Coverage Report

Created: 2025-07-07 10:01

/work/workdir/UnpackedTarball/harfbuzz/src/hb-subset-cff-common.hh
Line
Count
Source (jump to first uncovered line)
1
/*
2
 * Copyright © 2018 Adobe Inc.
3
 *
4
 *  This is part of HarfBuzz, a text shaping library.
5
 *
6
 * Permission is hereby granted, without written agreement and without
7
 * license or royalty fees, to use, copy, modify, and distribute this
8
 * software and its documentation for any purpose, provided that the
9
 * above copyright notice and the following two paragraphs appear in
10
 * all copies of this software.
11
 *
12
 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
13
 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
14
 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
15
 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
16
 * DAMAGE.
17
 *
18
 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
19
 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
20
 * FITNESS FOR A PARTICULAR PURPOSE.  THE SOFTWARE PROVIDED HEREUNDER IS
21
 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
22
 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
23
 *
24
 * Adobe Author(s): Michiharu Ariza
25
 */
26
27
#ifndef HB_SUBSET_CFF_COMMON_HH
28
#define HB_SUBSET_CFF_COMMON_HH
29
30
#include "hb.hh"
31
32
#include "hb-subset-plan.hh"
33
#include "hb-cff-interp-cs-common.hh"
34
35
namespace CFF {
36
37
/* Used for writing a temporary charstring */
38
struct str_encoder_t
39
{
40
  str_encoder_t (str_buff_t &buff_)
41
0
    : buff (buff_) {}
42
43
0
  void reset () { buff.reset (); }
44
45
  void encode_byte (unsigned char b)
46
0
  {
47
0
    if (likely ((signed) buff.length < buff.allocated))
48
0
      buff.arrayZ[buff.length++] = b;
49
0
    else
50
0
      buff.push (b);
51
0
  }
52
53
  void encode_int (int v)
54
0
  {
55
0
    if ((-1131 <= v) && (v <= 1131))
56
0
    {
57
0
      if ((-107 <= v) && (v <= 107))
58
0
  encode_byte (v + 139);
59
0
      else if (v > 0)
60
0
      {
61
0
  v -= 108;
62
0
  encode_byte ((v >> 8) + OpCode_TwoBytePosInt0);
63
0
  encode_byte (v & 0xFF);
64
0
      }
65
0
      else
66
0
      {
67
0
  v = -v - 108;
68
0
  encode_byte ((v >> 8) + OpCode_TwoByteNegInt0);
69
0
  encode_byte (v & 0xFF);
70
0
      }
71
0
    }
72
0
    else
73
0
    {
74
0
      if (unlikely (v < -32768))
75
0
  v = -32768;
76
0
      else if (unlikely (v > 32767))
77
0
  v = 32767;
78
0
      encode_byte (OpCode_shortint);
79
0
      encode_byte ((v >> 8) & 0xFF);
80
0
      encode_byte (v & 0xFF);
81
0
    }
82
0
  }
83
84
  // Encode number for CharString
85
  void encode_num_cs (const number_t& n)
86
0
  {
87
0
    if (n.in_int_range ())
88
0
    {
89
0
      encode_int (n.to_int ());
90
0
    }
91
0
    else
92
0
    {
93
0
      int32_t v = n.to_fixed ();
94
0
      encode_byte (OpCode_fixedcs);
95
0
      encode_byte ((v >> 24) & 0xFF);
96
0
      encode_byte ((v >> 16) & 0xFF);
97
0
      encode_byte ((v >> 8) & 0xFF);
98
0
      encode_byte (v & 0xFF);
99
0
    }
100
0
  }
101
102
  // Encode number for TopDict / Private
103
  void encode_num_tp (const number_t& n)
104
0
  {
105
0
    if (n.in_int_range ())
106
0
    {
107
0
      // TODO longint
108
0
      encode_int (n.to_int ());
109
0
    }
110
0
    else
111
0
    {
112
0
      // Sigh. BCD
113
0
      // https://learn.microsoft.com/en-us/typography/opentype/spec/cff2#table-5-nibble-definitions
114
0
      double v = n.to_real ();
115
0
      encode_byte (OpCode_BCD);
116
0
117
0
      // Based on:
118
0
      // https://github.com/fonttools/fonttools/blob/0738c41dfbcbc213ab9263f486ef0cccc6eb5ce5/Lib/fontTools/misc/psCharStrings.py#L267-L316
119
0
120
0
      char buf[16];
121
0
      /* FontTools has the following comment:
122
0
       *
123
0
       * # Note: 14 decimal digits seems to be the limitation for CFF real numbers
124
0
       * # in macOS. However, we use 8 here to match the implementation of AFDKO.
125
0
       *
126
0
       * We use 8 here to match FontTools X-).
127
0
       */
128
0
129
0
      hb_locale_t clocale HB_UNUSED;
130
0
      hb_locale_t oldlocale HB_UNUSED;
131
0
      oldlocale = hb_uselocale (clocale = newlocale (LC_ALL_MASK, "C", NULL));
132
0
      snprintf (buf, sizeof (buf), "%.8G", v);
133
0
      (void) hb_uselocale (((void) freelocale (clocale), oldlocale));
134
0
135
0
      char *s = buf;
136
0
      size_t len;
137
0
      char *comma = strchr (s, ',');
138
0
      if (comma) // Comma for some European locales in case no uselocale available.
139
0
  *comma = '.';
140
0
      if (s[0] == '0' && s[1] == '.')
141
0
  s++;
142
0
      else if (s[0] == '-' && s[1] == '0' && s[2] == '.')
