|
1 /* GRAPHITE2 LICENSING |
|
2 |
|
3 Copyright 2010, SIL International |
|
4 All rights reserved. |
|
5 |
|
6 This library is free software; you can redistribute it and/or modify |
|
7 it under the terms of the GNU Lesser General Public License as published |
|
8 by the Free Software Foundation; either version 2.1 of License, or |
|
9 (at your option) any later version. |
|
10 |
|
11 This program is distributed in the hope that it will be useful, |
|
12 but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
14 Lesser General Public License for more details. |
|
15 |
|
16 You should also have received a copy of the GNU Lesser General Public |
|
17 License along with this library in the file named "LICENSE". |
|
18 If not, write to the Free Software Foundation, 51 Franklin Street, |
|
19 Suite 500, Boston, MA 02110-1335, USA or visit their web page on the |
|
20 internet at http://www.fsf.org/licenses/lgpl.html. |
|
21 |
|
22 Alternatively, the contents of this file may be used under the terms of the |
|
23 Mozilla Public License (http://mozilla.org/MPL) or the GNU General Public |
|
24 License, as published by the Free Software Foundation, either version 2 |
|
25 of the License or (at your option) any later version. |
|
26 */ |
|
27 #include <cstdlib> |
|
28 #include "graphite2/Segment.h" |
|
29 #include "inc/debug.h" |
|
30 #include "inc/Endian.h" |
|
31 #include "inc/Silf.h" |
|
32 #include "inc/Segment.h" |
|
33 #include "inc/Rule.h" |
|
34 #include "inc/Error.h" |
|
35 |
|
36 |
|
37 using namespace graphite2; |
|
38 |
|
39 namespace { static const uint32 ERROROFFSET = 0xFFFFFFFF; } |
|
40 |
|
41 Silf::Silf() throw() |
|
42 : m_passes(0), |
|
43 m_pseudos(0), |
|
44 m_classOffsets(0), |
|
45 m_classData(0), |
|
46 m_justs(0), |
|
47 m_numPasses(0), |
|
48 m_numJusts(0), |
|
49 m_sPass(0), |
|
50 m_pPass(0), |
|
51 m_jPass(0), |
|
52 m_bPass(0), |
|
53 m_flags(0), |
|
54 m_aPseudo(0), |
|
55 m_aBreak(0), |
|
56 m_aUser(0), |
|
57 m_aBidi(0), |
|
58 m_aMirror(0), |
|
59 m_aPassBits(0), |
|
60 m_iMaxComp(0), |
|
61 m_aLig(0), |
|
62 m_numPseudo(0), |
|
63 m_nClass(0), |
|
64 m_nLinear(0), |
|
65 m_gEndLine(0) |
|
66 { |
|
67 memset(&m_silfinfo, 0, sizeof m_silfinfo); |
|
68 } |
|
69 |
|
70 Silf::~Silf() throw() |
|
71 { |
|
72 releaseBuffers(); |
|
73 } |
|
74 |
|
75 void Silf::releaseBuffers() throw() |
|
76 { |
|
77 delete [] m_passes; |
|
78 delete [] m_pseudos; |
|
79 free(m_classOffsets); |
|
80 free(m_classData); |
|
81 free(m_justs); |
|
82 m_passes= 0; |
|
83 m_pseudos = 0; |
|
84 m_classOffsets = 0; |
|
85 m_classData = 0; |
|
86 m_justs = 0; |
|
87 } |
|
88 |
|
89 |
|
90 bool Silf::readGraphite(const byte * const silf_start, size_t lSilf, Face& face, uint32 version) |
|
91 { |
|
92 const byte * p = silf_start, |
|
93 * const silf_end = p + lSilf; |
|
94 Error e; |
|
95 |
|
96 if (version >= 0x00030000) |
|
97 { |
|
98 if (e.test(lSilf < 28, E_BADSIZE)) { releaseBuffers(); return face.error(e); } |
|
99 be::skip<int32>(p); // ruleVersion |
|
100 be::skip<uint16>(p,2); // passOffset & pseudosOffset |
|
101 } |
|
102 else if (e.test(lSilf < 20, E_BADSIZE)) { releaseBuffers(); return face.error(e); } |
