gfx/graphite2/src/Silf.cpp

changeset 0
6474c204b198
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/gfx/graphite2/src/Silf.cpp	Wed Dec 31 06:09:35 2014 +0100
     1.3 @@ -0,0 +1,415 @@
     1.4 +/*  GRAPHITE2 LICENSING
     1.5 +
     1.6 +    Copyright 2010, SIL International
     1.7 +    All rights reserved.
     1.8 +
     1.9 +    This library is free software; you can redistribute it and/or modify
    1.10 +    it under the terms of the GNU Lesser General Public License as published
    1.11 +    by the Free Software Foundation; either version 2.1 of License, or
    1.12 +    (at your option) any later version.
    1.13 +
    1.14 +    This program is distributed in the hope that it will be useful,
    1.15 +    but WITHOUT ANY WARRANTY; without even the implied warranty of
    1.16 +    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    1.17 +    Lesser General Public License for more details.
    1.18 +
    1.19 +    You should also have received a copy of the GNU Lesser General Public
    1.20 +    License along with this library in the file named "LICENSE".
    1.21 +    If not, write to the Free Software Foundation, 51 Franklin Street, 
    1.22 +    Suite 500, Boston, MA 02110-1335, USA or visit their web page on the 
    1.23 +    internet at http://www.fsf.org/licenses/lgpl.html.
    1.24 +
    1.25 +Alternatively, the contents of this file may be used under the terms of the
    1.26 +Mozilla Public License (http://mozilla.org/MPL) or the GNU General Public
    1.27 +License, as published by the Free Software Foundation, either version 2
    1.28 +of the License or (at your option) any later version.
    1.29 +*/
    1.30 +#include <cstdlib>
    1.31 +#include "graphite2/Segment.h"
    1.32 +#include "inc/debug.h"
    1.33 +#include "inc/Endian.h"
    1.34 +#include "inc/Silf.h"
    1.35 +#include "inc/Segment.h"
    1.36 +#include "inc/Rule.h"
    1.37 +#include "inc/Error.h"
    1.38 +
    1.39 +
    1.40 +using namespace graphite2;
    1.41 +
    1.42 +namespace { static const uint32 ERROROFFSET = 0xFFFFFFFF; }
    1.43 +
    1.44 +Silf::Silf() throw()
    1.45 +: m_passes(0),
    1.46 +  m_pseudos(0),
    1.47 +  m_classOffsets(0),
    1.48 +  m_classData(0),
    1.49 +  m_justs(0),
    1.50 +  m_numPasses(0),
    1.51 +  m_numJusts(0),
    1.52 +  m_sPass(0),
    1.53 +  m_pPass(0),
    1.54 +  m_jPass(0),
    1.55 +  m_bPass(0),
    1.56 +  m_flags(0),
    1.57 +  m_aPseudo(0),
    1.58 +  m_aBreak(0),
    1.59 +  m_aUser(0),
    1.60 +  m_aBidi(0),
    1.61 +  m_aMirror(0),
    1.62 +  m_aPassBits(0),
    1.63 +  m_iMaxComp(0),
    1.64 +  m_aLig(0),
    1.65 +  m_numPseudo(0),
    1.66 +  m_nClass(0),
    1.67 +  m_nLinear(0),
    1.68 +  m_gEndLine(0)
    1.69 +{
    1.70 +    memset(&m_silfinfo, 0, sizeof m_silfinfo);
    1.71 +}
    1.72 +
    1.73 +Silf::~Silf() throw()
    1.74 +{
    1.75 +    releaseBuffers();
    1.76 +}
    1.77 +
    1.78 +void Silf::releaseBuffers() throw()
    1.79 +{
    1.80 +    delete [] m_passes;
    1.81 +    delete [] m_pseudos;
    1.82 +    free(m_classOffsets);
    1.83 +    free(m_classData);
