 * mx - va))
res = mx + my;
}
return res;
}
// Mark an area with a cost that can vary along the x or y axis. The region is expressed in terms of the centre of the target glyph in each axis
void ShiftCollider::addBox_slope(bool isx, const Rect &box, const BBox &bb, const SlantBox &sb, const Position &org, float weight, float m, bool minright, int axis)
{
float a, c;
switch (axis) {
case 0 :
if (box.bl.y < org.y + bb.ya && box.tr.y > org.y + bb.yi && box.width() > 0)
{
a = org.y + 0.5f * (bb.yi + bb.ya);
c = 0.5f * (bb.xi + bb.xa);
if (isx)
_ranges[axis].weighted<XY>(box.bl.x - c, box.tr.x - c, weight, a, m,
(minright ? box.tr.x : box.bl.x) - c, a, 0, false);
else
_ranges[axis].weighted<XY>(box.bl.x - c, box.tr.x - c, weight, a, 0, 0, org.y,
m * (a * a + sqr((minright ? box.tr.y : box.bl.y) - 0.5f * (bb.yi + bb.ya))), false);
}
break;
case 1 :
if (box.bl.x < org.x + bb.xa && box.tr.x > org.x + bb.xi && box.height() > 0)
{
a = org.x + 0.5f * (bb.xi + bb.xa);
c = 0.5f * (bb.yi + bb.ya);
if (isx)
_ranges[axis].weighted<XY>(box.bl.y - c, box.tr.y - c, weight, a, 0, 0, org.x,
m * (a * a + sqr((minright ? box.tr.x : box.bl.x) - 0.5f * (bb.xi + bb.xa))), false);
else
_ranges[axis].weighted<XY>(box.bl.y - c, box.tr.y - c, weight, a, m,
(minright ? box.tr.y : box.bl.y) - c, a, 0, false);
}
break;
case 2 :
if (box.bl.x - box.tr.y < org.x - org.y + sb.da && box.tr.x - box.bl.y > org.x - org.y + sb.di)
{
float d = org.x - org.y + 0.5f * (sb.di + sb.da);
c = 0.5f * (sb.si + sb.sa);
float smax = min(2 * box.tr.x - d, 2 * box.tr.y + d);
float smin = max(2 * box.bl.x - d, 2 * box.bl.y + d);
if (smin > smax) return;
float si;
a = d;
if (isx)
si = 2 * (minright ? box.tr.x : box.bl.x) - a;
else
si = 2 * (minright ? box.tr.y : box.bl.y) + a;
_ranges[axis].weighted<SD>(smin - c, smax - c, weight / 2, a, m / 2, si, 0, 0, isx);
}
break;
case 3 :
if (box.bl.x + box.bl.y < org.x + org.y + sb.sa && box.tr.x + box.tr.y > org.x + org.y + sb.si)
{
float s = org.x + org.y + 0.5f * (sb.si + sb.sa);
c = 0.5f * (sb.di + sb.da);
float dmax = min(2 * box.tr.x - s, s - 2 * box.bl.y);
float dmin = max(2 * box.bl.x - s, s - 2 * box.tr.y);
if (dmin > dmax) return;
float di;
a = s;
if (isx)
di = 2 * (minright ? box.tr.x : box.bl.x) - a;
else
di = 2 * (minright ? box.tr.y : box.bl.y) + a;
_ranges[axis].weighted<SD>(dmin - c, dmax - c, weight / 2, a, m / 2, di, 0, 0, !isx);
}
break;
default :
break;
}
return;
}
// Mark an area with an absolute cost, making it completely inaccessible.
inline void ShiftCollider::removeBox(const Rect &box, const BBox &bb, const SlantBox &sb, const Position &org, int axis)
{
float c;
switch (axis) {
case 0 :
if (box.bl.y < org.y + bb.ya && box.tr.y > org.y + bb.yi && box.width() > 0)
{
c = 0.5f * (bb.xi + bb.xa);
_ranges[axis].exclude(box.bl.x - c, box.tr.x - c);
}
break;
case 1 :
if (box.bl.x < org.x + bb.xa && box.tr.x > org.x + bb.xi && box.height() > 0)
{
c = 0.5f * (bb.yi + bb.ya);
_ranges[axis].exclude(box.bl.y - c, box.tr.y - c);
}
break;
case 2 :
if (box.bl.x - box.tr.y < org.x - org.y + sb.da && box.tr.x - box.bl.y > org.x - org.y + sb.di
&& box.width() > 0 && box.height() > 0)
{
float di = org.x - org.y + sb.di;
float da = org.x - org.y + sb.da;
float smax = sdm(di, da, box.tr.x, box.tr.y, std::greater<float>());
float smin = sdm(da, di, box.bl.x, box.bl.y, std::less<float>());
c = 0.5f * (sb.si + sb.sa);
_ranges[axis].exclude(smin - c, smax - c);
}
break;
case 3 :
if (box.bl.x + box.bl.y < org.x + org.y + sb.sa && box.tr.x + box.tr.y > org.x + org.y + sb.si
&& box.width() > 0 && box.height() > 0)
{
float si = org.x + org.y + sb.si;
float sa = org.x + org.y + sb.sa;
float dmax = sdm(si, sa, box.tr.x, -box.bl.y, std::greater<float>());
float dmin = sdm(sa, si, box.bl.x, -box.tr.y, std::less<float>());
c = 0.5f * (sb.di + sb.da);
_ranges[axis].exclude(dmin - c, dmax - c);
}
break;
default :
break;
}
return;
}
// Adjust the movement limits for the target to avoid having it collide
// with the given neighbor slot. Also determine if there is in fact a collision
// between the target and the given slot.