143
0
      {
144
0
  s[1] = '-';
145
0
  s++;
146
0
      }
147
0
      else if ((len = strlen (s)) > 3 && !strcmp (s + len - 3, "000"))
148
0
      {
149
0
  unsigned exponent = len - 3;
150
0
  char *s2 = s + exponent - 1;
151
0
  while (*s2 == '0' && exponent > 1)
152
0
  {
153
0
    s2--;
154
0
    exponent++;
155
0
  }
156
0
  snprintf (s2 + 1, sizeof (buf) - (s2 + 1 - buf), "E%u", exponent);
157
0
      }
158
0
      else
159
0
      {
160
0
  char *dot = strchr (s, '.');
161
0
  char *e = strchr (s, 'E');
162
0
  if (dot && e)
163
0
  {
164
0
    memmove (dot, dot + 1, e - (dot + 1));
165
0
    int exponent = atoi (e + 1);
166
0
    int new_exponent = exponent - (e - (dot + 1));
167
0
    if (new_exponent == 1)
168
0
    {
169
0
      e[-1] = '0';
170
0
      e[0] = '\0';
171
0
    }
172
0
    else
173
0
      snprintf (e - 1, sizeof (buf) - (e - 1 - buf), "E%d", new_exponent);
174
0
  }
175
0
      }
176
0
      if ((s[0] == '.' && s[1] == '0') || (s[0] == '-' && s[1] == '.' && s[2] == '0'))
177
0
      {
178
0
  int sign = s[0] == '-';
179
0
  char *s2 = s + sign + 1;
180
0
  while (*s2 == '0')
181
0
    s2++;
182
0
  len = strlen (s2);
183
0
  memmove (s + sign, s2, len);
184
0
  snprintf (s + sign + len, sizeof (buf) - (s + sign + len - buf), "E-%u", (unsigned) (strlen (s + sign) - 1));
185
0
      }
186
0
      hb_vector_t<char> nibbles;
187
0
      while (*s)
188
0
      {
189
0
  char c = s[0];
190
0
  s++;
191
0
192
0
  switch (c)
193
0
  {
194
0
    case 'E':
195
0
    {
196
0
      char c2 = *s;
197
0
      if (c2 == '-')
198
0
      {
199
0
        s++;
200
0
        nibbles.push (0x0C); // E-
201
0
      } else {
202
0
        if (c2 == '+')
203
0
    s++;
204
0
        nibbles.push (0x0B); // E
205
0
      }
206
0
      if (*s == '0')
207
0
        s++;
208
0
      continue;
209
0
    }
210
0
211
0
    case '.':
212
0
      nibbles.push (0x0A); // .
213
0
      continue;
214
0
215
0
    case '-':
216
0
      nibbles.push (0x0E); // -
217
0
      continue;
218
0
  }
219
0
220
0
  nibbles.push (c - '0');
221
0
      }
222
0
      nibbles.push (0x0F);
223
0
      if (nibbles.length % 2)
224
0
  nibbles.push (0x0F);
225
0
226
0
      unsigned count = nibbles.length;
227
0
      for (unsigned i = 0; i < count; i += 2)
228
0
        encode_byte ((nibbles[i] << 4) | nibbles[i+1]);
229
0
    }
230
0
  }
231
232
  void encode_op (op_code_t op)
233
0
  {
234
0
    if (Is_OpCode_ESC (op))
235
0
    {
236
0
      encode_byte (OpCode_escape);
237
0
      encode_byte (Unmake_OpCode_ESC (op));
238
0
    }
239
0
    else
240
0
      encode_byte (op);
241
0
  }
242
243
  void copy_str (const unsigned char *str, unsigned length)
244
0
  {
245
0
    assert ((signed) (buff.length + length) <= buff.allocated);
246
0
    hb_memcpy (buff.arrayZ + buff.length, str, length);
247
0
    buff.length += length;
248
0
  }
249
250
0
  bool in_error () const { return buff.in_error (); }
251
252
  protected:
253
254
  str_buff_t &buff;
255
};
256
257
struct cff_sub_table_info_t {
258
  cff_sub_table_info_t ()
259
    : fd_array_link (0),
260
      char_strings_link (0)
261
0
  {
262
0
    fd_select.init ();
263
0
  }
264
265
  table_info_t     fd_select;
266
  objidx_t         fd_array_link;
267
  objidx_t         char_strings_link;
268
};
269
270
template <typename OPSTR=op_str_t>
271
struct cff_top_dict_op_serializer_t : op_serializer_t
272
{
273
  bool serialize (hb_serialize_context_t *c,
274
      const OPSTR &opstr,
275
      const cff_sub_table_info_t &info) const
276
  {
277
    TRACE_SERIALIZE (this);
278
279
    switch (opstr.op)
280
    {
281
      case OpCode_CharStrings:
282
  return_trace (FontDict::serialize_link4_op(c, opstr.op, info.char_strings_link, whence_t::Absolute));
283
284
      case OpCode_FDArray:
285
  return_trace (FontDict::serialize_link4_op(c, opstr.op, info.fd_array_link, whence_t::Absolute));
286
287
      case OpCode_FDSelect:
288
  return_trace (FontDict::serialize_link4_op(c, opstr.op, info.fd_select.link, whence_t::Absolute));
289
290
      default:
291
  return_trace (copy_opstr (c, opstr));
292
    }
293
    return_trace (true);
294
  }
295
};
296
297
struct cff_font_dict_op_serializer_t : op_serializer_t
298
{
299
  bool serialize (hb_serialize_context_t *c,
300
      const op_str_t &opstr,
301
      const table_info_t &privateDictInfo) const
302
0
  {
303
0
    TRACE_SERIALIZE (this);
304
0
305
0
    if (opstr.op == OpCode_Private)
306
0
    {
307
0
      /* serialize the private dict size & offset as 2-byte & 4-byte integers */
308
0
      return_trace (UnsizedByteStr::serialize_int2 (c, privateDictInfo.size) &&
309
0
        Dict::serialize_link4_op (c, opstr.op, privateDictInfo.link, whence_t::Absolute));
310
0
    }
311
0
    else
312
0
    {
313
0
      unsigned char *d = c->allocate_size<unsigned char> (opstr.length);
314
0
      if (unlikely (!d)) return_trace (false);
315
0
      /* Faster than hb_memcpy for small strings. */
316
0