    1.84 +    free(m_justs);
    1.85 +    m_passes= 0;
    1.86 +    m_pseudos = 0;
    1.87 +    m_classOffsets = 0;
    1.88 +    m_classData = 0;
    1.89 +    m_justs = 0;
    1.90 +}
    1.91 +
    1.92 +
    1.93 +bool Silf::readGraphite(const byte * const silf_start, size_t lSilf, Face& face, uint32 version)
    1.94 +{
    1.95 +    const byte * p = silf_start,
    1.96 +               * const silf_end = p + lSilf;
    1.97 +    Error e;
    1.98 +
    1.99 +    if (version >= 0x00030000)
   1.100 +    {
   1.101 +        if (e.test(lSilf < 28, E_BADSIZE)) { releaseBuffers(); return face.error(e); }
   1.102 +        be::skip<int32>(p);    // ruleVersion
   1.103 +        be::skip<uint16>(p,2); // passOffset & pseudosOffset
   1.104 +    }
   1.105 +    else if (e.test(lSilf < 20, E_BADSIZE)) { releaseBuffers(); return face.error(e); }
   1.106 +    const uint16 maxGlyph = be::read<uint16>(p);
   1.107 +    m_silfinfo.extra_ascent = be::read<uint16>(p);
   1.108 +    m_silfinfo.extra_descent = be::read<uint16>(p);
   1.109 +    m_numPasses = be::read<uint8>(p);
   1.110 +    m_sPass     = be::read<uint8>(p);
   1.111 +    m_pPass     = be::read<uint8>(p);
   1.112 +    m_jPass     = be::read<uint8>(p);
   1.113 +    m_bPass     = be::read<uint8>(p);
   1.114 +    m_flags     = be::read<uint8>(p);
   1.115 +    be::skip<uint8>(p,2); //  max{Pre,Post}Context.
   1.116 +    m_aPseudo   = be::read<uint8>(p);
   1.117 +    m_aBreak    = be::read<uint8>(p);
   1.118 +    m_aBidi     = be::read<uint8>(p);
   1.119 +    m_aMirror   = be::read<uint8>(p);
   1.120 +    m_aPassBits = be::read<uint8>(p);
   1.121 +
   1.122 +    // Read Justification levels.
   1.123 +    m_numJusts  = be::read<uint8>(p);
   1.124 +    if (e.test(maxGlyph >= face.glyphs().numGlyphs(), E_BADMAXGLYPH)
   1.125 +        || e.test(p + m_numJusts * 8 >= silf_end, E_BADNUMJUSTS))
   1.126 +    {
   1.127 +        releaseBuffers(); return face.error(e);
   1.128 +    }
   1.129 +
   1.130 +    if (m_numJusts)
   1.131 +    {
   1.132 +        m_justs = gralloc<Justinfo>(m_numJusts);
   1.133 +        if (e.test(!m_justs, E_OUTOFMEM)) return face.error(e);
   1.134 +
   1.135 +        for (uint8 i = 0; i < m_numJusts; i++)
   1.136 +        {
   1.137 +            ::new(m_justs + i) Justinfo(p[0], p[1], p[2], p[3]);
   1.138 +            be::skip<byte>(p,8);
   1.139 +        }
   1.140 +    }
   1.141 +
   1.142 +    if (e.test(p + sizeof(uint16) + sizeof(uint8)*8 >= silf_end, E_BADENDJUSTS)) { releaseBuffers(); return face.error(e); }
   1.143 +    m_aLig      = be::read<uint16>(p);
   1.144 +    m_aUser     = be::read<uint8>(p);
   1.145 +    m_iMaxComp  = be::read<uint8>(p);
   1.146 +    be::skip<byte>(p,5);                        // direction and 4 reserved bytes
   1.147 +    be::skip<uint16>(p, be::read<uint8>(p));    // don't need critical features yet
   1.148 +    be::skip<byte>(p);                          // reserved
   1.149 +    if (e.test(p >= silf_end, E_BADCRITFEATURES))   { releaseBuffers(); return face.error(e); }
   1.150 +    be::skip<uint32>(p, be::read<uint8>(p));    // don't use scriptTag array.