bool ShiftCollider::mergeSlot(Segment & seg, Slot & slot, const SlotCollision *cslot, const Position &currShift,
bool isAfter, // slot is logically after _target
bool sameCluster, bool &hasCol, bool isExclusion,
GR_MAYBE_UNUSED json * const dbgout )
{
bool isCol = false;
const float sx = slot.origin().x - _origin.x + currShift.x;
const float sy = slot.origin().y - _origin.y + currShift.y;
const float sd = sx - sy;
const float ss = sx + sy;
float vmin, vmax;
float omin, omax, otmin, otmax;
float cmin, cmax; // target limits
float torg;
const GlyphCache &gc = seg.getFace()->glyphs();
auto const gid = slot.gid();
if (!gc.check(gid))
return false;
const BBox &bb = gc.getBoundingBBox(gid);
// SlotCollision * cslot = seg.collisionInfo(slot);
int orderFlags = 0;
bool sameClass = _seqProxClass == 0 && cslot->seqClass() == _seqClass;
if (sameCluster && _seqClass
&& (sameClass || (_seqProxClass != 0 && cslot->seqClass() == _seqProxClass)))
// Force the target glyph to be in the specified direction from the slot we're testing.
orderFlags = _seqOrder;
// short circuit if only interested in direct collision and we are out of range
if (orderFlags || (sx + bb.xa + _margin >= _limit.bl.x && sx + bb.xi - _margin <= _limit.tr.x)
|| (sy + bb.ya + _margin >= _limit.bl.y && sy + bb.yi - _margin <= _limit.tr.y))
{
const float tx = _currOffset.x + _currShift.x;
const float ty = _currOffset.y + _currShift.y;
const float td = tx - ty;
const float ts = tx + ty;
const SlantBox &sb = gc.getBoundingSlantBox(gid);
const unsigned short tgid = _target->gid();
const BBox &tbb = gc.getBoundingBBox(tgid);
const SlantBox &tsb = gc.getBoundingSlantBox(tgid);
float seq_above_wt = cslot->seqAboveWt();
float seq_below_wt = cslot->seqBelowWt();
float seq_valign_wt = cslot->seqValignWt();
float lmargin;
// if isAfter, invert orderFlags for diagonal orders.
if (isAfter)
{
// invert appropriate bits
orderFlags ^= (sameClass ? 0x3F : 0x3);
// consider 2 bits at a time, non overlapping. If both bits set, clear them
orderFlags = orderFlags ^ ((((orderFlags >> 1) & orderFlags) & 0x15) * 3);
}
#if !defined GRAPHITE2_NTRACING
if (dbgout)
dbgout->setenv(0, &slot);
#endif
// Process main bounding octabox.
for (int i = 0; i < 4; ++i)
{
switch (i) {
case 0 : // x direction
vmin = max(max(bb.xi - tbb.xa + sx, sb.di - tsb.da + ty + sd), sb.si - tsb.sa - ty + ss);
vmax = min(min(bb.xa - tbb.xi + sx, sb.da - tsb.di + ty + sd), sb.sa - tsb.si - ty + ss);
otmin = tbb.yi + ty;
otmax = tbb.ya + ty;
omin = bb.yi + sy;
omax = bb.ya + sy;
torg = _currOffset.x;
cmin = _limit.bl.x + torg;
cmax = _limit.tr.x - tbb.xi + tbb.xa + torg;
lmargin = _margin;
break;
case 1 : // y direction
vmin = max(max(bb.yi - tbb.ya + sy, tsb.di - sb.da + tx - sd), sb.si - tsb.sa - tx + ss);
vmax = min(min(bb.ya - tbb.yi + sy, tsb.da - sb.di + tx - sd), sb.sa - tsb.si - tx + ss);
otmin = tbb.xi + tx;
otmax = tbb.xa + tx;
omin = bb.xi + sx;
omax = bb.xa + sx;
torg = _currOffset.y;
cmin = _limit.bl.y + torg;
cmax = _limit.tr.y - tbb.yi + tbb.ya + torg;
lmargin = _margin;
break;
case 2 : // sum - moving along the positively-sloped vector, so the boundaries are the
// negatively-sloped boundaries.