      for (unsigned i = 0; i < opstr.length; i++)
317
0
  d[i] = opstr.ptr[i];
318
0
      //hb_memcpy (d, opstr.ptr, opstr.length);
319
0
    }
320
0
    return_trace (true);
321
0
  }
322
};
323
324
struct flatten_param_t
325
{
326
  str_buff_t     &flatStr;
327
  bool  drop_hints;
328
  const hb_subset_plan_t *plan;
329
};
330
331
template <typename ACC, typename ENV, typename OPSET, op_code_t endchar_op=OpCode_Invalid>
332
struct subr_flattener_t
333
{
334
  subr_flattener_t (const ACC &acc_,
335
        const hb_subset_plan_t *plan_)
336
       : acc (acc_), plan (plan_) {}
337
338
  bool flatten (str_buff_vec_t &flat_charstrings)
339
  {
340
    unsigned count = plan->num_output_glyphs ();
341
    if (!flat_charstrings.resize_exact (count))
342
      return false;
343
    for (unsigned int i = 0; i < count; i++)
344
    {
345
      hb_codepoint_t  glyph;
346
      if (!plan->old_gid_for_new_gid (i, &glyph))
347
      {
348
  /* add an endchar only charstring for a missing glyph if CFF1 */
349
  if (endchar_op != OpCode_Invalid) flat_charstrings[i].push (endchar_op);
350
  continue;
351
      }
352
      const hb_ubytes_t str = (*acc.charStrings)[glyph];
353
      unsigned int fd = acc.fdSelect->get_fd (glyph);
354
      if (unlikely (fd >= acc.fdCount))
355
  return false;
356
357
358
      ENV env (str, acc, fd,
359
         plan->normalized_coords.arrayZ, plan->normalized_coords.length);
360
      cs_interpreter_t<ENV, OPSET, flatten_param_t> interp (env);
361
      flatten_param_t  param = {
362
        flat_charstrings.arrayZ[i],
363
        (bool) (plan->flags & HB_SUBSET_FLAGS_NO_HINTING),
364
  plan
365
      };
366
      if (unlikely (!interp.interpret (param)))
367
  return false;
368
    }
369
    return true;
370
  }
371
372
  const ACC &acc;
373
  const hb_subset_plan_t *plan;
374
};
375
376
struct subr_closures_t
377
{
378
  subr_closures_t (unsigned int fd_count) : global_closure (), local_closures ()
379
0
  {
380
0
    local_closures.resize_exact (fd_count);
381
0
  }
382
383
  void reset ()
384
0
  {
385
0
    global_closure.clear();
386
0
    for (unsigned int i = 0; i < local_closures.length; i++)
387
0
      local_closures[i].clear();
388
0
  }
389
390
0
  bool in_error () const { return local_closures.in_error (); }
391
  hb_set_t  global_closure;
392
  hb_vector_t<hb_set_t> local_closures;
393
};
394
395
struct parsed_cs_op_t : op_str_t
396
{
397
  parsed_cs_op_t (unsigned int subr_num_ = 0) :
398
0
    subr_num (subr_num_) {}
399
400
0
  bool is_hinting () const { return hinting_flag; }
401
0
  void set_hinting ()       { hinting_flag = true; }
402
403
  /* The layout of this struct is designed to fit within the
404
   * padding of op_str_t! */
405
406
  protected:
407
  bool    hinting_flag = false;
408
409
  public:
410
  uint16_t subr_num;
411
};
412
413
struct parsed_cs_str_t : parsed_values_t<parsed_cs_op_t>
414
{
415
  parsed_cs_str_t () :
416
    parsed (false),
417
    hint_dropped (false),
418
    has_prefix_ (false),
419
    has_calls_ (false)
420
0
  {
421
0
    SUPER::init ();
422
0
  }
423
424
  void add_op (op_code_t op, const byte_str_ref_t& str_ref)
425
0
  {
426
0
    if (!is_parsed ())
427
0
      SUPER::add_op (op, str_ref);
428
0
  }
429
430
  void add_call_op (op_code_t op, const byte_str_ref_t& str_ref, unsigned int subr_num)
431
0
  {
432
0
    if (!is_parsed ())
433
0
    {
434
0
      has_calls_ = true;
435
0
436
0
      /* Pop the subroutine number. */
437
0
      values.pop ();
438
0
439
0
      SUPER::add_op (op, str_ref, {subr_num});
440
0
    }
441
0
  }
442
443
  void set_prefix (const number_t &num, op_code_t op = OpCode_Invalid)
444
0
  {
445
0
    has_prefix_ = true;
446
0
    prefix_op_ = op;
447
0
    prefix_num_ = num;
448
0
  }
449
450
  bool at_end (unsigned int pos) const
451
0
  {
452
0
    return ((pos + 1 >= values.length) /* CFF2 */
453
0
  || (values[pos + 1].op == OpCode_return));
454
0
  }
455
456
0
  bool is_parsed () const { return parsed; }
457
0
  void set_parsed ()      { parsed = true; }
458
459
0
  bool is_hint_dropped () const { return hint_dropped; }
460
0
  void set_hint_dropped ()      { hint_dropped = true; }
461
462
0
  bool is_vsindex_dropped () const { return vsindex_dropped; }
463
0
  void set_vsindex_dropped ()      { vsindex_dropped = true; }
464
465
0
  bool has_prefix () const          { return has_prefix_; }
466
0
  op_code_t prefix_op () const         { return prefix_op_; }
467
0
  const number_t &prefix_num () const { return prefix_num_; }
468
469
0
  bool has_calls () const          { return has_calls_; }
470
471
  void compact ()
472
0
  {
473
0
    unsigned count = values.length;
474
0
    if (!count) return;
475
0
    auto &opstr = values.arrayZ;
476
0
    unsigned j = 0;
477
0
    for (unsigned i = 1; i < count; i++)
478
0
    {
479
0
      /* See if we can combine op j and op i. */
480
0
      bool combine =
481
0
        (opstr[j].op != OpCode_callsubr && opstr[j].op != OpCode_callgsubr) &&
482
0