   1.151 +    if (e.test(p + sizeof(uint16) + sizeof(uint32) >= silf_end, E_BADSCRIPTTAGS)) { releaseBuffers(); return face.error(e); }
   1.152 +    m_gEndLine  = be::read<uint16>(p);          // lbGID
   1.153 +    const byte * o_passes = p,
   1.154 +               * const passes_start = silf_start + be::read<uint32>(p);
   1.155 +
   1.156 +    const size_t num_attrs = face.glyphs().numAttrs();
   1.157 +    if (e.test(m_aPseudo   >= num_attrs, E_BADAPSEUDO)
   1.158 +        || e.test(m_aBreak >= num_attrs, E_BADABREAK)
   1.159 +        || e.test(m_aBidi  >= num_attrs, E_BADABIDI)
   1.160 +        || e.test(m_aMirror>= num_attrs, E_BADAMIRROR)
   1.161 +        || e.test(m_numPasses > 128, E_BADNUMPASSES) || e.test(passes_start >= silf_end, E_BADPASSESSTART)
   1.162 +        || e.test(m_pPass < m_sPass, E_BADPASSBOUND) || e.test(m_pPass > m_numPasses, E_BADPPASS) || e.test(m_sPass > m_numPasses, E_BADSPASS)
   1.163 +        || e.test(m_jPass < m_pPass, E_BADJPASSBOUND) || e.test(m_jPass > m_numPasses, E_BADJPASS)
   1.164 +        || e.test((m_bPass != 0xFF && (m_bPass < m_jPass || m_bPass > m_numPasses)), E_BADBPASS)
   1.165 +        || e.test(m_aLig > 127, E_BADALIG))
   1.166 +    {
   1.167 +        releaseBuffers();
   1.168 +        return face.error(e);
   1.169 +    }
   1.170 +    be::skip<uint32>(p, m_numPasses);
   1.171 +    if (e.test(p + sizeof(uint16) >= passes_start, E_BADPASSESSTART)) { releaseBuffers(); return face.error(e); }
   1.172 +    m_numPseudo = be::read<uint16>(p);
   1.173 +    be::skip<uint16>(p, 3); // searchPseudo, pseudoSelector, pseudoShift
   1.174 +    if (e.test(p + m_numPseudo*(sizeof(uint32) + sizeof(uint16)) >= passes_start, E_BADNUMPSEUDO))
   1.175 +    {
   1.176 +        releaseBuffers(); return face.error(e);
   1.177 +    }
   1.178 +    m_pseudos = new Pseudo[m_numPseudo];
   1.179 +    for (int i = 0; i < m_numPseudo; i++)
   1.180 +    {
   1.181 +        m_pseudos[i].uid = be::read<uint32>(p);
   1.182 +        m_pseudos[i].gid = be::read<uint16>(p);
   1.183 +    }
   1.184 +
   1.185 +    const size_t clen = readClassMap(p, passes_start - p, version, e);
   1.186 +    if (e || e.test(p + clen > passes_start, E_BADPASSESSTART)) { releaseBuffers(); return face.error(e); }
   1.187 +
   1.188 +    m_passes = new Pass[m_numPasses];
   1.189 +    for (size_t i = 0; i < m_numPasses; ++i)
   1.190 +    {
   1.191 +        const byte * const pass_start = silf_start + be::read<uint32>(o_passes),
   1.192 +                   * const pass_end = silf_start + be::peek<uint32>(o_passes);
   1.193 +        face.error_context((face.error_context() & 0xFF00) + EC_ASILF + (i << 16));
   1.194 +        if (e.test(pass_start > pass_end, E_BADPASSSTART) || e.test(pass_end > silf_end, E_BADPASSEND)) {
   1.195 +            releaseBuffers(); return face.error(e);
   1.196 +        }
   1.197 +
   1.198 +        m_passes[i].init(this);
   1.199 +        if (!m_passes[i].readPass(pass_start, pass_end - pass_start, pass_start - silf_start, face, e))
   1.200 +        {
   1.201 +            releaseBuffers();
   1.202 +            return false;
   1.203 +        }
   1.204 +    }
   1.205 +
   1.206 +    // fill in gr_faceinfo
   1.207 +    m_silfinfo.upem = face.glyphs().unitsPerEm();
   1.208 +    m_silfinfo.has_bidi_pass = (m_bPass != 0xFF);
   1.209 +    m_silfinfo.justifies = (m_numJusts != 0) || (m_jPass < m_pPass);
   1.210 +    m_silfinfo.line_ends = (m_flags & 1);
   1.211 +    m_silfinfo.space_contextuals = gr_faceinfo::gr_space_contextuals((m_flags >> 2) & 0x7);
   1.212 +    return true;
   1.213 +}
   1.214 +
   1.215 +template<typename T> inline uint32 Silf::readClassOffsets(const byte *&p, size_t data_len, Error &e)
   1.216 +{
   1.217 +    const T cls_off = 2*sizeof(uint16) + sizeof(T)*(m_nClass+1);
   1.218 +    const size_t max_off = (be::peek<T>(p + sizeof(T)*m_nClass) - cls_off)/sizeof(uint16);
   1.219 +    // Check that the last+1 offset is less than or equal to the class map length.