vmin = max(max(sb.si - tsb.sa + ss, 2 * (bb.yi - tbb.ya + sy) + td), 2 * (bb.xi - tbb.xa + sx) - td);
vmax = min(min(sb.sa - tsb.si + ss, 2 * (bb.ya - tbb.yi + sy) + td), 2 * (bb.xa - tbb.xi + sx) - td);
otmin = tsb.di + td;
otmax = tsb.da + td;
omin = sb.di + sd;
omax = sb.da + sd;
torg = _currOffset.x + _currOffset.y;
cmin = _limit.bl.x + _limit.bl.y + torg;
cmax = _limit.tr.x + _limit.tr.y - tsb.si + tsb.sa + torg;
lmargin = _margin / ISQRT2;
break;
case 3 : // diff - moving along the negatively-sloped vector, so the boundaries are the
// positively-sloped boundaries.
vmin = max(max(sb.di - tsb.da + sd, 2 * (bb.xi - tbb.xa + sx) - ts), -2 * (bb.ya - tbb.yi + sy) + ts);
vmax = min(min(sb.da - tsb.di + sd, 2 * (bb.xa - tbb.xi + sx) - ts), -2 * (bb.yi - tbb.ya + sy) + ts);
otmin = tsb.si + ts;
otmax = tsb.sa + ts;
omin = sb.si + ss;
omax = sb.sa + ss;
torg = _currOffset.x - _currOffset.y;
cmin = _limit.bl.x - _limit.tr.y + torg;
cmax = _limit.tr.x - _limit.bl.y - tsb.di + tsb.da + torg;
lmargin = _margin / ISQRT2;
break;
default :
continue;
}
#if !defined GRAPHITE2_NTRACING
if (dbgout)
dbgout->setenv(1, reinterpret_cast<void *>(-1));
#define DBGTAG(x) if (dbgout) dbgout->setenv(1, reinterpret_cast<void *>(-x));
#else
#define DBGTAG(x)
#endif
if (orderFlags)
{
Position org(tx, ty);
float xminf = _limit.bl.x + _currOffset.x + tbb.xi;
float xpinf = _limit.tr.x + _currOffset.x + tbb.xa;
float ypinf = _limit.tr.y + _currOffset.y + tbb.ya;
float yminf = _limit.bl.y + _currOffset.y + tbb.yi;
switch (orderFlags) {
case SlotCollision::SEQ_ORDER_RIGHTUP :
{
float r1Xedge = cslot->seqAboveXoff() + 0.5f * (bb.xi + bb.xa) + sx;
float r3Xedge = cslot->seqBelowXlim() + bb.xa + sx + 0.5f * (tbb.xa - tbb.xi);
float r2Yedge = 0.5f * (bb.yi + bb.ya) + sy;
// DBGTAG(1x) means the regions are up and right
// region 1
DBGTAG(11)
addBox_slope(true, Rect(Position(xminf, r2Yedge), Position(r1Xedge, ypinf)),
tbb, tsb, org, 0, seq_above_wt, true, i);
// region 2
DBGTAG(12)
removeBox(Rect(Position(xminf, yminf), Position(r3Xedge, r2Yedge)), tbb, tsb, org, i);
// region 3, which end is zero is irrelevant since m weight is 0
DBGTAG(13)
addBox_slope(true, Rect(Position(r3Xedge, yminf), Position(xpinf, r2Yedge - cslot->seqValignHt())),
tbb, tsb, org, seq_below_wt, 0, true, i);
// region 4
DBGTAG(14)
addBox_slope(false, Rect(Position(sx + bb.xi, r2Yedge), Position(xpinf, r2Yedge + cslot->seqValignHt())),
tbb, tsb, org, 0, seq_valign_wt, true, i);
// region 5
DBGTAG(15)
addBox_slope(false, Rect(Position(sx + bb.xi, r2Yedge - cslot->seqValignHt()), Position(xpinf, r2Yedge)),
tbb, tsb, org, seq_below_wt, seq_valign_wt, false, i);
break;
}
case SlotCollision::SEQ_ORDER_LEFTDOWN :
{
float r1Xedge = 0.5f * (bb.xi + bb.xa) + cslot->seqAboveXoff() + sx;
float r3Xedge = bb.xi - cslot->seqBelowXlim() + sx - 0.5f * (tbb.xa - tbb.xi);
float r2Yedge = 0.5f * (bb.yi + bb.ya) + sy;
// DBGTAG(2x) means the regions are up and right
// region 1
DBGTAG(21)
addBox_slope(true, Rect(Position(r1Xedge, yminf), Position(xpinf, r2Yedge)),
tbb, tsb, org, 0, seq_above_wt, false, i);
// region 2
DBGTAG(22)
removeBox(Rect(Position(r3Xedge, r2Yedge), Position(xpinf, ypinf)), tbb, tsb, org, i);
// region 3
DBGTAG(23)
addBox_slope(true, Rect(Position(xminf, r2Yedge - cslot->seqValignHt()), Position(r3Xedge, ypinf)),
tbb, tsb, org, seq_below_wt, 0, false, i);
// region 4
DBGTAG(24)
addBox_slope(false, Rect(Position(xminf, r2Yedge), Position(sx + bb.xa, r2Yedge + cslot->seqValignHt())),
tbb, tsb, org, 0, seq_valign_wt, true, i);
// region 5
DBGTAG(25)
addBox_slope(false, Rect(Position(xminf, r2Yedge - cslot->seqValignHt()),