        (opstr[i].op != OpCode_callsubr && opstr[i].op != OpCode_callgsubr) &&
483
0
        (opstr[j].is_hinting () == opstr[i].is_hinting ()) &&
484
0
        (opstr[j].ptr + opstr[j].length == opstr[i].ptr) &&
485
0
        (opstr[j].length + opstr[i].length <= 255);
486
0
487
0
      if (combine)
488
0
      {
489
0
  opstr[j].length += opstr[i].length;
490
0
  opstr[j].op = OpCode_Invalid;
491
0
      }
492
0
      else
493
0
      {
494
0
  opstr[++j] = opstr[i];
495
0
      }
496
0
    }
497
0
    values.shrink (j + 1);
498
0
  }
499
500
  protected:
501
  bool    parsed : 1;
502
  bool    hint_dropped : 1;
503
  bool    vsindex_dropped : 1;
504
  bool    has_prefix_ : 1;
505
  bool    has_calls_ : 1;
506
  op_code_t prefix_op_;
507
  number_t  prefix_num_;
508
509
  private:
510
  typedef parsed_values_t<parsed_cs_op_t> SUPER;
511
};
512
513
struct parsed_cs_str_vec_t : hb_vector_t<parsed_cs_str_t>
514
{
515
  private:
516
  typedef hb_vector_t<parsed_cs_str_t> SUPER;
517
};
518
519
struct cff_subset_accelerator_t
520
{
521
  static cff_subset_accelerator_t* create (
522
      hb_blob_t* original_blob,
523
      const parsed_cs_str_vec_t& parsed_charstrings,
524
      const parsed_cs_str_vec_t& parsed_global_subrs,
525
0
      const hb_vector_t<parsed_cs_str_vec_t>& parsed_local_subrs) {
526
0
    cff_subset_accelerator_t* accel =
527
0
        (cff_subset_accelerator_t*) hb_malloc (sizeof(cff_subset_accelerator_t));
528
0
    if (unlikely (!accel)) return nullptr;
529
0
    new (accel) cff_subset_accelerator_t (original_blob,
530
0
                                          parsed_charstrings,
531
0
                                          parsed_global_subrs,
532
0
                                          parsed_local_subrs);
533
0
    return accel;
534
0
  }
535
536
0
  static void destroy (void* value) {
537
0
    if (!value) return;
538
0
539
0
    cff_subset_accelerator_t* accel = (cff_subset_accelerator_t*) value;
540
0
    accel->~cff_subset_accelerator_t ();
541
0
    hb_free (accel);
542
0
  }
543
544
  cff_subset_accelerator_t(
545
      hb_blob_t* original_blob_,
546
      const parsed_cs_str_vec_t& parsed_charstrings_,
547
      const parsed_cs_str_vec_t& parsed_global_subrs_,
548
      const hb_vector_t<parsed_cs_str_vec_t>& parsed_local_subrs_)
549
0
  {
550
0
    parsed_charstrings = parsed_charstrings_;
551
0
    parsed_global_subrs = parsed_global_subrs_;
552
0
    parsed_local_subrs = parsed_local_subrs_;
553
0
554
0
    // the parsed charstrings point to memory in the original CFF table so we must hold a reference
555
0
    // to it to keep the memory valid.
556
0
    original_blob = hb_blob_reference (original_blob_);
557
0
  }
558
559
  ~cff_subset_accelerator_t()
560
0
  {
561
0
    hb_blob_destroy (original_blob);
562
0
    auto *mapping = glyph_to_sid_map.get_relaxed ();
563
0
    if (mapping)
564
0
    {
565
0
      mapping->~glyph_to_sid_map_t ();
566
0
      hb_free (mapping);
567
0
    }
568
0
  }
569
570
  parsed_cs_str_vec_t parsed_charstrings;
571
  parsed_cs_str_vec_t parsed_global_subrs;
572
  hb_vector_t<parsed_cs_str_vec_t> parsed_local_subrs;
573
  mutable hb_atomic_t<glyph_to_sid_map_t *> glyph_to_sid_map;
574
575
 private:
576
  hb_blob_t* original_blob;
577
};
578
579
struct subr_subset_param_t
580
{
581
  subr_subset_param_t (parsed_cs_str_t *parsed_charstring_,
582
           parsed_cs_str_vec_t *parsed_global_subrs_,
583
           parsed_cs_str_vec_t *parsed_local_subrs_,
584
           hb_set_t *global_closure_,
585
           hb_set_t *local_closure_,
586
           bool drop_hints_) :
587
      current_parsed_str (parsed_charstring_),
588
      parsed_charstring (parsed_charstring_),
589
      parsed_global_subrs (parsed_global_subrs_),
590
      parsed_local_subrs (parsed_local_subrs_),
591
      global_closure (global_closure_),
592
      local_closure (local_closure_),
593
0
      drop_hints (drop_hints_) {}
594
595
  parsed_cs_str_t *get_parsed_str_for_context (call_context_t &context)
596
0
  {
597
0
    switch (context.type)
598
0
    {
599
0
      case CSType_CharString:
600
0
  return parsed_charstring;
601
0
602
0
      case CSType_LocalSubr:
603
0
  if (likely (context.subr_num < parsed_local_subrs->length))
604
0
    return &(*parsed_local_subrs)[context.subr_num];
605
0
  break;
606
0
607
0
      case CSType_GlobalSubr:
608
0
  if (likely (context.subr_num < parsed_global_subrs->length))
609
0
    return &(*parsed_global_subrs)[context.subr_num];
610
0
  break;
611
0
    }
612
0
    return nullptr;
613
0
  }
614
615
  template <typename ENV>
616
  void set_current_str (ENV &env, bool calling)
617
  {
618
    parsed_cs_str_t *parsed_str = get_parsed_str_for_context (env.context);
619
    if (unlikely (!parsed_str))
620
    {
621
      env.set_error ();
622
      return;
623
    }
624
    /* If the called subroutine is parsed partially but not completely yet,
625
     * it must be because we are calling it recursively.
