/*
 * Copyright © 2014  Google, Inc.
 *
 *  This is part of HarfBuzz, a text shaping library.
 *
 * Permission is hereby granted, without written agreement and without
 * license or royalty fees, to use, copy, modify, and distribute this
 * software and its documentation for any purpose, provided that the
 * above copyright notice and the following two paragraphs appear in
 * all copies of this software.
 *
 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
 * DAMAGE.
 *
 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
 * FITNESS FOR A PARTICULAR PURPOSE.  THE SOFTWARE PROVIDED HEREUNDER IS
 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
 *
 * Google Author(s): Behdad Esfahbod
 */

#ifndef HB_OT_CMAP_TABLE_HH
#define HB_OT_CMAP_TABLE_HH

#include "hb-ot-os2-table.hh"
#include "hb-ot-shaper-arabic-pua.hh"
#include "hb-open-type.hh"
#include "hb-set.hh"

/*
 * cmap -- Character to Glyph Index Mapping
 * https://docs.microsoft.com/en-us/typography/opentype/spec/cmap
 */
#define HB_OT_TAG_cmap HB_TAG('c','m','a','p')

namespace OT {


struct CmapSubtableFormat0
{
  bool get_glyph (hb_codepoint_t codepoint, hb_codepoint_t *glyph) const
  {
    hb_codepoint_t gid = codepoint < 256 ? glyphIdArray[codepoint] : 0;
    if (unlikely (!gid))
      return false;
    *glyph = gid;
    return true;
  }

  unsigned get_language () const
  {
    return language;
  }

  void collect_unicodes (hb_set_t *out) const
  {
    for (unsigned int i = 0; i < 256; i++)
      if (glyphIdArray[i])
	out->add (i);
  }

  void collect_mapping (hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping /* OUT */) const
  {
    for (unsigned i = 0; i < 256; i++)
      if (glyphIdArray[i])
      {
	hb_codepoint_t glyph = glyphIdArray[i];
	unicodes->add (i);
	mapping->set (i, glyph);
      }
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this));
  }

  protected:
  HBUINT16	format;		/* Format number is set to 0. */
  HBUINT16	length;		/* Byte length of this subtable. */
  HBUINT16	language;	/* Ignore. */
  HBUINT8	glyphIdArray[256];/* An array that maps character
				 * code to glyph index values. */
  public:
  DEFINE_SIZE_STATIC (6 + 256);
};

struct CmapSubtableFormat4
{


  template<typename Iterator,
      typename Writer,
	   hb_requires (hb_is_iterator (Iterator))>
  void to_ranges (Iterator it, Writer& range_writer)
  {
    hb_codepoint_t start_cp = 0, prev_run_start_cp = 0, run_start_cp = 0, end_cp = 0, last_gid = 0;
    int run_length = 0 , delta = 0, prev_delta = 0;

    enum {
      FIRST_SUB_RANGE,
      FOLLOWING_SUB_RANGE,
    } mode;

    while (it) {
      // Start a new range
      {
        const auto& pair = *it;
        start_cp = pair.first;
        prev_run_start_cp = start_cp;
        run_start_cp = start_cp;
        end_cp = start_cp;
        last_gid = pair.second;
        run_length = 1;
        prev_delta = 0;
      }

      delta = last_gid - start_cp;
      mode = FIRST_SUB_RANGE;
      it++;

      while (it) {
        // Process range
        const auto& pair = *it;
        hb_codepoint_t next_cp = pair.first;
        hb_codepoint_t next_gid = pair.second;
        if (next_cp != end_cp + 1) {
          // Current range is over, stop processing.
          break;
        }

        if (next_gid == last_gid + 1) {
          // The current run continues.
          end_cp = next_cp;
          run_length++;
          last_gid = next_gid;
          it++;
          continue;
        }

        // A new run is starting, decide if we want to commit the current run.
        int split_cost = (mode == FIRST_SUB_RANGE) ? 8 : 16;
        int run_cost = run_length * 2;
        if (run_cost >= split_cost) {
          commit_current_range(start_cp,
                               prev_run_start_cp,
                               run_start_cp,
                               end_cp,
                               delta,
                               prev_delta,
                               split_cost,
                               range_writer);
          start_cp = next_cp;
        }

        // Start the new run
        mode = FOLLOWING_SUB_RANGE;
        prev_run_start_cp = run_start_cp;
        run_start_cp = next_cp;
        end_cp = next_cp;
        prev_delta = delta;
        delta = next_gid - run_start_cp;
        run_length = 1;
        last_gid = next_gid;
        it++;
      }

      // Finalize range
      commit_current_range (start_cp,
                            prev_run_start_cp,
                            run_start_cp,
                            end_cp,
                            delta,
                            prev_delta,
                            8,
                            range_writer);
    }

    if (likely (end_cp != 0xFFFF)) {
      range_writer (0xFFFF, 0xFFFF, 1);
    }
  }

  /*
   * Writes the current range as either one or two ranges depending on what is most efficient.
   */
  template<typename Writer>
  void commit_current_range (hb_codepoint_t start,
                             hb_codepoint_t prev_run_start,
                             hb_codepoint_t run_start,
                             hb_codepoint_t end,
                             int run_delta,
                             int previous_run_delta,
                             int split_cost,
                             Writer& range_writer) {
    bool should_split = false;
    if (start < run_start && run_start < end) {
      int run_cost = (end - run_start + 1) * 2;
      if (run_cost >= split_cost) {
        should_split = true;
      }
    }

    // TODO(grieger): handle case where delta is legitimately 0, mark range offset array instead?
    if (should_split) {
      if (start == prev_run_start)
        range_writer (start, run_start - 1, previous_run_delta);
      else
        range_writer (start, run_start - 1, 0);
      range_writer (run_start, end, run_delta);
      return;
    }


    if (start == run_start) {
      // Range is only a run
      range_writer (start, end, run_delta);
      return;
    }