Position(sx + bb.xa, r2Yedge)), tbb, tsb, org, seq_below_wt, seq_valign_wt, false, i);
break;
}
case SlotCollision::SEQ_ORDER_NOABOVE : // enforce neighboring glyph being above
DBGTAG(31);
removeBox(Rect(Position(bb.xi - tbb.xa + sx, sy + bb.ya),
Position(bb.xa - tbb.xi + sx, ypinf)), tbb, tsb, org, i);
break;
case SlotCollision::SEQ_ORDER_NOBELOW : // enforce neighboring glyph being below
DBGTAG(32);
removeBox(Rect(Position(bb.xi - tbb.xa + sx, yminf),
Position(bb.xa - tbb.xi + sx, sy + bb.yi)), tbb, tsb, org, i);
break;
case SlotCollision::SEQ_ORDER_NOLEFT : // enforce neighboring glyph being to the left
DBGTAG(33)
removeBox(Rect(Position(xminf, bb.yi - tbb.ya + sy),
Position(bb.xi - tbb.xa + sx, bb.ya - tbb.yi + sy)), tbb, tsb, org, i);
break;
case SlotCollision::SEQ_ORDER_NORIGHT : // enforce neighboring glyph being to the right
DBGTAG(34)
removeBox(Rect(Position(bb.xa - tbb.xi + sx, bb.yi - tbb.ya + sy),
Position(xpinf, bb.ya - tbb.yi + sy)), tbb, tsb, org, i);
break;
default :
break;
}
}
if (vmax < cmin - lmargin || vmin > cmax + lmargin || omax < otmin - lmargin || omin > otmax + lmargin)
continue;
// Process sub-boxes that are defined for this glyph.
// We only need to do this if there was in fact a collision with the main octabox.
uint8_t numsub = gc.numSubBounds(gid);
if (numsub > 0)
{
bool anyhits = false;
for (int j = 0; j < numsub; ++j)
{
const BBox &sbb = gc.getSubBoundingBBox(gid, j);
const SlantBox &ssb = gc.getSubBoundingSlantBox(gid, j);
switch (i) {
case 0 : // x
vmin = max(max(sbb.xi-tbb.xa+sx, ssb.di-tsb.da+sd+ty), ssb.si-tsb.sa+ss-ty);
vmax = min(min(sbb.xa-tbb.xi+sx, ssb.da-tsb.di+sd+ty), ssb.sa-tsb.si+ss-ty);
omin = sbb.yi + sy;
omax = sbb.ya + sy;
break;
case 1 : // y
vmin = max(max(sbb.yi-tbb.ya+sy, tsb.di-ssb.da-sd+tx), ssb.si-tsb.sa+ss-tx);
vmax = min(min(sbb.ya-tbb.yi+sy, tsb.da-ssb.di-sd+tx), ssb.sa-tsb.si+ss-tx);
omin = sbb.xi + sx;
omax = sbb.xa + sx;
break;
case 2 : // sum
vmin = max(max(ssb.si-tsb.sa+ss, 2*(sbb.yi-tbb.ya+sy)+td), 2*(sbb.xi-tbb.xa+sx)-td);
vmax = min(min(ssb.sa-tsb.si+ss, 2*(sbb.ya-tbb.yi+sy)+td), 2*(sbb.xa-tbb.xi+sx)-td);
omin = ssb.di + sd;
omax = ssb.da + sd;
break;
case 3 : // diff
vmin = max(max(ssb.di-tsb.da+sd, 2*(sbb.xi-tbb.xa+sx)-ts), -2*(sbb.ya-tbb.yi+sy)+ts);
vmax = min(min(ssb.da-tsb.di+sd, 2*(sbb.xa-tbb.xi+sx)-ts), -2*(sbb.yi-tbb.ya+sy)+ts);
omin = ssb.si + ss;
omax = ssb.sa + ss;
break;
}
if (vmax < cmin - lmargin || vmin > cmax + lmargin || omax < otmin - lmargin || omin > otmax + lmargin)
continue;
#if !defined GRAPHITE2_NTRACING
if (dbgout)
dbgout->setenv(1, reinterpret_cast<void *>(j));
#endif
if (omin > otmax)
_ranges[i].weightedAxis(i, vmin - lmargin, vmax + lmargin, 0, 0, 0, 0, 0,
sqr(lmargin - omin + otmax) * _marginWt, false);
else if (omax < otmin)
_ranges[i].weightedAxis(i, vmin - lmargin, vmax + lmargin, 0, 0, 0, 0, 0,
sqr(lmargin - otmin + omax) * _marginWt, false);
else
_ranges[i].exclude_with_margins(vmin, vmax, i);
anyhits = true;
}
if (anyhits)
isCol = true;
}
else // no sub-boxes
{
#if !defined GRAPHITE2_NTRACING
if (dbgout)
dbgout->setenv(1, reinterpret_cast<void *>(-1));
#endif
isCol = true;
if (omin > otmax)
_ranges[i].weightedAxis(i, vmin - lmargin, vmax + lmargin, 0, 0, 0, 0, 0,
sqr(lmargin - omin + otmax) * _marginWt, false);
else if (omax < otmin)
_ranges[i].weightedAxis(i, vmin - lmargin, vmax + lmargin, 0, 0, 0, 0, 0,
sqr(lmargin - otmin + omax) * _marginWt, false);
else
_ranges[i].exclude_with_margins(vmin, vmax, i);
}
}
}
bool res = true;
if (cslot->exclGlyph() > 0 && gc.check(cslot->exclGlyph()) && !isExclusion)
{
// Set up the bogus slot representing the exclusion glyph.