626
     * Handle it as an error. */
627
    if (unlikely (calling && !parsed_str->is_parsed () && (parsed_str->values.length > 0)))
628
      env.set_error ();
629
    else
630
    {
631
      if (!parsed_str->is_parsed ())
632
        parsed_str->alloc (env.str_ref.total_size ());
633
      current_parsed_str = parsed_str;
634
    }
635
  }
636
637
  parsed_cs_str_t *current_parsed_str;
638
639
  parsed_cs_str_t *parsed_charstring;
640
  parsed_cs_str_vec_t *parsed_global_subrs;
641
  parsed_cs_str_vec_t *parsed_local_subrs;
642
  hb_set_t      *global_closure;
643
  hb_set_t      *local_closure;
644
  bool    drop_hints;
645
};
646
647
struct subr_remap_t : hb_inc_bimap_t
648
{
649
  void create (const hb_set_t *closure)
650
0
  {
651
0
    /* create a remapping of subroutine numbers from old to new.
652
0
     * no optimization based on usage counts. fonttools doesn't appear doing that either.
653
0
     */
654
0
655
0
    alloc (closure->get_population ());
656
0
    for (auto old_num : *closure)
657
0
      add (old_num);
658
0
659
0
    if (get_population () < 1240)
660
0
      bias = 107;
661
0
    else if (get_population () < 33900)
662
0
      bias = 1131;
663
0
    else
664
0
      bias = 32768;
665
0
  }
666
667
  int biased_num (unsigned int old_num) const
668
0
  {
669
0
    hb_codepoint_t new_num = get (old_num);
670
0
    return (int)new_num - bias;
671
0
  }
672
673
  protected:
674
  int bias;
675
};
676
677
struct subr_remaps_t
678
{
679
  subr_remaps_t (unsigned int fdCount)
680
0
  {
681
0
    local_remaps.resize (fdCount);
682
0
  }
683
684
  bool in_error()
685
0
  {
686
0
    return local_remaps.in_error ();
687
0
  }
688
689
  void create (subr_closures_t& closures)
690
0
  {
691
0
    global_remap.create (&closures.global_closure);
692
0
    for (unsigned int i = 0; i < local_remaps.length; i++)
693
0
      local_remaps.arrayZ[i].create (&closures.local_closures[i]);
694
0
  }
695
696
  subr_remap_t         global_remap;
697
  hb_vector_t<subr_remap_t>  local_remaps;
698
};
699
700
template <typename SUBSETTER, typename SUBRS, typename ACC, typename ENV, typename OPSET, op_code_t endchar_op=OpCode_Invalid>
701
struct subr_subsetter_t
702
{
703
  subr_subsetter_t (ACC &acc_, const hb_subset_plan_t *plan_)
704
      : acc (acc_), plan (plan_), closures(acc_.fdCount),
705
        remaps(acc_.fdCount)
706
  {}
707
708
  /* Subroutine subsetting with --no-desubroutinize runs in phases:
709
   *
710
   * 1. execute charstrings/subroutines to determine subroutine closures
711
   * 2. parse out all operators and numbers
712
   * 3. mark hint operators and operands for removal if --no-hinting
713
   * 4. re-encode all charstrings and subroutines with new subroutine numbers
714
   *
715
   * Phases #1 and #2 are done at the same time in collect_subrs ().
716
   * Phase #3 walks charstrings/subroutines forward then backward (hence parsing required),
717
   * because we can't tell if a number belongs to a hint op until we see the first moveto.
718
   *
719
   * Assumption: a callsubr/callgsubr operator must immediately follow a (biased) subroutine number
720
   * within the same charstring/subroutine, e.g., not split across a charstring and a subroutine.
721
   */
722
  bool subset (void)
723
  {
724
    unsigned fd_count = acc.fdCount;
725
    const cff_subset_accelerator_t* cff_accelerator = nullptr;
726
    if (acc.cff_accelerator) {
727
      cff_accelerator = acc.cff_accelerator;
728
      fd_count = cff_accelerator->parsed_local_subrs.length;
729
    }
730
731
    if (cff_accelerator) {
732
      // If we are not dropping hinting then charstrings are not modified so we can
733
      // just use a reference to the cached copies.
734
      cached_charstrings.resize_exact (plan->num_output_glyphs ());
735
      parsed_global_subrs = &cff_accelerator->parsed_global_subrs;
736
      parsed_local_subrs = &cff_accelerator->parsed_local_subrs;
737
    } else {
738
      parsed_charstrings.resize_exact (plan->num_output_glyphs ());
739
      parsed_global_subrs_storage.resize_exact (acc.globalSubrs->count);
740
741
      if (unlikely (!parsed_local_subrs_storage.resize (fd_count))) return false;
742
743
      for (unsigned int i = 0; i < acc.fdCount; i++)
744
      {
745
        unsigned count = acc.privateDicts[i].localSubrs->count;
746
        parsed_local_subrs_storage[i].resize (count);
747
        if (unlikely (parsed_local_subrs_storage[i].in_error ())) return false;
748
      }
749
750
      parsed_global_subrs = &parsed_global_subrs_storage;
751
      parsed_local_subrs = &parsed_local_subrs_storage;
752
    }
753
754
    if (unlikely (remaps.in_error()
755
                  || cached_charstrings.in_error ()
756
                  || parsed_charstrings.in_error ()
757
                  || parsed_global_subrs->in_error ()
758
                  || closures.in_error ())) {
759
      return false;
760
    }
761
762
    /* phase 1 & 2 */
763
    for (auto _ : plan->new_to_old_gid_list)
764
    {
765
      hb_codepoint_t new_glyph = _.first;
766
      hb_codepoint_t old_glyph = _.second;
767
768
      const hb_ubytes_t str = (*acc.charStrings)[old_glyph];
769
      unsigned int fd = acc.fdSelect->get_fd (old_glyph);
770
      if (unlikely (fd >= acc.fdCount))
771
        return false;
772
773
      if (cff_accelerator)
774
      {
775
        // parsed string already exists in accelerator, copy it and move
776
        // on.