   1.220 +    if (e.test(be::peek<T>(p) != cls_off, E_MISALIGNEDCLASSES) 
   1.221 +            || e.test(max_off > (data_len - cls_off)/sizeof(uint16), E_HIGHCLASSOFFSET))
   1.222 +        return ERROROFFSET;
   1.223 +
   1.224 +    // Read in all the offsets.
   1.225 +    m_classOffsets = gralloc<uint32>(m_nClass+1);
   1.226 +    if (e.test(!m_classOffsets, E_OUTOFMEM)) return ERROROFFSET;
   1.227 +    for (uint32 * o = m_classOffsets, * const o_end = o + m_nClass + 1; o != o_end; ++o)
   1.228 +    {
   1.229 +        *o = (be::read<T>(p) - cls_off)/sizeof(uint16);
   1.230 +        if (e.test(*o > max_off, E_HIGHCLASSOFFSET))
   1.231 +            return ERROROFFSET;
   1.232 +    }
   1.233 +    return max_off;
   1.234 +}
   1.235 +
   1.236 +size_t Silf::readClassMap(const byte *p, size_t data_len, uint32 version, Error &e)
   1.237 +{
   1.238 +    if (e.test(data_len < sizeof(uint16)*2, E_BADCLASSSIZE)) return ERROROFFSET;
   1.239 +
   1.240 +    m_nClass  = be::read<uint16>(p);
   1.241 +    m_nLinear = be::read<uint16>(p);
   1.242 +
   1.243 +    // Check that numLinear < numClass,
   1.244 +    // that there is at least enough data for numClasses offsets.
   1.245 +    if (e.test(m_nLinear > m_nClass, E_TOOMANYLINEAR)
   1.246 +     || e.test((m_nClass + 1) * (version >= 0x00040000 ? sizeof(uint32) : sizeof(uint16)) > (data_len - 4), E_CLASSESTOOBIG))
   1.247 +        return ERROROFFSET;
   1.248 +
   1.249 +    uint32 max_off;
   1.250 +    if (version >= 0x00040000)
   1.251 +        max_off = readClassOffsets<uint32>(p, data_len, e);
   1.252 +    else
   1.253 +        max_off = readClassOffsets<uint16>(p, data_len, e);
   1.254 +
   1.255 +    if (max_off == ERROROFFSET) return ERROROFFSET;
   1.256 +
   1.257 +    // Check the linear offsets are sane, these must be monotonically increasing.