Slot exclSlot;
exclSlot.glyph(seg, cslot->exclGlyph());
exclSlot.position_shift(slot.origin() + cslot->exclOffset());
SlotCollision exclInfo(seg, exclSlot);
res &= mergeSlot(seg, exclSlot, &exclInfo, currShift, isAfter, sameCluster, isCol, true, dbgout);
}
hasCol |= isCol;
return res;
} // end of ShiftCollider::mergeSlot
// Figure out where to move the target glyph to, and return the amount to shift by.
Position ShiftCollider::resolve(GR_MAYBE_UNUSED Segment &seg, bool &isCol, GR_MAYBE_UNUSED json * const dbgout)
{
float tbase;
float totalCost = (float)(std::numeric_limits<float>::max() / 2);
Position resultPos = Position(0, 0);
#if !defined GRAPHITE2_NTRACING
int bestAxis = -1;
if (dbgout)
{
outputJsonDbgStartSlot(dbgout, seg);
*dbgout << "vectors" << json::array;
}
#endif
isCol = true;
for (int i = 0; i < 4; ++i)
{
float bestCost = -1;
float bestPos;
// Calculate the margin depending on whether we are moving diagonally or not:
switch (i) {
case 0 : // x direction
tbase = _currOffset.x;
break;
case 1 : // y direction
tbase = _currOffset.y;
break;
case 2 : // sum (negatively-sloped diagonals)
tbase = _currOffset.x + _currOffset.y;
break;
case 3 : // diff (positively-sloped diagonals)
tbase = _currOffset.x - _currOffset.y;
break;
}
Position testp;
bestPos = _ranges[i].closest(0, bestCost) - tbase; // Get the best relative position
#if !defined GRAPHITE2_NTRACING
if (dbgout)
outputJsonDbgOneVector(dbgout, seg, i, tbase, bestCost, bestPos) ;
#endif
if (bestCost >= 0.0f)
{
isCol = false;
switch (i) {
case 0 : testp = Position(bestPos, _currShift.y); break;
case 1 : testp = Position(_currShift.x, bestPos); break;
case 2 : testp = Position(0.5f * (_currShift.x - _currShift.y + bestPos), 0.5f * (_currShift.y - _currShift.x + bestPos)); break;
case 3 : testp = Position(0.5f * (_currShift.x + _currShift.y + bestPos), 0.5f * (_currShift.x + _currShift.y - bestPos)); break;
}
if (bestCost < totalCost - 0.01f)
{
totalCost = bestCost;
resultPos = testp;
#if !defined GRAPHITE2_NTRACING
bestAxis = i;
#endif
}
}
} // end of loop over 4 directions
#if !defined GRAPHITE2_NTRACING
if (dbgout)
outputJsonDbgEndSlot(dbgout, resultPos, bestAxis, isCol);
#endif
return resultPos;
} // end of ShiftCollider::resolve
#if !defined GRAPHITE2_NTRACING
void ShiftCollider::outputJsonDbg(json * const dbgout, Segment & seg, int axis)
{
int axisMax = axis;
if (axis < 0) // output all axes
{
*dbgout << "gid" << _target->gid()
<< "limit" << _limit
<< "target" << json::object
<< "origin" << _target->origin()
<< "margin" << _margin
<< "bbox" << seg.theGlyphBBoxTemporary(_target->gid())
<< "slantbox" << seg.getFace()->glyphs().slant(_target->gid())
<< json::close; // target object
*dbgout << "ranges" << json::array;
axis = 0;
axisMax = 3;
}
for (int iAxis = axis; iAxis <= axisMax; ++iAxis)
{
*dbgout << json::flat << json::array << _ranges[iAxis].position();
for (Zones::const_iterator s = _ranges[iAxis].begin(), e = _ranges[iAxis].end(); s != e; ++s)
*dbgout << json::flat << json::array
<< Position(s->x, s->xm) << s->sm << s->smx << s->c
<< json::close;
*dbgout << json::close;
}
if (axis < axisMax) // looped through the _ranges array for all axes
*dbgout << json::close; // ranges array
}
void ShiftCollider::outputJsonDbgStartSlot(json * const dbgout, Segment &seg)
{
*dbgout << json::object // slot - not closed till the end of the caller method
<< "slot" << objectid(SlotBuffer::const_iterator::from(_target))
<< "gid" << _target->gid()
<< "limit" << _limit
<< "target" << json::object
<< "origin" << _origin
<< "currShift" << _currShift
<< "currOffset" << seg.collisionInfo(*_target)->offset()
<< "bbox" << seg.theGlyphBBoxTemporary(_target->gid())
<< "slantBox" << seg.getFace()->glyphs().slant(_target->gid())
<< "fix" << "shift";
*dbgout << json::close; // target object