777
        if (cached_charstrings)
778
          cached_charstrings[new_glyph] = &cff_accelerator->parsed_charstrings[old_glyph];
779
        else
780
          parsed_charstrings[new_glyph] = cff_accelerator->parsed_charstrings[old_glyph];
781
782
        continue;
783
      }
784
785
      ENV env (str, acc, fd);
786
      cs_interpreter_t<ENV, OPSET, subr_subset_param_t> interp (env);
787
788
      parsed_charstrings[new_glyph].alloc (str.length);
789
      subr_subset_param_t  param (&parsed_charstrings[new_glyph],
790
                                  &parsed_global_subrs_storage,
791
                                  &parsed_local_subrs_storage[fd],
792
                                  &closures.global_closure,
793
                                  &closures.local_closures[fd],
794
                                  plan->flags & HB_SUBSET_FLAGS_NO_HINTING);
795
796
      if (unlikely (!interp.interpret (param)))
797
        return false;
798
799
      /* complete parsed string esp. copy CFF1 width or CFF2 vsindex to the parsed charstring for encoding */
800
      SUBSETTER::complete_parsed_str (interp.env, param, parsed_charstrings[new_glyph]);
801
802
      /* mark hint ops and arguments for drop */
803
      if ((plan->flags & HB_SUBSET_FLAGS_NO_HINTING) || plan->inprogress_accelerator)
804
      {
805
  subr_subset_param_t  param (&parsed_charstrings[new_glyph],
806
            &parsed_global_subrs_storage,
807
            &parsed_local_subrs_storage[fd],
808
            &closures.global_closure,
809
            &closures.local_closures[fd],
810
            plan->flags & HB_SUBSET_FLAGS_NO_HINTING);
811
812
  drop_hints_param_t  drop;
813
  if (drop_hints_in_str (parsed_charstrings[new_glyph], param, drop))
814
  {
815
    parsed_charstrings[new_glyph].set_hint_dropped ();
816
    if (drop.vsindex_dropped)
817
      parsed_charstrings[new_glyph].set_vsindex_dropped ();
818
  }
819
      }
820
821
      /* Doing this here one by one instead of compacting all at the end
822
       * has massive peak-memory saving.
823
       *
824
       * The compacting both saves memory and makes further operations
825
       * faster.
826
       */
827
      parsed_charstrings[new_glyph].compact ();
828
    }
829
830
    /* Since parsed strings were loaded from accelerator, we still need
831
     * to compute the subroutine closures which would have normally happened during
832
     * parsing.
833
     *
834
     * Or if we are dropping hinting, redo closure to get actually used subrs.
835
     */
836
    if ((cff_accelerator ||
837
  (!cff_accelerator && plan->flags & HB_SUBSET_FLAGS_NO_HINTING)) &&
838
        !closure_subroutines(*parsed_global_subrs,
839
                             *parsed_local_subrs))
840
      return false;
841
842
    remaps.create (closures);
843
844
    populate_subset_accelerator ();
845
    return true;
846
  }
847
848
  bool encode_charstrings (str_buff_vec_t &buffArray, bool encode_prefix = true) const
849
  {
850
    unsigned num_glyphs = plan->num_output_glyphs ();
851
    if (unlikely (!buffArray.resize_exact (num_glyphs)))
852
      return false;
853
    hb_codepoint_t last = 0;
854
    for (auto _ : plan->new_to_old_gid_list)
855
    {
856
      hb_codepoint_t gid = _.first;
857
      hb_codepoint_t old_glyph = _.second;
858
859
      if (endchar_op != OpCode_Invalid)
860
        for (; last < gid; last++)
861
  {
862
    // Hack to point vector to static string.
863
    auto &b = buffArray.arrayZ[last];
864
    b.length = 1;
865
    b.arrayZ = const_cast<unsigned char *>(endchar_str);
866
  }
867
868
      last++; // Skip over gid
869
      unsigned int  fd = acc.fdSelect->get_fd (old_glyph);
870
      if (unlikely (fd >= acc.fdCount))
871
  return false;
872
      if (unlikely (!encode_str (get_parsed_charstring (gid), fd, buffArray.arrayZ[gid], encode_prefix)))
873
  return false;
874
    }
875
    if (endchar_op != OpCode_Invalid)
876
      for (; last < num_glyphs; last++)
877
      {
878
  // Hack to point vector to static string.