    // Write only a single non-run range.
    range_writer (start, end, 0);
  }

  template<typename Iterator,
	   hb_requires (hb_is_iterator (Iterator))>
  unsigned serialize_find_segcount (Iterator it) {
    struct Counter {
      unsigned segcount = 0;

      void operator() (hb_codepoint_t start,
                       hb_codepoint_t end,
                       int delta) {
        segcount++;
      }
    } counter;

    to_ranges (+it, counter);
    return counter.segcount;
  }


  template<typename Iterator,
	   hb_requires (hb_is_iterator (Iterator))>
  bool serialize_start_end_delta_arrays (hb_serialize_context_t *c,
                                         Iterator it,
                                         int segcount)
  {
    struct Writer {
      hb_serialize_context_t *serializer_;
      HBUINT16* end_code_;
      HBUINT16* start_code_;
      HBINT16* id_delta_;
      int index_;

      Writer(hb_serialize_context_t *serializer)
          : serializer_(serializer),
            end_code_(nullptr),
            start_code_(nullptr),
            id_delta_(nullptr),
            index_ (0) {}
      void operator() (hb_codepoint_t start,
                       hb_codepoint_t end,
                       int delta) {
        start_code_[index_] = start;
        end_code_[index_] = end;
        id_delta_[index_] = delta;
        index_++;
      }
    } writer(c);

    writer.end_code_ = c->allocate_size<HBUINT16> (HBUINT16::static_size * segcount);
    c->allocate_size<HBUINT16> (2); // padding
    writer.start_code_ = c->allocate_size<HBUINT16> (HBUINT16::static_size * segcount);
    writer.id_delta_ = c->allocate_size<HBINT16> (HBINT16::static_size * segcount);

    if (unlikely (!writer.end_code_ || !writer.start_code_ || !writer.id_delta_)) return false;

    to_ranges (+it, writer);
    return true;
  }

  template<typename Iterator,
          hb_requires (hb_is_iterator (Iterator))>
  HBUINT16* serialize_rangeoffset_glyid (hb_serialize_context_t *c,
                                         Iterator it,
					 HBUINT16 *endCode,
					 HBUINT16 *startCode,
					 HBINT16 *idDelta,
					 unsigned segcount)
  {
    hb_map_t cp_to_gid { it };

    HBUINT16 *idRangeOffset = c->allocate_size<HBUINT16> (HBUINT16::static_size * segcount);
    if (unlikely (!c->check_success (idRangeOffset))) return nullptr;
    if (unlikely ((char *)idRangeOffset - (char *)idDelta != (int) segcount * (int) HBINT16::static_size)) return nullptr;

    for (unsigned i : + hb_range (segcount)
		      | hb_filter ([&] (const unsigned _) { return idDelta[_] == 0; }))
    {
      idRangeOffset[i] = 2 * (c->start_embed<HBUINT16> () - idRangeOffset - i);
      for (hb_codepoint_t cp = startCode[i]; cp <= endCode[i]; cp++)
      {
        HBUINT16 gid;
        gid = cp_to_gid[cp];
        c->copy<HBUINT16> (gid);
      }
    }

    return idRangeOffset;
  }

  template<typename Iterator,
	   hb_requires (hb_is_iterator (Iterator))>
  void serialize (hb_serialize_context_t *c,
		  Iterator it)
  {
    auto format4_iter =
    + it
    | hb_filter ([&] (const hb_pair_t<hb_codepoint_t, hb_codepoint_t> _)
		 { return _.first <= 0xFFFF; })
    ;

    if (!format4_iter) return;

    unsigned table_initpos = c->length ();
    if (unlikely (!c->extend_min (this))) return;
    this->format = 4;

    hb_vector_t<hb_pair_t<hb_codepoint_t, hb_codepoint_t>> cp_to_gid {
      format4_iter
    };

    //serialize endCode[], startCode[], idDelta[]
    HBUINT16* endCode = c->start_embed<HBUINT16> ();
    unsigned segcount = serialize_find_segcount (cp_to_gid.iter());
    if (unlikely (!serialize_start_end_delta_arrays (c, cp_to_gid.iter(), segcount)))
      return;

    HBUINT16 *startCode = endCode + segcount + 1;
    HBINT16 *idDelta = ((HBINT16*)startCode) + segcount;

    HBUINT16 *idRangeOffset = serialize_rangeoffset_glyid (c,
                                                           cp_to_gid.iter (),
                                                           endCode,
                                                           startCode,
                                                           idDelta,
                                                           segcount);
    if (unlikely (!c->check_success (idRangeOffset))) return;

    this->length = c->length () - table_initpos;
    if ((long long) this->length != (long long) c->length () - table_initpos)
    {
      // Length overflowed. Discard the current object before setting the error condition, otherwise
      // discard is a noop which prevents the higher level code from reverting the serializer to the
      // pre-error state in cmap4 overflow handling code.
      c->pop_discard ();
      c->err (HB_SERIALIZE_ERROR_INT_OVERFLOW);
      return;
    }

    this->segCountX2 = segcount * 2;
    this->entrySelector = hb_max (1u, hb_bit_storage (segcount)) - 1;
    this->searchRange = 2 * (1u << this->entrySelector);
    this->rangeShift = segcount * 2 > this->searchRange
		       ? 2 * segcount - this->searchRange
		       : 0;
  }

  unsigned get_language () const
  {
    return language;
  }

  struct accelerator_t
  {
    accelerator_t () {}
    accelerator_t (const CmapSubtableFormat4 *subtable) { init (subtable); }

    void init (const CmapSubtableFormat4 *subtable)
    {
      segCount = subtable->segCountX2 / 2;
      endCount = subtable->values.arrayZ;
      startCount = endCount + segCount + 1;
      idDelta = startCount + segCount;
      idRangeOffset = idDelta + segCount;
      glyphIdArray = idRangeOffset + segCount;
      glyphIdArrayLength = (subtable->length - 16 - 8 * segCount) / 2;
    }

    bool get_glyph (hb_codepoint_t codepoint, hb_codepoint_t *glyph) const
    {
      struct CustomRange
      {
	int cmp (hb_codepoint_t k,
		 unsigned distance) const
	{
	  if (k > last) return +1;
	  if (k < (&last)[distance]) return -1;
	  return 0;
	}
	HBUINT16 last;
      };

      const HBUINT16 *found = hb_bsearch (codepoint,
					  this->endCount,
					  this->segCount,
					  2,
					  _hb_cmp_method<hb_codepoint_t, CustomRange, unsigned>,
					  this->segCount + 1);
      if (unlikely (!found))
	return false;
      unsigned int i = found - endCount;

      hb_codepoint_t gid;
      unsigned int rangeOffset = this->idRangeOffset[i];
      if (rangeOffset == 0)
	gid = codepoint + this->idDelta[i];
      else
      {
	/* Somebody has been smoking... */
	unsigned int index = rangeOffset / 2 + (codepoint - this->startCount[i]) + i - this->segCount;
	if (unlikely (index >= this->glyphIdArrayLength))
	  return false;
	gid = this->glyphIdArray[index];
	if (unlikely (!gid))
	  return false;
	gid += this->idDelta[i];
      }
      gid &= 0xFFFFu;
      if (unlikely (!gid))
	return false;
      *glyph = gid;
      return true;
    }