   1.258 +    for (const uint32 *o = m_classOffsets, * const o_end = o + m_nLinear; o != o_end; ++o)
   1.259 +        if (e.test(o[0] > o[1], E_BADCLASSOFFSET))
   1.260 +            return ERROROFFSET;
   1.261 +
   1.262 +    // Fortunately the class data is all uint16s so we can decode these now
   1.263 +    m_classData = gralloc<uint16>(max_off);
   1.264 +    if (e.test(!m_classData, E_OUTOFMEM)) return ERROROFFSET;
   1.265 +    for (uint16 *d = m_classData, * const d_end = d + max_off; d != d_end; ++d)
   1.266 +        *d = be::read<uint16>(p);
   1.267 +
   1.268 +    // Check the lookup class invariants for each non-linear class
   1.269 +    for (const uint32 *o = m_classOffsets + m_nLinear, * const o_end = m_classOffsets + m_nClass; o != o_end; ++o)
   1.270 +    {
   1.271 +        const uint16 * lookup = m_classData + *o;
   1.272 +        if (e.test(*o > max_off - 4, E_HIGHCLASSOFFSET)                        // LookupClass doesn't stretch over max_off
   1.273 +         || e.test(lookup[0] == 0                                                   // A LookupClass with no looks is a suspicious thing ...
   1.274 +                    || lookup[0] > (max_off - *o - 4)/2                             // numIDs lookup pairs fits within (start of LookupClass' lookups array, max_off]
   1.275 +                    || lookup[3] != lookup[0] - lookup[1], E_BADCLASSLOOKUPINFO))   // rangeShift:   numIDs  - searchRange
   1.276 +            return ERROROFFSET;
   1.277 +    }
   1.278 +
   1.279 +    return max_off;
   1.280 +}
   1.281 +
   1.282 +uint16 Silf::findPseudo(uint32 uid) const
   1.283 +{
   1.284 +    for (int i = 0; i < m_numPseudo; i++)
   1.285 +        if (m_pseudos[i].uid == uid) return m_pseudos[i].gid;
   1.286 +    return 0;
   1.287 +}
   1.288 +
   1.289 +uint16 Silf::findClassIndex(uint16 cid, uint16 gid) const
   1.290 +{
   1.291 +    if (cid > m_nClass) return -1;
   1.292 +
   1.293 +    const uint16 * cls = m_classData + m_classOffsets[cid];
   1.294 +    if (cid < m_nLinear)        // output class being used for input, shouldn't happen
   1.295 +    {
   1.296 +        for (unsigned int i = 0, n = m_classOffsets[cid + 1]; i < n; ++i, ++cls)
   1.297 +            if (*cls == gid) return i;
   1.298 +        return -1;
   1.299 +    }
   1.300 +    else
   1.301 +    {
   1.302 +        const uint16 *  min = cls + 4,      // lookups array
   1.303 +                     *  max = min + cls[0]*2; // lookups aray is numIDs (cls[0]) uint16 pairs long
   1.304 +        do
   1.305 +        {
   1.306 +            const uint16 * p = min + (-2 & ((max-min)/2));
   1.307 +            if  (p[0] > gid)    max = p;
   1.308 +            else                min = p;
   1.309 +        }
   1.310 +        while (max - min > 2);
   1.311 +        return min[0] == gid ? min[1] : -1;
   1.312 +    }
   1.313 +}
   1.314 +
   1.315 +uint16 Silf::getClassGlyph(uint16 cid, unsigned int index) const
   1.316 +{
   1.317 +    if (cid > m_nClass) return 0;
   1.318 +
   1.319 +    uint32 loc = m_classOffsets[cid];
   1.320 +    if (cid < m_nLinear)
   1.321 +    {
   1.322 +        if (index < m_classOffsets[cid + 1] - loc)
   1.323 +            return m_classData[index + loc];
   1.324 +    }
   1.325 +    else        // input class being used for output. Shouldn't happen
   1.326 +    {
   1.327 +        for (unsigned int i = loc + 4; i < m_classOffsets[cid + 1]; i += 2)
   1.328 +            if (m_classData[i + 1] == index) return m_classData[i];
   1.329 +    }
   1.330 +    return 0;
   1.331 +}
   1.332 +
   1.333 +
   1.334 +bool Silf::runGraphite(Segment *seg, uint8 firstPass, uint8 lastPass, int dobidi) const