879
  auto &b = buffArray.arrayZ[last];
880
  b.length = 1;
881
  b.arrayZ = const_cast<unsigned char *>(endchar_str);
882
      }
883
884
    return true;
885
  }
886
887
  bool encode_subrs (const parsed_cs_str_vec_t &subrs, const subr_remap_t& remap, unsigned int fd, str_buff_vec_t &buffArray) const
888
  {
889
    unsigned int  count = remap.get_population ();
890
891
    if (unlikely (!buffArray.resize_exact (count)))
892
      return false;
893
    for (unsigned int new_num = 0; new_num < count; new_num++)
894
    {
895
      hb_codepoint_t old_num = remap.backward (new_num);
896
      assert (old_num != CFF_UNDEF_CODE);
897
898
      if (unlikely (!encode_str (subrs[old_num], fd, buffArray[new_num])))
899
  return false;
900
    }
901
    return true;
902
  }
903
904
  bool encode_globalsubrs (str_buff_vec_t &buffArray)
905
  {
906
    return encode_subrs (*parsed_global_subrs, remaps.global_remap, 0, buffArray);
907
  }
908
909
  bool encode_localsubrs (unsigned int fd, str_buff_vec_t &buffArray) const
910
  {
911
    return encode_subrs ((*parsed_local_subrs)[fd], remaps.local_remaps[fd], fd, buffArray);
912
  }
913
914
  protected:
915
  struct drop_hints_param_t
916
  {
917
    drop_hints_param_t ()
918
      : seen_moveto (false),
919
  ends_in_hint (false),
920
  all_dropped (false),
921
  vsindex_dropped (false) {}
922
923
    bool  seen_moveto;
924
    bool  ends_in_hint;
925
    bool  all_dropped;
926
    bool  vsindex_dropped;
927
  };
928
929
  bool drop_hints_in_subr (parsed_cs_str_t &str, unsigned int pos,
930
         parsed_cs_str_vec_t &subrs, unsigned int subr_num,
931
         const subr_subset_param_t &param, drop_hints_param_t &drop)
932
  {
933
    drop.ends_in_hint = false;
934
    bool has_hint = drop_hints_in_str (subrs[subr_num], param, drop);
935
936
    /* if this subr ends with a stem hint (i.e., not a number; potential argument for moveto),
937
     * then this entire subroutine must be a hint. drop its call. */
938
    if (drop.ends_in_hint)
939
    {
940
      str.values[pos].set_hinting ();
941
      /* if this subr call is at the end of the parent subr, propagate the flag
942
       * otherwise reset the flag */
943
      if (!str.at_end (pos))
944
  drop.ends_in_hint = false;
945
    }
946
    else if (drop.all_dropped)
947
    {
948
      str.values[pos].set_hinting ();
949
    }
950
951
    return has_hint;
952
  }
953
954
  /* returns true if it sees a hint op before the first moveto */
955
  bool drop_hints_in_str (parsed_cs_str_t &str, const subr_subset_param_t &param, drop_hints_param_t &drop)
956
  {
957
    bool  seen_hint = false;
958
959
    unsigned count = str.values.length;
960
    auto *values = str.values.arrayZ;
961
    for (unsigned int pos = 0; pos < count; pos++)
962
    {
963
      bool  has_hint = false;
964
      switch (values[pos].op)
965
      {
966
  case OpCode_callsubr:
967
    has_hint = drop_hints_in_subr (str, pos,
968
          *param.parsed_local_subrs, values[pos].subr_num,
969
          param, drop);
970
    break;
971
972
  case OpCode_callgsubr:
973
    has_hint = drop_hints_in_subr (str, pos,
974
          *param.parsed_global_subrs, values[pos].subr_num,
975
          param, drop);
976
    break;
977
978
  case OpCode_rmoveto:
979
  case OpCode_hmoveto:
980
  case OpCode_vmoveto:
981
    drop.seen_moveto = true;
982
    break;
983
984
  case OpCode_hintmask:
985
  case OpCode_cntrmask:
986
    if (drop.seen_moveto)
987
    {
988
      values[pos].set_hinting ();
989
      break;
990
    }
991
    HB_FALLTHROUGH;
992
993
  case OpCode_hstemhm:
994
  case OpCode_vstemhm:
995
  case OpCode_hstem:
996
  case OpCode_vstem:
997
    has_hint = true;
998
    values[pos].set_hinting ();
999
    if (str.at_end (pos))
1000
      drop.ends_in_hint = true;
1001
    break;
1002
1003
  case OpCode_dotsection:
1004
    values[pos].set_hinting ();
1005
    break;
1006
1007
  default:
1008
    /* NONE */
1009
    break;
1010
      }
1011
      if (has_hint)
1012
      {
1013
  for (int i = pos - 1; i >= 0; i--)
1014
  {
1015
    parsed_cs_op_t  &csop = values[(unsigned)i];
1016
    if (csop.is_hinting ())
1017
      break;
1018
    csop.set_hinting ();
1019
    if (csop.op == OpCode_vsindexcs)
1020
      drop.vsindex_dropped = true;
1021
  }
1022
  seen_hint |= has_hint;
1023
      }
1024
    }
1025
1026
    /* Raise all_dropped flag if all operators except return are dropped from a subr.
1027
     * It may happen even after seeing the first moveto if a subr contains
1028
     * only (usually one) hintmask operator, then calls to this subr can be dropped.
1029
     */
1030
    drop.all_dropped = true;
1031
    for (unsigned int pos = 0; pos < count; pos++)
1032
    {
1033
      parsed_cs_op_t  &csop = values[pos];
1034
      if (csop.op == OpCode_return)
1035
  break;
1036
      if (!csop.is_hinting ())
1037
      {
1038
  drop.all_dropped = false;
1039
  break;
1040
      }
1041
    }
1042
1043
    return seen_hint;
1044
  }
1045
1046
  bool closure_subroutines (const parsed_cs_str_vec_t& global_subrs,
1047
                            const hb_vector_t<parsed_cs_str_vec_t>& local_subrs)
1048
  {
1049
    closures.reset ();
1050
    for (auto _ : plan->new_to_old_gid_list)
1051
    {
1052
      hb_codepoint_t new_glyph = _.first;
1053
      hb_codepoint_t old_glyph = _.second;
1054
      unsigned int fd = acc.fdSelect->get_fd (old_glyph);
1055
      if (unlikely (fd >= acc.fdCount))
1056
        return false;
1057
1058
      // Note: const cast is safe here because the collect_subr_refs_in_str only performs a
1059
      //       closure and does not modify any of the charstrings.