    HB_INTERNAL static bool get_glyph_func (const void *obj, hb_codepoint_t codepoint, hb_codepoint_t *glyph)
    { return ((const accelerator_t *) obj)->get_glyph (codepoint, glyph); }

    void collect_unicodes (hb_set_t *out) const
    {
      unsigned int count = this->segCount;
      if (count && this->startCount[count - 1] == 0xFFFFu)
	count--; /* Skip sentinel segment. */
      for (unsigned int i = 0; i < count; i++)
      {
	hb_codepoint_t start = this->startCount[i];
	hb_codepoint_t end = this->endCount[i];
	unsigned int rangeOffset = this->idRangeOffset[i];
        out->add_range(start, end);
	if (rangeOffset == 0)
	{
	  for (hb_codepoint_t codepoint = start; codepoint <= end; codepoint++)
	  {
	    hb_codepoint_t gid = (codepoint + this->idDelta[i]) & 0xFFFFu;
	    if (unlikely (!gid))
              out->del(codepoint);
	  }
	}
	else
	{
	  for (hb_codepoint_t codepoint = start; codepoint <= end; codepoint++)
	  {
	    unsigned int index = rangeOffset / 2 + (codepoint - this->startCount[i]) + i - this->segCount;
	    if (unlikely (index >= this->glyphIdArrayLength))
            {
              out->del_range (codepoint, end);
	      break;
            }
	    hb_codepoint_t gid = this->glyphIdArray[index];
	    if (unlikely (!gid))
              out->del(codepoint);
	  }
	}
      }
    }

    void collect_mapping (hb_set_t *unicodes, /* OUT */
			  hb_map_t *mapping /* OUT */) const
    {
      // TODO(grieger): optimize similar to collect_unicodes
      // (ie. use add_range())
      unsigned count = this->segCount;
      if (count && this->startCount[count - 1] == 0xFFFFu)
	count--; /* Skip sentinel segment. */
      for (unsigned i = 0; i < count; i++)
      {
	hb_codepoint_t start = this->startCount[i];
	hb_codepoint_t end = this->endCount[i];
	unsigned rangeOffset = this->idRangeOffset[i];
	if (rangeOffset == 0)
	{
	  for (hb_codepoint_t codepoint = start; codepoint <= end; codepoint++)
	  {
	    hb_codepoint_t gid = (codepoint + this->idDelta[i]) & 0xFFFFu;
	    if (unlikely (!gid))
	      continue;
	    unicodes->add (codepoint);
	    mapping->set (codepoint, gid);
	  }
	}
	else
	{
	  for (hb_codepoint_t codepoint = start; codepoint <= end; codepoint++)
	  {
	    unsigned index = rangeOffset / 2 + (codepoint - this->startCount[i]) + i - this->segCount;
	    if (unlikely (index >= this->glyphIdArrayLength))
	      break;
	    hb_codepoint_t gid = this->glyphIdArray[index];
	    if (unlikely (!gid))
	      continue;
	    unicodes->add (codepoint);
	    mapping->set (codepoint, gid);
	  }
	}
      }
    }

    const HBUINT16 *endCount;
    const HBUINT16 *startCount;
    const HBUINT16 *idDelta;
    const HBUINT16 *idRangeOffset;
    const HBUINT16 *glyphIdArray;
    unsigned int segCount;
    unsigned int glyphIdArrayLength;
  };

  bool get_glyph (hb_codepoint_t codepoint, hb_codepoint_t *glyph) const
  {
    accelerator_t accel (this);
    return accel.get_glyph_func (&accel, codepoint, glyph);
  }
  void collect_unicodes (hb_set_t *out) const
  {
    accelerator_t accel (this);
    accel.collect_unicodes (out);
  }

  void collect_mapping (hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping /* OUT */) const
  {
    accelerator_t accel (this);
    accel.collect_mapping (unicodes, mapping);
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    if (unlikely (!c->check_struct (this)))
      return_trace (false);

    if (unlikely (!c->check_range (this, length)))
    {
      /* Some broken fonts have too long of a "length" value.
       * If that is the case, just change the value to truncate
       * the subtable at the end of the blob. */
      uint16_t new_length = (uint16_t) hb_min ((uintptr_t) 65535,
					       (uintptr_t) (c->end -
							    (char *) this));
      if (!c->try_set (&length, new_length))
	return_trace (false);
    }

    return_trace (16 + 4 * (unsigned int) segCountX2 <= length);
  }


  protected:
  HBUINT16	format;		/* Format number is set to 4. */
  HBUINT16	length;		/* This is the length in bytes of the
				 * subtable. */
  HBUINT16	language;	/* Ignore. */
  HBUINT16	segCountX2;	/* 2 x segCount. */
  HBUINT16	searchRange;	/* 2 * (2**floor(log2(segCount))) */
  HBUINT16	entrySelector;	/* log2(searchRange/2) */
  HBUINT16	rangeShift;	/* 2 x segCount - searchRange */