|
103 const uint16 maxGlyph = be::read<uint16>(p); |
|
104 m_silfinfo.extra_ascent = be::read<uint16>(p); |
|
105 m_silfinfo.extra_descent = be::read<uint16>(p); |
|
106 m_numPasses = be::read<uint8>(p); |
|
107 m_sPass = be::read<uint8>(p); |
|
108 m_pPass = be::read<uint8>(p); |
|
109 m_jPass = be::read<uint8>(p); |
|
110 m_bPass = be::read<uint8>(p); |
|
111 m_flags = be::read<uint8>(p); |
|
112 be::skip<uint8>(p,2); // max{Pre,Post}Context. |
|
113 m_aPseudo = be::read<uint8>(p); |
|
114 m_aBreak = be::read<uint8>(p); |
|
115 m_aBidi = be::read<uint8>(p); |
|
116 m_aMirror = be::read<uint8>(p); |
|
117 m_aPassBits = be::read<uint8>(p); |
|
118 |
|
119 // Read Justification levels. |
|
120 m_numJusts = be::read<uint8>(p); |
|
121 if (e.test(maxGlyph >= face.glyphs().numGlyphs(), E_BADMAXGLYPH) |
|
122 || e.test(p + m_numJusts * 8 >= silf_end, E_BADNUMJUSTS)) |
|
123 { |
|
124 releaseBuffers(); return face.error(e); |
|
125 } |
|
126 |
|
127 if (m_numJusts) |
|
128 { |
|
129 m_justs = gralloc<Justinfo>(m_numJusts); |
|
130 if (e.test(!m_justs, E_OUTOFMEM)) return face.error(e); |
|
131 |
|
132 for (uint8 i = 0; i < m_numJusts; i++) |
|
133 { |
|
134 ::new(m_justs + i) Justinfo(p[0], p[1], p[2], p[3]); |
|
135 be::skip<byte>(p,8); |
|
136 } |
|
137 } |
|
138 |
|
139 if (e.test(p + sizeof(uint16) + sizeof(uint8)*8 >= silf_end, E_BADENDJUSTS)) { releaseBuffers(); return face.error(e); } |
|
140 m_aLig = be::read<uint16>(p); |
|
141 m_aUser = be::read<uint8>(p); |
|
142 m_iMaxComp = be::read<uint8>(p); |
|
143 be::skip<byte>(p,5); // direction and 4 reserved bytes |
|
144 be::skip<uint16>(p, be::read<uint8>(p)); // don't need critical features yet |
|
145 be::skip<byte>(p); // reserved |
|
146 if (e.test(p >= silf_end, E_BADCRITFEATURES)) { releaseBuffers(); return face.error(e); } |
|
147 be::skip<uint32>(p, be::read<uint8>(p)); // don't use scriptTag array. |
|
148 if (e.test(p + sizeof(uint16) + sizeof(uint32) >= silf_end, E_BADSCRIPTTAGS)) { releaseBuffers(); return face.error(e); } |
|
149 m_gEndLine = be::read<uint16>(p); // lbGID |
|
150 const byte * o_passes = p, |
|
151 * const passes_start = silf_start + be::read<uint32>(p); |
|
152 |
|
153 const size_t num_attrs = face.glyphs().numAttrs(); |
|
154 if (e.test(m_aPseudo >= num_attrs, E_BADAPSEUDO) |
|
155 || e.test(m_aBreak >= num_attrs, E_BADABREAK) |
|
156 || e.test(m_aBidi >= num_attrs, E_BADABIDI) |
|
157 || e.test(m_aMirror>= num_attrs, E_BADAMIRROR) |
|
158 || e.test(m_numPasses > 128, E_BADNUMPASSES) || e.test(passes_start >= silf_end, E_BADPASSESSTART) |
|
159 || e.test(m_pPass < m_sPass, E_BADPASSBOUND) || e.test(m_pPass > m_numPasses, E_BADPPASS) || e.test(m_sPass > m_numPasses, E_BADSPASS) |
|
160 || e.test(m_jPass < m_pPass, E_BADJPASSBOUND) || e.test(m_jPass > m_numPasses, E_BADJPASS) |
|
161 || e.test((m_bPass != 0xFF && (m_bPass < m_jPass || m_bPass > m_numPasses)), E_BADBPASS) |
|
162 || e.test(m_aLig > 127, E_BADALIG)) |
|
163 { |
|
164 releaseBuffers(); |
|
165 return face.error(e); |
|
166 } |
|
167 be::skip<uint32>(p, m_numPasses); |
|