}
void ShiftCollider::outputJsonDbgEndSlot(GR_MAYBE_UNUSED json * const dbgout,
Position resultPos, int bestAxis, bool isCol)
{
*dbgout << json::close // vectors array
<< "result" << resultPos
//<< "scraping" << _scraping[bestAxis]
<< "bestAxis" << bestAxis
<< "stillBad" << isCol
<< json::close; // slot object
}
void ShiftCollider::outputJsonDbgOneVector(json * const dbgout, Segment &seg, int axis,
float tleft, float bestCost, float bestVal)
{
const char * label;
switch (axis)
{
case 0: label = "x"; break;
case 1: label = "y"; break;
case 2: label = "sum (NE-SW)"; break;
case 3: label = "diff (NW-SE)"; break;
default: label = "???"; break;
}
*dbgout << json::object // vector
<< "direction" << label
<< "targetMin" << tleft;
outputJsonDbgRemovals(dbgout, axis, seg);
*dbgout << "ranges";
outputJsonDbg(dbgout, seg, axis);
*dbgout << "bestCost" << bestCost
<< "bestVal" << bestVal + tleft
<< json::close; // vectors object
}
void ShiftCollider::outputJsonDbgRemovals(json * const dbgout, int axis, Segment &seg)
{
*dbgout << "removals" << json::array;
_ranges[axis].jsonDbgOut(seg);
*dbgout << json::close; // removals array
}
#endif // !defined GRAPHITE2_NTRACING
//// KERN-COLLIDER ////
inline
static float localmax (float al, float au, float bl, float bu, float x)
{
if (al < bl)
{ if (au < bu) return au < x ? au : x; }
else if (au > bu) return bl < x ? bl : x;
return x;
}
inline
static float localmin(float al, float au, float bl, float bu, float x)
{
if (bl > al)
{ if (bu > au) return bl > x ? bl : x; }
else if (au > bu) return al > x ? al : x;
return x;
}
// Return the given edge of the glyph at height y, taking any slant box into account.
static float get_edge(Segment & seg, const Slot & s, const Position &shift, float y, float width, float margin, bool isRight)
{
const GlyphCache &gc = seg.getFace()->glyphs();
unsigned short gid = s.gid();
float sx = s.origin().x + shift.x;
float sy = s.origin().y + shift.y;
uint8_t numsub = gc.numSubBounds(gid);
float res = isRight ? (float)-1e38 : (float)1e38;
if (numsub > 0)
{
for (int i = 0; i < numsub; ++i)
{
const BBox &sbb = gc.getSubBoundingBBox(gid, i);
const SlantBox &ssb = gc.getSubBoundingSlantBox(gid, i);
if (sy + sbb.yi - margin > y + width / 2 || sy + sbb.ya + margin < y - width / 2)
continue;
if (isRight)
{
float x = sx + sbb.xa + margin;
if (x > res)
{
float td = sx - sy + ssb.da + margin + y;
float ts = sx + sy + ssb.sa + margin - y;
x = localmax(td - width / 2, td + width / 2, ts - width / 2, ts + width / 2, x);
if (x > res)
res = x;
}
}
else
{
float x = sx + sbb.xi - margin;
if (x < res)
{
float td = sx - sy + ssb.di - margin + y;
float ts = sx + sy + ssb.si - margin - y;
x = localmin(td - width / 2, td + width / 2, ts - width / 2, ts + width / 2, x);
if (x < res)
res = x;
}
}
}
}
else
{
const BBox &bb = gc.getBoundingBBox(gid);
const SlantBox &sb = gc.getBoundingSlantBox(gid);
if (sy + bb.yi - margin > y + width / 2 || sy + bb.ya + margin < y - width / 2)
return res;
float td = sx - sy + y;
float ts = sx + sy - y;
if (isRight)
res = localmax(td + sb.da - width / 2, td + sb.da + width / 2, ts + sb.sa - width / 2, ts + sb.sa + width / 2, sx + bb.xa) + margin;
else
res = localmin(td + sb.di - width / 2, td + sb.di + width / 2, ts + sb.si - width / 2, ts + sb.si + width / 2, sx + bb.xi) - margin;
}
return res;
}
bool KernCollider::initSlot(Segment & seg, Slot & aSlot, const Rect &limit, float margin,
const Position &currShift, const Position &offsetPrev, int dir,
float ymin, float ymax, GR_MAYBE_UNUSED json * const dbgout)
{
auto & gc = seg.getFace()->glyphs();
auto const * base = aSlot.base();
int numSlices;
if (margin < 10) margin = 10;
_limit = limit;
_offsetPrev = offsetPrev; // kern from a previous pass
// Calculate the height of the glyph and how many horizontal slices to use.