  UnsizedArrayOf<HBUINT16>
		values;
#if 0
  HBUINT16	endCount[segCount];	/* End characterCode for each segment,
					 * last=0xFFFFu. */
  HBUINT16	reservedPad;		/* Set to 0. */
  HBUINT16	startCount[segCount];	/* Start character code for each segment. */
  HBINT16		idDelta[segCount];	/* Delta for all character codes in segment. */
  HBUINT16	idRangeOffset[segCount];/* Offsets into glyphIdArray or 0 */
  UnsizedArrayOf<HBUINT16>
		glyphIdArray;	/* Glyph index array (arbitrary length) */
#endif

  public:
  DEFINE_SIZE_ARRAY (14, values);
};

struct CmapSubtableLongGroup
{
  friend struct CmapSubtableFormat12;
  friend struct CmapSubtableFormat13;
  template<typename U>
  friend struct CmapSubtableLongSegmented;
  friend struct cmap;

  int cmp (hb_codepoint_t codepoint) const
  {
    if (codepoint < startCharCode) return -1;
    if (codepoint > endCharCode)   return +1;
    return 0;
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this));
  }

  private:
  HBUINT32		startCharCode;	/* First character code in this group. */
  HBUINT32		endCharCode;	/* Last character code in this group. */
  HBUINT32		glyphID;	/* Glyph index; interpretation depends on
					 * subtable format. */
  public:
  DEFINE_SIZE_STATIC (12);
};
DECLARE_NULL_NAMESPACE_BYTES (OT, CmapSubtableLongGroup);

template <typename UINT>
struct CmapSubtableTrimmed
{
  bool get_glyph (hb_codepoint_t codepoint, hb_codepoint_t *glyph) const
  {
    /* Rely on our implicit array bound-checking. */
    hb_codepoint_t gid = glyphIdArray[codepoint - startCharCode];
    if (unlikely (!gid))
      return false;
    *glyph = gid;
    return true;
  }

  unsigned get_language () const
  {
    return language;
  }

  void collect_unicodes (hb_set_t *out) const
  {
    hb_codepoint_t start = startCharCode;
    unsigned int count = glyphIdArray.len;
    for (unsigned int i = 0; i < count; i++)
      if (glyphIdArray[i])
	out->add (start + i);
  }

  void collect_mapping (hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping /* OUT */) const
  {
    hb_codepoint_t start_cp = startCharCode;
    unsigned count = glyphIdArray.len;
    for (unsigned i = 0; i < count; i++)
      if (glyphIdArray[i])
      {
	hb_codepoint_t unicode = start_cp + i;
	hb_codepoint_t glyphid = glyphIdArray[i];
	unicodes->add (unicode);
	mapping->set (unicode, glyphid);
      }
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this) && glyphIdArray.sanitize (c));
  }

  protected:
  UINT		formatReserved;	/* Subtable format and (maybe) padding. */
  UINT		length;		/* Byte length of this subtable. */
  UINT		language;	/* Ignore. */
  UINT		startCharCode;	/* First character code covered. */
  ArrayOf<HBGlyphID16, UINT>
		glyphIdArray;	/* Array of glyph index values for character
				 * codes in the range. */
  public:
  DEFINE_SIZE_ARRAY (5 * sizeof (UINT), glyphIdArray);
};

struct CmapSubtableFormat6  : CmapSubtableTrimmed<HBUINT16> {};
struct CmapSubtableFormat10 : CmapSubtableTrimmed<HBUINT32> {};

template <typename T>
struct CmapSubtableLongSegmented
{
  friend struct cmap;

  bool get_glyph (hb_codepoint_t codepoint, hb_codepoint_t *glyph) const
  {
    hb_codepoint_t gid = T::group_get_glyph (groups.bsearch (codepoint), codepoint);
    if (unlikely (!gid))
      return false;
    *glyph = gid;
    return true;
  }

  unsigned get_language () const
  {
    return language;
  }

  void collect_unicodes (hb_set_t *out, unsigned int num_glyphs) const
  {
    for (unsigned int i = 0; i < this->groups.len; i++)
    {
      hb_codepoint_t start = this->groups[i].startCharCode;
      hb_codepoint_t end = hb_min ((hb_codepoint_t) this->groups[i].endCharCode,
				   (hb_codepoint_t) HB_UNICODE_MAX);
      hb_codepoint_t gid = this->groups[i].glyphID;
      if (!gid)
      {
	/* Intention is: if (hb_is_same (T, CmapSubtableFormat13)) continue; */
	if (! T::group_get_glyph (this->groups[i], end)) continue;
	start++;
	gid++;
      }
      if (unlikely ((unsigned int) gid >= num_glyphs)) continue;
      if (unlikely ((unsigned int) (gid + end - start) >= num_glyphs))
	end = start + (hb_codepoint_t) num_glyphs - gid;

      out->add_range (start, hb_min (end, 0x10FFFFu));
    }
  }

  void collect_mapping (hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping, /* OUT */
			unsigned num_glyphs) const
  {
    hb_codepoint_t last_end = 0;
    for (unsigned i = 0; i < this->groups.len; i++)
    {
      hb_codepoint_t start = this->groups[i].startCharCode;
      hb_codepoint_t end = hb_min ((hb_codepoint_t) this->groups[i].endCharCode,
				   (hb_codepoint_t) HB_UNICODE_MAX);
      if (unlikely (start > end || start < last_end)) {
        // Range is not in order and is invalid, skip it.
        continue;
      }
      last_end = end;


      hb_codepoint_t gid = this->groups[i].glyphID;
      if (!gid)
      {
	/* Intention is: if (hb_is_same (T, CmapSubtableFormat13)) continue; */
	if (! T::group_get_glyph (this->groups[i], end)) continue;
	start++;
	gid++;
      }
      if (unlikely ((unsigned int) gid >= num_glyphs)) continue;
      if (unlikely ((unsigned int) (gid + end - start) >= num_glyphs))
	end = start + (hb_codepoint_t) num_glyphs - gid;

      for (unsigned cp = start; cp <= end; cp++)
      {
	unicodes->add (cp);
	mapping->set (cp, gid);
	gid++;
      }
    }
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this) && groups.sanitize (c));
  }

  protected:
  HBUINT16	format;		/* Subtable format; set to 12. */
  HBUINT16	reserved;	/* Reserved; set to 0. */
  HBUINT32	length;		/* Byte length of this subtable. */
  HBUINT32	language;	/* Ignore. */
  SortedArray32Of<CmapSubtableLongGroup>
		groups;		/* Groupings. */
  public:
  DEFINE_SIZE_ARRAY (16, groups);
};

struct CmapSubtableFormat12 : CmapSubtableLongSegmented<CmapSubtableFormat12>
{
  static hb_codepoint_t group_get_glyph (const CmapSubtableLongGroup &group,
					 hb_codepoint_t u)
  { return likely (group.startCharCode <= group.endCharCode) ?
	   group.glyphID + (u - group.startCharCode) : 0; }