168 if (e.test(p + sizeof(uint16) >= passes_start, E_BADPASSESSTART)) { releaseBuffers(); return face.error(e); } |
|
169 m_numPseudo = be::read<uint16>(p); |
|
170 be::skip<uint16>(p, 3); // searchPseudo, pseudoSelector, pseudoShift |
|
171 if (e.test(p + m_numPseudo*(sizeof(uint32) + sizeof(uint16)) >= passes_start, E_BADNUMPSEUDO)) |
|
172 { |
|
173 releaseBuffers(); return face.error(e); |
|
174 } |
|
175 m_pseudos = new Pseudo[m_numPseudo]; |
|
176 for (int i = 0; i < m_numPseudo; i++) |
|
177 { |
|
178 m_pseudos[i].uid = be::read<uint32>(p); |
|
179 m_pseudos[i].gid = be::read<uint16>(p); |
|
180 } |
|
181 |
|
182 const size_t clen = readClassMap(p, passes_start - p, version, e); |
|
183 if (e || e.test(p + clen > passes_start, E_BADPASSESSTART)) { releaseBuffers(); return face.error(e); } |
|
184 |
|
185 m_passes = new Pass[m_numPasses]; |
|
186 for (size_t i = 0; i < m_numPasses; ++i) |
|
187 { |
|
188 const byte * const pass_start = silf_start + be::read<uint32>(o_passes), |
|
189 * const pass_end = silf_start + be::peek<uint32>(o_passes); |
|
190 face.error_context((face.error_context() & 0xFF00) + EC_ASILF + (i << 16)); |
|
191 if (e.test(pass_start > pass_end, E_BADPASSSTART) || e.test(pass_end > silf_end, E_BADPASSEND)) { |
|
192 releaseBuffers(); return face.error(e); |
|
193 } |
|
194 |
|
195 m_passes[i].init(this); |
|
196 if (!m_passes[i].readPass(pass_start, pass_end - pass_start, pass_start - silf_start, face, e)) |
|
197 { |
|
198 releaseBuffers(); |
|
199 return false; |
|
200 } |
|
201 } |
|
202 |
|
203 // fill in gr_faceinfo |
|
204 m_silfinfo.upem = face.glyphs().unitsPerEm(); |
|
205 m_silfinfo.has_bidi_pass = (m_bPass != 0xFF); |
|
206 m_silfinfo.justifies = (m_numJusts != 0) || (m_jPass < m_pPass); |
|
207 m_silfinfo.line_ends = (m_flags & 1); |
|
208 m_silfinfo.space_contextuals = gr_faceinfo::gr_space_contextuals((m_flags >> 2) & 0x7); |
|
209 return true; |
|
210 } |
|
211 |
|
212 template<typename T> inline uint32 Silf::readClassOffsets(const byte *&p, size_t data_len, Error &e) |
|
213 { |
|
214 const T cls_off = 2*sizeof(uint16) + sizeof(T)*(m_nClass+1); |
|
215 const size_t max_off = (be::peek<T>(p + sizeof(T)*m_nClass) - cls_off)/sizeof(uint16); |
|
216 // Check that the last+1 offset is less than or equal to the class map length. |
|
217 if (e.test(be::peek<T>(p) != cls_off, E_MISALIGNEDCLASSES) |
|
218 || e.test(max_off > (data_len - cls_off)/sizeof(uint16), E_HIGHCLASSOFFSET)) |
|
219 return ERROROFFSET; |
|
220 |
|
221 // Read in all the offsets. |
|
222 m_classOffsets = gralloc<uint32>(m_nClass+1); |
|
223 if (e.test(!m_classOffsets, E_OUTOFMEM)) return ERROROFFSET; |
|
224 for (uint32 * o = m_classOffsets, * const o_end = o + m_nClass + 1; o != o_end; ++o) |
|
225 { |
|
226 *o = (be::read<T>(p) - cls_off)/sizeof(uint16); |
|
227 if (e.test(*o > max_off, E_HIGHCLASSOFFSET)) |
|
228 return ERROROFFSET; |
|
229 } |
|
230 return max_off; |
|
231 } |
|
232 |
|
233 size_t Silf::readClassMap(const byte *p, size_t data_len, uint32 version, Error &e) |
|
234 { |
|
235 if (e.test(data_len < sizeof(uint16)*2, E_BADCLASSSIZE)) return ERROROFFSET; |
|
236 |
|
237 m_nClass = be::read<uint16>(p); |
|