if (_maxy >= 1e37f)
{
_sliceWidth = margin / 1.5f;
_maxy = ymax + margin;
_miny = ymin - margin;
numSlices = int((_maxy - _miny + 2) / (_sliceWidth / 1.5f) + 1.f); // +2 helps with rounding errors
_edges.clear();
_edges.insert(_edges.begin(), numSlices, (dir & 1) ? 1e38f : -1e38f);
_xbound = (dir & 1) ? (float)1e38f : (float)-1e38f;
}
else if (_maxy != ymax || _miny != ymin)
{
if (_miny != ymin)
{
numSlices = int((ymin - margin - _miny) / _sliceWidth - 1);
_miny += numSlices * _sliceWidth;
if (numSlices < 0)
_edges.insert(_edges.begin(), -numSlices, (dir & 1) ? 1e38f : -1e38f);
else if ((unsigned)numSlices < _edges.size()) // this shouldn't fire since we always grow the range
{
vector<float>::iterator e = _edges.begin();
while (numSlices--)
++e;
_edges.erase(_edges.begin(), e);
}
}
if (_maxy != ymax)
{
numSlices = int((ymax + margin - _miny) / _sliceWidth + 1);
_maxy = numSlices * _sliceWidth + _miny;
if (numSlices > (int)_edges.size())
_edges.insert(_edges.end(), numSlices - _edges.size(), (dir & 1) ? 1e38f : -1e38f);
else if (numSlices < (int)_edges.size()) // this shouldn't fire since we always grow the range
{
while ((int)_edges.size() > numSlices)
_edges.pop_back();
}
}
goto done;
}
numSlices = int(_edges.size());
#if !defined GRAPHITE2_NTRACING
// Debugging
_seg = &seg;
_slotNear.clear();
_slotNear.insert(_slotNear.begin(), numSlices, NULL);
_nearEdges.clear();
_nearEdges.insert(_nearEdges.begin(), numSlices, (dir & 1) ? -1e38f : +1e38f);
#endif
// Determine the trailing edge of each slice (ie, left edge for a RTL glyph).
for (auto s = base->cluster(), end = base->end(); s != end; ++s)
{
SlotCollision *c = seg.collisionInfo(*s);
if (!gc.check(s->gid()))
return false;
const BBox &bs = gc.getBoundingBBox(s->gid());
float x = s->origin().x + c->shift().x + ((dir & 1) ? bs.xi : bs.xa);
// Loop over slices.
// Note smin might not be zero if glyph s is not at the bottom of the cluster; similarly for smax.
float toffset = c->shift().y - _miny + 1 + s->origin().y;
int smin = max(0, int((bs.yi + toffset) / _sliceWidth));
int smax = min(numSlices - 1, int((bs.ya + toffset) / _sliceWidth + 1));
for (int i = smin; i <= smax; ++i)
{
float t;
float y = _miny - 1 + (i + .5f) * _sliceWidth; // vertical center of slice
if ((dir & 1) && x < _edges[i])
{
t = get_edge(seg, *s, c->shift(), y, _sliceWidth, margin, false);
if (t < _edges[i])
{
_edges[i] = t;
if (t < _xbound)
_xbound = t;
}
}
else if (!(dir & 1) && x > _edges[i])
{
t = get_edge(seg, *s, c->shift(), y, _sliceWidth, margin, true);
if (t > _edges[i])
{
_edges[i] = t;
if (t > _xbound)
_xbound = t;
}
}
}
}
done:
_mingap = (float)1e37; // less than 1e38 s.t. 1e38-_mingap is really big
_target = &aSlot;
_margin = margin;
_currShift = currShift;
return true;
} // end of KernCollider::initSlot
// Determine how much the target slot needs to kern away from the given slot.
// In other words, merge information from given slot's position with what the target slot knows
// about how it can kern.
// Return false if we know there is no collision, true if we think there might be one.
bool KernCollider::mergeSlot(Segment & seg, Slot & slot, const Position &currShift, float currSpace, int dir, GR_MAYBE_UNUSED json * const dbgout)
{
int rtl = (dir & 1) * 2 - 1;
if (!seg.getFace()->glyphs().check(slot.gid()))
return false;
const Rect &bb = seg.theGlyphBBoxTemporary(slot.gid());
const float sx = slot.origin().x + currShift.x;
float x = (sx + (rtl > 0 ? bb.tr.x : bb.bl.x)) * rtl;
// this isn't going to reduce _mingap so skip
if (_hit && x < rtl * (_xbound - _mingap - currSpace))
return false;
const float sy = slot.origin().y + currShift.y;
int smin = max(1, int((bb.bl.y + (1 - _miny + sy)) / _sliceWidth + 1)) - 1;
int smax = min((int)_edges.size() - 2, int((bb.tr.y + (1 - _miny + sy)) / _sliceWidth + 1)) + 1;
if (smin > smax)
return false;
bool collides = false;
bool nooverlap = true;
for (int i = smin; i <= smax; ++i)
{
float here = _edges[i] * rtl;
if (here > (float)9e37)
continue;
if (!_hit || x > here - _mingap - currSpace)
{
float y = (float)(_miny - 1 + (i + .5f) * _sliceWidth); // vertical center of slice
// 2 * currSpace to account for the space that is already separating them and the space we want to add
float m = get_edge(seg, slot, currShift, y, _sliceWidth, 0., rtl > 0) * rtl + 2 * currSpace;
if (m < (float)-8e37) // only true if the glyph has a gap in it
continue;
nooverlap = false;
float t = here - m;
// _mingap is positive to shrink
if (t < _mingap || (!_hit && !collides))
{
_mingap = t;
collides = true;
}
#if !defined GRAPHITE2_NTRACING
// Debugging - remember the closest neighboring edge for this slice.