  template<typename Iterator,
	   hb_requires (hb_is_iterator (Iterator))>
  void serialize (hb_serialize_context_t *c,
		  Iterator it)
  {
    if (!it) return;
    unsigned table_initpos = c->length ();
    if (unlikely (!c->extend_min (this))) return;

    hb_codepoint_t startCharCode = (hb_codepoint_t) -1, endCharCode = (hb_codepoint_t) -1;
    hb_codepoint_t glyphID = 0;

    for (const auto& _ : +it)
    {
      if (startCharCode == (hb_codepoint_t) -1)
      {
	startCharCode = _.first;
	endCharCode = _.first;
	glyphID = _.second;
      }
      else if (!_is_gid_consecutive (endCharCode, startCharCode, glyphID, _.first, _.second))
      {
	CmapSubtableLongGroup  grouprecord;
	grouprecord.startCharCode = startCharCode;
	grouprecord.endCharCode = endCharCode;
	grouprecord.glyphID = glyphID;
	c->copy<CmapSubtableLongGroup> (grouprecord);

	startCharCode = _.first;
	endCharCode = _.first;
	glyphID = _.second;
      }
      else
	endCharCode = _.first;
    }

    CmapSubtableLongGroup record;
    record.startCharCode = startCharCode;
    record.endCharCode = endCharCode;
    record.glyphID = glyphID;
    c->copy<CmapSubtableLongGroup> (record);

    this->format = 12;
    this->reserved = 0;
    this->length = c->length () - table_initpos;
    this->groups.len = (this->length - min_size) / CmapSubtableLongGroup::static_size;
  }

  static size_t get_sub_table_size (const hb_sorted_vector_t<CmapSubtableLongGroup> &groups_data)
  { return 16 + 12 * groups_data.length; }

  private:
  static bool _is_gid_consecutive (hb_codepoint_t endCharCode,
				   hb_codepoint_t startCharCode,
				   hb_codepoint_t glyphID,
				   hb_codepoint_t cp,
				   hb_codepoint_t new_gid)
  {
    return (cp - 1 == endCharCode) &&
	new_gid == glyphID + (cp - startCharCode);
  }

};

struct CmapSubtableFormat13 : CmapSubtableLongSegmented<CmapSubtableFormat13>
{
  static hb_codepoint_t group_get_glyph (const CmapSubtableLongGroup &group,
					 hb_codepoint_t u HB_UNUSED)
  { return group.glyphID; }
};

typedef enum
{
  GLYPH_VARIANT_NOT_FOUND = 0,
  GLYPH_VARIANT_FOUND = 1,
  GLYPH_VARIANT_USE_DEFAULT = 2
} glyph_variant_t;

struct UnicodeValueRange
{
  int cmp (const hb_codepoint_t &codepoint) const
  {
    if (codepoint < startUnicodeValue) return -1;
    if (codepoint > startUnicodeValue + additionalCount) return +1;
    return 0;
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this));
  }

  HBUINT24	startUnicodeValue;	/* First value in this range. */
  HBUINT8	additionalCount;	/* Number of additional values in this
					 * range. */
  public:
  DEFINE_SIZE_STATIC (4);
};

struct DefaultUVS : SortedArray32Of<UnicodeValueRange>
{
  void collect_unicodes (hb_set_t *out) const
  {
    unsigned int count = len;
    for (unsigned int i = 0; i < count; i++)
    {
      hb_codepoint_t first = arrayZ[i].startUnicodeValue;
      hb_codepoint_t last = hb_min ((hb_codepoint_t) (first + arrayZ[i].additionalCount),
				    (hb_codepoint_t) HB_UNICODE_MAX);
      out->add_range (first, last);
    }
  }

  DefaultUVS* copy (hb_serialize_context_t *c,
		    const hb_set_t *unicodes) const
  {
    DefaultUVS *out = c->start_embed<DefaultUVS> ();
    if (unlikely (!out)) return nullptr;
    auto snap = c->snapshot ();

    HBUINT32 len;
    len = 0;
    if (unlikely (!c->copy<HBUINT32> (len))) return nullptr;
    unsigned init_len = c->length ();

    if (this->len > unicodes->get_population () * hb_bit_storage ((unsigned) this->len))
    {
      hb_codepoint_t start = HB_SET_VALUE_INVALID;
      hb_codepoint_t end = HB_SET_VALUE_INVALID;

      for (hb_codepoint_t u = HB_SET_VALUE_INVALID;
	   unicodes->next (&u);)
      {
        if (!as_array ().bsearch (u))
	  continue;
	if (start == HB_SET_VALUE_INVALID)
	{
	  start = u;
	  end = start - 1;
	}
	if (end + 1 != u || end - start == 255)
        {
	  UnicodeValueRange rec;
	  rec.startUnicodeValue = start;
	  rec.additionalCount = end - start;
	  c->copy<UnicodeValueRange> (rec);
	  start = u;
	}
	end = u;
      }
      if (start != HB_SET_VALUE_INVALID)
      {
	UnicodeValueRange rec;
	rec.startUnicodeValue = start;
	rec.additionalCount = end - start;
	c->copy<UnicodeValueRange> (rec);
      }

    }
    else
    {
      hb_codepoint_t lastCode = HB_SET_VALUE_INVALID;
      int count = -1;

      for (const UnicodeValueRange& _ : *this)
      {
	hb_codepoint_t curEntry = (hb_codepoint_t) (_.startUnicodeValue - 1);
	hb_codepoint_t end = curEntry + _.additionalCount + 2;

	for (; unicodes->next (&curEntry) && curEntry < end;)
	{
	  count += 1;
	  if (lastCode == HB_SET_VALUE_INVALID)
	    lastCode = curEntry;
	  else if (lastCode + count != curEntry)
	  {
	    UnicodeValueRange rec;
	    rec.startUnicodeValue = lastCode;
	    rec.additionalCount = count - 1;
	    c->copy<UnicodeValueRange> (rec);

	    lastCode = curEntry;
	    count = 0;
	  }
	}
      }

      if (lastCode != HB_MAP_VALUE_INVALID)
      {
	UnicodeValueRange rec;
	rec.startUnicodeValue = lastCode;
	rec.additionalCount = count;
	c->copy<UnicodeValueRange> (rec);
      }
    }

    if (c->length () - init_len == 0)
    {
      c->revert (snap);
      return nullptr;
    }
    else
    {
      if (unlikely (!c->check_assign (out->len,
                                      (c->length () - init_len) / UnicodeValueRange::static_size,
                                      HB_SERIALIZE_ERROR_INT_OVERFLOW))) return nullptr;
      return out;
    }
  }

  public:
  DEFINE_SIZE_ARRAY (4, *this);
};

struct UVSMapping
{
  int cmp (const hb_codepoint_t &codepoint) const
  { return unicodeValue.cmp (codepoint); }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this));
  }