238 m_nLinear = be::read<uint16>(p); |
|
239 |
|
240 // Check that numLinear < numClass, |
|
241 // that there is at least enough data for numClasses offsets. |
|
242 if (e.test(m_nLinear > m_nClass, E_TOOMANYLINEAR) |
|
243 || e.test((m_nClass + 1) * (version >= 0x00040000 ? sizeof(uint32) : sizeof(uint16)) > (data_len - 4), E_CLASSESTOOBIG)) |
|
244 return ERROROFFSET; |
|
245 |
|
246 uint32 max_off; |
|
247 if (version >= 0x00040000) |
|
248 max_off = readClassOffsets<uint32>(p, data_len, e); |
|
249 else |
|
250 max_off = readClassOffsets<uint16>(p, data_len, e); |
|
251 |
|
252 if (max_off == ERROROFFSET) return ERROROFFSET; |
|
253 |
|
254 // Check the linear offsets are sane, these must be monotonically increasing. |
|
255 for (const uint32 *o = m_classOffsets, * const o_end = o + m_nLinear; o != o_end; ++o) |
|
256 if (e.test(o[0] > o[1], E_BADCLASSOFFSET)) |
|
257 return ERROROFFSET; |
|
258 |
|
259 // Fortunately the class data is all uint16s so we can decode these now |
|
260 m_classData = gralloc<uint16>(max_off); |
|
261 if (e.test(!m_classData, E_OUTOFMEM)) return ERROROFFSET; |
|
262 for (uint16 *d = m_classData, * const d_end = d + max_off; d != d_end; ++d) |
|
263 *d = be::read<uint16>(p); |
|
264 |
|
265 // Check the lookup class invariants for each non-linear class |
|
266 for (const uint32 *o = m_classOffsets + m_nLinear, * const o_end = m_classOffsets + m_nClass; o != o_end; ++o) |
|
267 { |
|
268 const uint16 * lookup = m_classData + *o; |
|
269 if (e.test(*o > max_off - 4, E_HIGHCLASSOFFSET) // LookupClass doesn't stretch over max_off |
|
270 || e.test(lookup[0] == 0 // A LookupClass with no looks is a suspicious thing ... |
|
271 || lookup[0] > (max_off - *o - 4)/2 // numIDs lookup pairs fits within (start of LookupClass' lookups array, max_off] |
|
272 || lookup[3] != lookup[0] - lookup[1], E_BADCLASSLOOKUPINFO)) // rangeShift: numIDs - searchRange |
|
273 return ERROROFFSET; |
|
274 } |
|
275 |
|
276 return max_off; |
|
277 } |
|
278 |
|
279 uint16 Silf::findPseudo(uint32 uid) const |
|
280 { |
|
281 for (int i = 0; i < m_numPseudo; i++) |
|
282 if (m_pseudos[i].uid == uid) return m_pseudos[i].gid; |
|
283 return 0; |
|
284 } |
|
285 |
|
286 uint16 Silf::findClassIndex(uint16 cid, uint16 gid) const |
|
287 { |
|
288 if (cid > m_nClass) return -1; |
|
289 |
|
290 const uint16 * cls = m_classData + m_classOffsets[cid]; |
|
291 if (cid < m_nLinear) // output class being used for input, shouldn't happen |
|
292 { |
|
293 for (unsigned int i = 0, n = m_classOffsets[cid + 1]; i < n; ++i, ++cls) |
|
294 if (*cls == gid) return i; |
|
295 return -1; |
|
296 } |
|
297 else |
|
298 { |
|
299 const uint16 * min = cls + 4, // lookups array |
|
300 * max = min + cls[0]*2; // lookups aray is numIDs (cls[0]) uint16 pairs long |
|
301 do |
|
302 { |
|
303 const uint16 * p = min + (-2 & ((max-min)/2)); |
|
304 if (p[0] > gid) max = p; |
|
305 else min = p; |
|
306 } |
|
307 while (max - min > 2); |
|
308 return min[0] == gid ? min[1] : -1; |
|
309 } |
|
310 } |
|
311 |
|
312 uint16 Silf::getClassGlyph(uint16 cid, unsigned int index) const |
|
313 { |
|
314 if (cid > m_nClass) return 0; |