   1.335 +{
   1.336 +    assert(seg != 0);
   1.337 +    SlotMap            map(*seg);
   1.338 +    FiniteStateMachine fsm(map, seg->getFace()->logger());
   1.339 +    vm::Machine        m(map);
   1.340 +    unsigned int       initSize = seg->slotCount();
   1.341 +    uint8              lbidi = m_bPass;
   1.342 +#if !defined GRAPHITE2_NTRACING
   1.343 +    json * const dbgout = seg->getFace()->logger();
   1.344 +#endif
   1.345 +
   1.346 +    if (lastPass == 0)
   1.347 +    {
   1.348 +        if (firstPass == lastPass && lbidi == 0xFF)
   1.349 +            return true;
   1.350 +        lastPass = m_numPasses;
   1.351 +    }
   1.352 +    if (firstPass <= lbidi && lastPass >= lbidi && dobidi)
   1.353 +        lastPass++;
   1.354 +    else
   1.355 +        lbidi = 0xFF;
   1.356 +
   1.357 +    for (size_t i = firstPass; i < lastPass; ++i)
   1.358 +    {
   1.359 +        // bidi and mirroring
   1.360 +        if (i == lbidi)
   1.361 +        {
   1.362 +#if !defined GRAPHITE2_NTRACING
   1.363 +            if (dbgout)
   1.364 +            {
   1.365 +                *dbgout << json::item << json::object
   1.366 +                            << "id"     << -1
   1.367 +                            << "slots"  << json::array;
   1.368 +                seg->positionSlots(0);
   1.369 +                for(Slot * s = seg->first(); s; s = s->next())
   1.370 +                    *dbgout     << dslot(seg, s);
   1.371 +                *dbgout         << json::close
   1.372 +                            << "rules"  << json::array << json::close
   1.373 +                            << json::close;
   1.374 +            }
   1.375 +#endif
   1.376 +
   1.377 +            if (!(seg->dir() & 2))
   1.378 +                seg->bidiPass(m_aBidi, seg->dir() & 1, m_aMirror);
   1.379 +            else if (m_aMirror)
   1.380 +            {
   1.381 +                Slot * s;
   1.382 +                for (s = seg->first(); s; s = s->next())
   1.383 +                {
   1.384 +                    unsigned short g = seg->glyphAttr(s->gid(), m_aMirror);
   1.385 +                    if (g && (!(seg->dir() & 4) || !seg->glyphAttr(s->gid(), m_aMirror + 1)))
   1.386 +                        s->setGlyph(seg, g);
   1.387 +                }
   1.388 +            }
   1.389 +        --i;
   1.390 +        --lastPass;
   1.391 +        lbidi = 0xFF;
   1.392 +        continue;
   1.393 +        }
   1.394 +
   1.395 +#if !defined GRAPHITE2_NTRACING
   1.396 +        if (dbgout)
   1.397 +        {
   1.398 +            *dbgout << json::item << json::object
   1.399 +                        << "id"     << i+1
   1.400 +                        << "slots"  << json::array;
   1.401 +            seg->positionSlots(0);
   1.402 +            for(Slot * s = seg->first(); s; s = s->next())
   1.403 +                *dbgout     << dslot(seg, s);
   1.404 +            *dbgout         << json::close;
   1.405 +        }
   1.406 +#endif
   1.407 +
   1.408 +        // test whether to reorder, prepare for positioning
   1.409 +        if (i >= 32 || (seg->passBits() & (1 << i)) == 0)
   1.410 +            m_passes[i].runGraphite(m, fsm);
   1.411 +        // only subsitution passes can change segment length, cached subsegments are short for their text
   1.412 +        if (m.status() != vm::Machine::finished
   1.413 +            || (i < m_pPass && (seg->slotCount() > initSize * MAX_SEG_GROWTH_FACTOR
   1.414 +            || (seg->slotCount() && seg->slotCount() * MAX_SEG_GROWTH_FACTOR < initSize))))
   1.415 +            return false;
   1.416 +    }
   1.417 +    return true;
   1.418 +}

mercurial