  HBUINT24	unicodeValue;	/* Base Unicode value of the UVS */
  HBGlyphID16	glyphID;	/* Glyph ID of the UVS */
  public:
  DEFINE_SIZE_STATIC (5);
};

struct NonDefaultUVS : SortedArray32Of<UVSMapping>
{
  void collect_unicodes (hb_set_t *out) const
  {
    for (const auto& a : as_array ())
      out->add (a.unicodeValue);
  }

  void collect_mapping (hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping /* OUT */) const
  {
    for (const auto& a : as_array ())
    {
      hb_codepoint_t unicode = a.unicodeValue;
      hb_codepoint_t glyphid = a.glyphID;
      unicodes->add (unicode);
      mapping->set (unicode, glyphid);
    }
  }

  void closure_glyphs (const hb_set_t      *unicodes,
		       hb_set_t            *glyphset) const
  {
    + as_array ()
    | hb_filter (unicodes, &UVSMapping::unicodeValue)
    | hb_map (&UVSMapping::glyphID)
    | hb_sink (glyphset)
    ;
  }

  NonDefaultUVS* copy (hb_serialize_context_t *c,
		       const hb_set_t *unicodes,
		       const hb_set_t *glyphs_requested,
		       const hb_map_t *glyph_map) const
  {
    NonDefaultUVS *out = c->start_embed<NonDefaultUVS> ();
    if (unlikely (!out)) return nullptr;

    auto it =
    + as_array ()
    | hb_filter ([&] (const UVSMapping& _)
		 {
		   return unicodes->has (_.unicodeValue) || glyphs_requested->has (_.glyphID);
		 })
    ;

    if (!it) return nullptr;

    HBUINT32 len;
    len = it.len ();
    if (unlikely (!c->copy<HBUINT32> (len))) return nullptr;

    for (const UVSMapping& _ : it)
    {
      UVSMapping mapping;
      mapping.unicodeValue = _.unicodeValue;
      mapping.glyphID = glyph_map->get (_.glyphID);
      c->copy<UVSMapping> (mapping);
    }

    return out;
  }

  public:
  DEFINE_SIZE_ARRAY (4, *this);
};

struct VariationSelectorRecord
{
  glyph_variant_t get_glyph (hb_codepoint_t codepoint,
			     hb_codepoint_t *glyph,
			     const void *base) const
  {
    if ((base+defaultUVS).bfind (codepoint))
      return GLYPH_VARIANT_USE_DEFAULT;
    const UVSMapping &nonDefault = (base+nonDefaultUVS).bsearch (codepoint);
    if (nonDefault.glyphID)
    {
      *glyph = nonDefault.glyphID;
       return GLYPH_VARIANT_FOUND;
    }
    return GLYPH_VARIANT_NOT_FOUND;
  }

  VariationSelectorRecord(const VariationSelectorRecord& other)
  {
    *this = other;
  }

  void operator= (const VariationSelectorRecord& other)
  {
    varSelector = other.varSelector;
    HBUINT32 offset = other.defaultUVS;
    defaultUVS = offset;
    offset = other.nonDefaultUVS;
    nonDefaultUVS = offset;
  }

  void collect_unicodes (hb_set_t *out, const void *base) const
  {
    (base+defaultUVS).collect_unicodes (out);
    (base+nonDefaultUVS).collect_unicodes (out);
  }

  void collect_mapping (const void *base,
			hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping /* OUT */) const
  {
    (base+defaultUVS).collect_unicodes (unicodes);
    (base+nonDefaultUVS).collect_mapping (unicodes, mapping);
  }

  int cmp (const hb_codepoint_t &variation_selector) const
  { return varSelector.cmp (variation_selector); }

  bool sanitize (hb_sanitize_context_t *c, const void *base) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this) &&
		  defaultUVS.sanitize (c, base) &&
		  nonDefaultUVS.sanitize (c, base));
  }

  hb_pair_t<unsigned, unsigned>
  copy (hb_serialize_context_t *c,
	const hb_set_t *unicodes,
	const hb_set_t *glyphs_requested,
	const hb_map_t *glyph_map,
	const void *base) const
  {
    auto snap = c->snapshot ();
    auto *out = c->embed<VariationSelectorRecord> (*this);
    if (unlikely (!out)) return hb_pair (0, 0);

    out->defaultUVS = 0;
    out->nonDefaultUVS = 0;

    unsigned non_default_uvs_objidx = 0;
    if (nonDefaultUVS != 0)
    {
      c->push ();
      if (c->copy (base+nonDefaultUVS, unicodes, glyphs_requested, glyph_map))
	non_default_uvs_objidx = c->pop_pack ();
      else c->pop_discard ();
    }

    unsigned default_uvs_objidx = 0;
    if (defaultUVS != 0)
    {
      c->push ();
      if (c->copy (base+defaultUVS, unicodes))
	default_uvs_objidx = c->pop_pack ();
      else c->pop_discard ();
    }


    if (!default_uvs_objidx && !non_default_uvs_objidx)
      c->revert (snap);

    return hb_pair (default_uvs_objidx, non_default_uvs_objidx);
  }

  HBUINT24	varSelector;	/* Variation selector. */
  Offset32To<DefaultUVS>
		defaultUVS;	/* Offset to Default UVS Table.  May be 0. */
  Offset32To<NonDefaultUVS>
		nonDefaultUVS;	/* Offset to Non-Default UVS Table.  May be 0. */
  public:
  DEFINE_SIZE_STATIC (11);
};