1060
      subr_subset_param_t  param (const_cast<parsed_cs_str_t*> (&get_parsed_charstring (new_glyph)),
1061
                                  const_cast<parsed_cs_str_vec_t*> (&global_subrs),
1062
                                  const_cast<parsed_cs_str_vec_t*> (&local_subrs[fd]),
1063
                                  &closures.global_closure,
1064
                                  &closures.local_closures[fd],
1065
                                  plan->flags & HB_SUBSET_FLAGS_NO_HINTING);
1066
      collect_subr_refs_in_str (get_parsed_charstring (new_glyph), param);
1067
    }
1068
1069
    return true;
1070
  }
1071
1072
  void collect_subr_refs_in_subr (unsigned int subr_num, parsed_cs_str_vec_t &subrs,
1073
          hb_set_t *closure,
1074
          const subr_subset_param_t &param)
1075
  {
1076
    if (closure->has (subr_num))
1077
      return;
1078
    closure->add (subr_num);
1079
    collect_subr_refs_in_str (subrs[subr_num], param);
1080
  }
1081
1082
  void collect_subr_refs_in_str (const parsed_cs_str_t &str,
1083
                                 const subr_subset_param_t &param)
1084
  {
1085
    if (!str.has_calls ())
1086
      return;
1087
1088
    for (auto &opstr : str.values)
1089
    {
1090
      if (!param.drop_hints || !opstr.is_hinting ())
1091
      {
1092
  switch (opstr.op)
1093
  {
1094
    case OpCode_callsubr:
1095
      collect_subr_refs_in_subr (opstr.subr_num, *param.parsed_local_subrs,
1096
               param.local_closure, param);
1097
      break;
1098
1099
    case OpCode_callgsubr:
1100
      collect_subr_refs_in_subr (opstr.subr_num, *param.parsed_global_subrs,
1101
               param.global_closure, param);
1102
      break;
1103
1104
    default: break;
1105
  }
1106
      }
1107
    }
1108
  }
1109
1110
  bool encode_str (const parsed_cs_str_t &str, const unsigned int fd, str_buff_t &buff, bool encode_prefix = true) const
1111
  {
1112
    str_encoder_t  encoder (buff);
1113
    encoder.reset ();
1114
    bool hinting = !(plan->flags & HB_SUBSET_FLAGS_NO_HINTING);
1115
    /* if a prefix (CFF1 width or CFF2 vsindex) has been removed along with hints,
1116
     * re-insert it at the beginning of charstreing */
1117
    if (encode_prefix && str.has_prefix () && !hinting && str.is_hint_dropped ())
1118
    {
1119
      encoder.encode_num_cs (str.prefix_num ());
1120
      if (str.prefix_op () != OpCode_Invalid)
1121
  encoder.encode_op (str.prefix_op ());
1122
    }
1123
1124
    unsigned size = 0;
1125
    for (auto &opstr : str.values)
1126
    {
1127
      size += opstr.length;
1128
      if (opstr.op == OpCode_callsubr || opstr.op == OpCode_callgsubr)
1129
        size += 3;
1130
    }
1131
    if (!buff.alloc_exact (buff.length + size))
1132
      return false;
1133
1134
    for (auto &opstr : str.values)
1135
    {
1136
      if (hinting || !opstr.is_hinting ())
1137
      {
1138
  switch (opstr.op)
1139
  {
1140
    case OpCode_callsubr:
1141
      encoder.encode_int (remaps.local_remaps[fd].biased_num (opstr.subr_num));
1142
      encoder.copy_str (opstr.ptr, opstr.length);
1143
      break;
1144
1145
    case OpCode_callgsubr:
1146
      encoder.encode_int (remaps.global_remap.biased_num (opstr.subr_num));
1147
      encoder.copy_str (opstr.ptr, opstr.length);
1148
      break;
1149
1150
    default:
1151
      encoder.copy_str (opstr.ptr, opstr.length);
1152
      break;
1153
  }
1154
      }
1155
    }
1156
    return !encoder.in_error ();
1157
  }
1158
1159
  void compact_parsed_subrs () const
1160
  {
1161
    for (auto &cs : parsed_global_subrs_storage)
1162
      cs.compact ();
1163
    for (auto &vec : parsed_local_subrs_storage)
1164
      for (auto &cs : vec)
1165
  cs.compact ();
1166
  }
1167
1168
  void populate_subset_accelerator () const
1169
  {
1170
    if (!plan->inprogress_accelerator) return;
1171
1172
    compact_parsed_subrs ();
1173
1174
    acc.cff_accelerator =
1175
        cff_subset_accelerator_t::create(acc.blob,
1176
                                         parsed_charstrings,
1177
                                         parsed_global_subrs_storage,
1178
                                         parsed_local_subrs_storage);
1179
  }
1180
1181
  const parsed_cs_str_t& get_parsed_charstring (unsigned i) const
1182
  {
1183
    if (cached_charstrings) return *(cached_charstrings[i]);
1184
    return parsed_charstrings[i];
1185
  }
1186
1187
  protected:
1188
  const ACC     &acc;
1189
  const hb_subset_plan_t  *plan;
1190
1191
  subr_closures_t   closures;
1192
1193
  hb_vector_t<const parsed_cs_str_t*>     cached_charstrings;
1194
  const parsed_cs_str_vec_t*              parsed_global_subrs;
1195
  const hb_vector_t<parsed_cs_str_vec_t>* parsed_local_subrs;
1196
1197
  subr_remaps_t     remaps;
1198
1199
  private:
1200
1201
  parsed_cs_str_vec_t   parsed_charstrings;
1202
  parsed_cs_str_vec_t   parsed_global_subrs_storage;
1203
  hb_vector_t<parsed_cs_str_vec_t>  parsed_local_subrs_storage;
1204
  typedef typename SUBRS::count_type subr_count_type;
1205
};
1206
1207
} /* namespace CFF */
1208
1209
HB_INTERNAL bool
1210
hb_plan_subset_cff_fdselect (const hb_subset_plan_t *plan,
1211
          unsigned int fdCount,
1212
          const CFF::FDSelect &src, /* IN */
1213
          unsigned int &subset_fd_count /* OUT */,
1214
          unsigned int &subset_fdselect_size /* OUT */,
1215
          unsigned int &subset_fdselect_format /* OUT */,
1216
          hb_vector_t<CFF::code_pair_t> &fdselect_ranges /* OUT */,
1217
          hb_inc_bimap_t &fdmap /* OUT */);
1218
1219
HB_INTERNAL bool
1220
hb_serialize_cff_fdselect (hb_serialize_context_t *c,
1221
        unsigned int num_glyphs,
1222
        const CFF::FDSelect &src,
1223
        unsigned int fd_count,
1224
        unsigned int fdselect_format,
1225
        unsigned int size,
1226
        const hb_vector_t<CFF::code_pair_t> &fdselect_ranges);
1227
1228
#endif /* HB_SUBSET_CFF_COMMON_HH */