|
315 |
|
316 uint32 loc = m_classOffsets[cid]; |
|
317 if (cid < m_nLinear) |
|
318 { |
|
319 if (index < m_classOffsets[cid + 1] - loc) |
|
320 return m_classData[index + loc]; |
|
321 } |
|
322 else // input class being used for output. Shouldn't happen |
|
323 { |
|
324 for (unsigned int i = loc + 4; i < m_classOffsets[cid + 1]; i += 2) |
|
325 if (m_classData[i + 1] == index) return m_classData[i]; |
|
326 } |
|
327 return 0; |
|
328 } |
|
329 |
|
330 |
|
331 bool Silf::runGraphite(Segment *seg, uint8 firstPass, uint8 lastPass, int dobidi) const |
|
332 { |
|
333 assert(seg != 0); |
|
334 SlotMap map(*seg); |
|
335 FiniteStateMachine fsm(map, seg->getFace()->logger()); |
|
336 vm::Machine m(map); |
|
337 unsigned int initSize = seg->slotCount(); |
|
338 uint8 lbidi = m_bPass; |
|
339 #if !defined GRAPHITE2_NTRACING |
|
340 json * const dbgout = seg->getFace()->logger(); |
|
341 #endif |
|
342 |
|
343 if (lastPass == 0) |
|
344 { |
|
345 if (firstPass == lastPass && lbidi == 0xFF) |
|
346 return true; |
|
347 lastPass = m_numPasses; |
|
348 } |
|
349 if (firstPass <= lbidi && lastPass >= lbidi && dobidi) |
|
350 lastPass++; |
|
351 else |
|
352 lbidi = 0xFF; |
|
353 |
|
354 for (size_t i = firstPass; i < lastPass; ++i) |
|
355 { |
|
356 // bidi and mirroring |
|
357 if (i == lbidi) |
|
358 { |
|
359 #if !defined GRAPHITE2_NTRACING |
|
360 if (dbgout) |
|
361 { |
|
362 *dbgout << json::item << json::object |
|
363 << "id" << -1 |
|
364 << "slots" << json::array; |
|
365 seg->positionSlots(0); |
|
366 for(Slot * s = seg->first(); s; s = s->next()) |
|
367 *dbgout << dslot(seg, s); |
|
368 *dbgout << json::close |
|
369 << "rules" << json::array << json::close |
|
370 << json::close; |
|
371 } |
|
372 #endif |
|
373 |
|
374 if (!(seg->dir() & 2)) |
|
375 seg->bidiPass(m_aBidi, seg->dir() & 1, m_aMirror); |
|
376 else if (m_aMirror) |
|
377 { |
|
378 Slot * s; |
|
379 for (s = seg->first(); s; s = s->next()) |
|
380 { |
|
381 unsigned short g = seg->glyphAttr(s->gid(), m_aMirror); |
|
382 if (g && (!(seg->dir() & 4) || !seg->glyphAttr(s->gid(), m_aMirror + 1))) |
|
383 s->setGlyph(seg, g); |
|
384 } |
|
385 } |
|
386 --i; |
|
387 --lastPass; |
|
388 lbidi = 0xFF; |
|
389 continue; |
|
390 } |
|
391 |
|
392 #if !defined GRAPHITE2_NTRACING |
|
393 if (dbgout) |
|
394 { |
|
395 *dbgout << json::item << json::object |
|
396 << "id" << i+1 |
|
397 << "slots" << json::array; |
|
398 seg->positionSlots(0); |
|
399 for(Slot * s = seg->first(); s; s = s->next()) |
|
400 *dbgout << dslot(seg, s); |
|
401 *dbgout << json::close; |
|
402 } |
|
403 #endif |
|
404 |
|
405 // test whether to reorder, prepare for positioning |
|
406 if (i >= 32 || (seg->passBits() & (1 << i)) == 0) |
|
407 m_passes[i].runGraphite(m, fsm); |
|
408 // only subsitution passes can change segment length, cached subsegments are short for their text |
|
409 if (m.status() != vm::Machine::finished |
|
410 || (i < m_pPass && (seg->slotCount() > initSize * MAX_SEG_GROWTH_FACTOR |
|
411 || (seg->slotCount() && seg->slotCount() * MAX_SEG_GROWTH_FACTOR < initSize)))) |
|
412 return false; |
|
413 } |
|
414 return true; |
|
415 } |