struct CmapSubtableFormat14
{
  glyph_variant_t get_glyph_variant (hb_codepoint_t codepoint,
				     hb_codepoint_t variation_selector,
				     hb_codepoint_t *glyph) const
  { return record.bsearch (variation_selector).get_glyph (codepoint, glyph, this); }

  void collect_variation_selectors (hb_set_t *out) const
  {
    for (const auto& a : record.as_array ())
      out->add (a.varSelector);
  }
  void collect_variation_unicodes (hb_codepoint_t variation_selector,
				   hb_set_t *out) const
  { record.bsearch (variation_selector).collect_unicodes (out, this); }

  void serialize (hb_serialize_context_t *c,
		  const hb_set_t *unicodes,
		  const hb_set_t *glyphs_requested,
		  const hb_map_t *glyph_map,
		  const void *base)
  {
    auto snap = c->snapshot ();
    unsigned table_initpos = c->length ();
    const char* init_tail = c->tail;

    if (unlikely (!c->extend_min (this))) return;
    this->format = 14;

    auto src_tbl = reinterpret_cast<const CmapSubtableFormat14*> (base);

    /*
     * Some versions of OTS require that offsets are in order. Due to the use
     * of push()/pop_pack() serializing the variation records in order results
     * in the offsets being in reverse order (first record has the largest
     * offset). While this is perfectly valid, it will cause some versions of
     * OTS to consider this table bad.
     *
     * So to prevent this issue we serialize the variation records in reverse
     * order, so that the offsets are ordered from small to large. Since
     * variation records are supposed to be in increasing order of varSelector
     * we then have to reverse the order of the written variation selector
     * records after everything is finalized.
     */
    hb_vector_t<hb_pair_t<unsigned, unsigned>> obj_indices;
    for (int i = src_tbl->record.len - 1; i >= 0; i--)
    {
      hb_pair_t<unsigned, unsigned> result = src_tbl->record[i].copy (c, unicodes, glyphs_requested, glyph_map, base);
      if (result.first || result.second)
	obj_indices.push (result);
    }

    if (c->length () - table_initpos == CmapSubtableFormat14::min_size)
    {
      c->revert (snap);
      return;
    }

    if (unlikely (!c->check_success (!obj_indices.in_error ())))
      return;

    int tail_len = init_tail - c->tail;
    c->check_assign (this->length, c->length () - table_initpos + tail_len,
                     HB_SERIALIZE_ERROR_INT_OVERFLOW);
    c->check_assign (this->record.len,
		     (c->length () - table_initpos - CmapSubtableFormat14::min_size) /
		     VariationSelectorRecord::static_size,
                     HB_SERIALIZE_ERROR_INT_OVERFLOW);

    /* Correct the incorrect write order by reversing the order of the variation
       records array. */
    _reverse_variation_records ();

    /* Now that records are in the right order, we can set up the offsets. */
    _add_links_to_variation_records (c, obj_indices);
  }

  void _reverse_variation_records ()
  {
    record.as_array ().reverse ();
  }

  void _add_links_to_variation_records (hb_serialize_context_t *c,
					const hb_vector_t<hb_pair_t<unsigned, unsigned>>& obj_indices)
  {
    for (unsigned i = 0; i < obj_indices.length; i++)
    {
      /*
       * Since the record array has been reversed (see comments in copy())
       * but obj_indices has not been, the indices at obj_indices[i]
       * are for the variation record at record[j].
       */
      int j = obj_indices.length - 1 - i;
      c->add_link (record[j].defaultUVS, obj_indices[i].first);
      c->add_link (record[j].nonDefaultUVS, obj_indices[i].second);
    }
  }

  void closure_glyphs (const hb_set_t      *unicodes,
		       hb_set_t            *glyphset) const
  {
    + hb_iter (record)
    | hb_filter (hb_bool, &VariationSelectorRecord::nonDefaultUVS)
    | hb_map (&VariationSelectorRecord::nonDefaultUVS)
    | hb_map (hb_add (this))
    | hb_apply ([=] (const NonDefaultUVS& _) { _.closure_glyphs (unicodes, glyphset); })
    ;
  }

  void collect_unicodes (hb_set_t *out) const
  {
    for (const VariationSelectorRecord& _ : record)
      _.collect_unicodes (out, this);
  }

  void collect_mapping (hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping /* OUT */) const
  {
    for (const VariationSelectorRecord& _ : record)
      _.collect_mapping (this, unicodes, mapping);
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this) &&
		  record.sanitize (c, this));
  }

  protected:
  HBUINT16	format;		/* Format number is set to 14. */
  HBUINT32	length;		/* Byte length of this subtable. */
  SortedArray32Of<VariationSelectorRecord>
		record;		/* Variation selector records; sorted
				 * in increasing order of `varSelector'. */
  public:
  DEFINE_SIZE_ARRAY (10, record);
};

struct CmapSubtable
{
  /* Note: We intentionally do NOT implement subtable formats 2 and 8. */

  bool get_glyph (hb_codepoint_t codepoint,
		  hb_codepoint_t *glyph) const
  {
    switch (u.format) {
    case  0: return u.format0 .get_glyph (codepoint, glyph);
    case  4: return u.format4 .get_glyph (codepoint, glyph);
    case  6: return u.format6 .get_glyph (codepoint, glyph);
    case 10: return u.format10.get_glyph (codepoint, glyph);
    case 12: return u.format12.get_glyph (codepoint, glyph);
    case 13: return u.format13.get_glyph (codepoint, glyph);
    case 14:
    default: return false;
    }
  }
  void collect_unicodes (hb_set_t *out, unsigned int num_glyphs = UINT_MAX) const
  {
    switch (u.format) {
    case  0: u.format0 .collect_unicodes (out); return;
    case  4: u.format4 .collect_unicodes (out); return;
    case  6: u.format6 .collect_unicodes (out); return;
    case 10: u.format10.collect_unicodes (out); return;
    case 12: u.format12.collect_unicodes (out, num_glyphs); return;
    case 13: u.format13.collect_unicodes (out, num_glyphs); return;
    case 14:
    default: return;
    }
  }