if (m > rtl * _nearEdges[i])
{
_slotNear[i] = &slot;
_nearEdges[i] = m * rtl;
}
#endif
}
else
nooverlap = false;
}
if (nooverlap)
_mingap = max(_mingap, _xbound - rtl * (currSpace + _margin + x));
if (collides && !nooverlap)
_hit = true;
return collides | nooverlap; // note that true is not a necessarily reliable value
} // end of KernCollider::mergeSlot
// Return the amount to kern by.
Position KernCollider::resolve(GR_MAYBE_UNUSED Segment & seg, GR_MAYBE_UNUSED Slot & slot,
int dir, GR_MAYBE_UNUSED json * const dbgout)
{
float resultNeeded = (1 - 2 * (dir & 1)) * _mingap;
// float resultNeeded = (1 - 2 * (dir & 1)) * (_mingap - margin);
float result = min(_limit.tr.x - _offsetPrev.x, max(resultNeeded, _limit.bl.x - _offsetPrev.x));
#if !defined GRAPHITE2_NTRACING
if (dbgout)
{
*dbgout << json::object // slot
<< "slot" << objectid(SlotBuffer::const_iterator::from(_target))
<< "gid" << _target->gid()
<< "limit" << _limit
<< "miny" << _miny
<< "maxy" << _maxy
<< "slicewidth" << _sliceWidth
<< "target" << json::object
<< "origin" << _target->origin()
//<< "currShift" << _currShift
<< "offsetPrev" << _offsetPrev
<< "bbox" << seg.theGlyphBBoxTemporary(_target->gid())
<< "slantBox" << seg.getFace()->glyphs().slant(_target->gid())
<< "fix" << "kern"
<< json::close; // target object
*dbgout << "slices" << json::array;
for (int is = 0; is < (int)_edges.size(); is++)
{
*dbgout << json::flat << json::object
<< "i" << is
<< "targetEdge" << _edges[is]
<< "neighbor" << objectid(SlotBuffer::const_iterator::from(_slotNear[is]))
<< "nearEdge" << _nearEdges[is]
<< json::close;
}
*dbgout << json::close; // slices array
*dbgout
<< "xbound" << _xbound
<< "minGap" << _mingap
<< "needed" << resultNeeded
<< "result" << result
<< "stillBad" << (result != resultNeeded)
<< json::close; // slot object
}
#endif
return Position(result, 0.);
} // end of KernCollider::resolve
void KernCollider::shift(const Position &mv, int dir)
{
for (vector<float>::iterator e = _edges.begin(); e != _edges.end(); ++e)
*e += mv.x;
_xbound += (1 - 2 * (dir & 1)) * mv.x;
}
//// SLOT-COLLISION ////
// Initialize the collision attributes for the given slot.
SlotCollision::SlotCollision(Segment &seg, Slot &slot)
{
initFromSlot(seg, slot);
}
void SlotCollision::initFromSlot(Segment &seg, Slot &slot)
{
// Initialize slot attributes from glyph attributes.
// The order here must match the order in the grcompiler code,
// GrcSymbolTable::AssignInternalGlyphAttrIDs.
uint16_t gid = slot.gid();
uint16_t aCol = seg.silf()->aCollision(); // flags attr ID
const GlyphFace * glyphFace = seg.getFace()->glyphs().glyphSafe(gid);
if (!glyphFace)
return;
const sparse &p = glyphFace->attrs();
_flags = p[aCol];
_limit = Rect(Position(int16_t(p[aCol+1]), int16_t(p[aCol+2])),
Position(int16_t(p[aCol+3]), int16_t(p[aCol+4])));
_margin = p[aCol+5];
_marginWt = p[aCol+6];
_seqClass = p[aCol+7];
_seqProxClass = p[aCol+8];
_seqOrder = p[aCol+9];
_seqAboveXoff = p[aCol+10];
_seqAboveWt = p[aCol+11];
_seqBelowXlim = p[aCol+12];
_seqBelowWt = p[aCol+13];
_seqValignHt = p[aCol+14];
_seqValignWt = p[aCol+15];
// These attributes do not have corresponding glyph attribute:
_exclGlyph = 0;
_exclOffset = Position(0, 0);
}
float SlotCollision::getKern(int dir) const
{
if ((_flags & SlotCollision::COLL_KERN) != 0)
return float(_shift.x * ((dir & 1) ? -1 : 1));
else
return 0;
}
bool SlotCollision::ignore() const
{
return ((flags() & SlotCollision::COLL_IGNORE) || (flags() & SlotCollision::COLL_ISSPACE));
}