  void collect_mapping (hb_set_t *unicodes, /* OUT */
			hb_map_t *mapping, /* OUT */
			unsigned num_glyphs = UINT_MAX) const
  {
    switch (u.format) {
    case  0: u.format0 .collect_mapping (unicodes, mapping); return;
    case  4: u.format4 .collect_mapping (unicodes, mapping); return;
    case  6: u.format6 .collect_mapping (unicodes, mapping); return;
    case 10: u.format10.collect_mapping (unicodes, mapping); return;
    case 12: u.format12.collect_mapping (unicodes, mapping, num_glyphs); return;
    case 13: u.format13.collect_mapping (unicodes, mapping, num_glyphs); return;
    case 14:
    default: return;
    }
  }

  unsigned get_language () const
  {
    switch (u.format) {
    case  0: return u.format0 .get_language ();
    case  4: return u.format4 .get_language ();
    case  6: return u.format6 .get_language ();
    case 10: return u.format10.get_language ();
    case 12: return u.format12.get_language ();
    case 13: return u.format13.get_language ();
    case 14:
    default: return 0;
    }
  }

  template<typename Iterator,
	   hb_requires (hb_is_iterator (Iterator))>
  void serialize (hb_serialize_context_t *c,
		  Iterator it,
		  unsigned format,
		  const hb_subset_plan_t *plan,
		  const void *base)
  {
    switch (format) {
    case  4: return u.format4.serialize (c, it);
    case 12: return u.format12.serialize (c, it);
    case 14: return u.format14.serialize (c, plan->unicodes, plan->glyphs_requested, plan->glyph_map, base);
    default: return;
    }
  }

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    if (!u.format.sanitize (c)) return_trace (false);
    switch (u.format) {
    case  0: return_trace (u.format0 .sanitize (c));
    case  4: return_trace (u.format4 .sanitize (c));
    case  6: return_trace (u.format6 .sanitize (c));
    case 10: return_trace (u.format10.sanitize (c));
    case 12: return_trace (u.format12.sanitize (c));
    case 13: return_trace (u.format13.sanitize (c));
    case 14: return_trace (u.format14.sanitize (c));
    default:return_trace (true);
    }
  }

  public:
  union {
  HBUINT16		format;		/* Format identifier */
  CmapSubtableFormat0	format0;
  CmapSubtableFormat4	format4;
  CmapSubtableFormat6	format6;
  CmapSubtableFormat10	format10;
  CmapSubtableFormat12	format12;
  CmapSubtableFormat13	format13;
  CmapSubtableFormat14	format14;
  } u;
  public:
  DEFINE_SIZE_UNION (2, format);
};


struct EncodingRecord
{
  int cmp (const EncodingRecord &other) const
  {
    int ret;
    ret = platformID.cmp (other.platformID);
    if (ret) return ret;
    ret = encodingID.cmp (other.encodingID);
    if (ret) return ret;
    return 0;
  }

  bool sanitize (hb_sanitize_context_t *c, const void *base) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this) &&
		  subtable.sanitize (c, base));
  }

  template<typename Iterator,
	   hb_requires (hb_is_iterator (Iterator))>
  EncodingRecord* copy (hb_serialize_context_t *c,
			Iterator it,
			unsigned format,
			const void *base,
			const hb_subset_plan_t *plan,
			/* INOUT */ unsigned *objidx) const
  {
    TRACE_SERIALIZE (this);
    auto snap = c->snapshot ();
    auto *out = c->embed (this);
    if (unlikely (!out)) return_trace (nullptr);
    out->subtable = 0;

    if (*objidx == 0)
    {
      CmapSubtable *cmapsubtable = c->push<CmapSubtable> ();
      unsigned origin_length = c->length ();
      cmapsubtable->serialize (c, it, format, plan, &(base+subtable));
      if (c->length () - origin_length > 0) *objidx = c->pop_pack ();
      else c->pop_discard ();
    }

    if (*objidx == 0)
    {
      c->revert (snap);
      return_trace (nullptr);
    }

    c->add_link (out->subtable, *objidx);
    return_trace (out);
  }

  HBUINT16	platformID;	/* Platform ID. */
  HBUINT16	encodingID;	/* Platform-specific encoding ID. */
  Offset32To<CmapSubtable>
		subtable;	/* Byte offset from beginning of table to the subtable for this encoding. */
  public:
  DEFINE_SIZE_STATIC (8);
};

struct cmap;

struct SubtableUnicodesCache {

 private:
  hb_blob_ptr_t<cmap> base_blob;
  const char* base;
  hb_hashmap_t<unsigned, hb::unique_ptr<hb_set_t>> cached_unicodes;

 public:

  static SubtableUnicodesCache* create (hb_blob_ptr_t<cmap> source_table)
  {
    SubtableUnicodesCache* cache =
        (SubtableUnicodesCache*) hb_malloc (sizeof(SubtableUnicodesCache));
    new (cache) SubtableUnicodesCache (source_table);
    return cache;
  }

  static void destroy (void* value) {
    if (!value) return;

    SubtableUnicodesCache* cache = (SubtableUnicodesCache*) value;
    cache->~SubtableUnicodesCache ();
    hb_free (cache);
  }

  SubtableUnicodesCache(const void* cmap_base)
      : base_blob(),
        base ((const char*) cmap_base),
        cached_unicodes ()
  {}

  SubtableUnicodesCache(hb_blob_ptr_t<cmap> base_blob_)
      : base_blob(base_blob_),
        base ((const char *) base_blob.get()),
        cached_unicodes ()
  {}

  ~SubtableUnicodesCache()
  {
    base_blob.destroy ();
  }

  bool same_base(const void* other) const
  {
    return other == (const void*) base;
  }

  const hb_set_t* set_for (const EncodingRecord* record,
                           SubtableUnicodesCache& mutable_cache) const
  {
    if (cached_unicodes.has ((unsigned) ((const char *) record - base)))
      return cached_unicodes.get ((unsigned) ((const char *) record - base));

    return mutable_cache.set_for (record);
  }

  const hb_set_t* set_for (const EncodingRecord* record)
  {
    if (!cached_unicodes.has ((unsigned) ((const char *) record - base)))
    {
      hb_set_t *s = hb_set_create ();
      if (unlikely (s->in_error ()))
	return hb_set_get_empty ();

      (base+record->subtable).collect_unicodes (s);

      if (unlikely (!cached_unicodes.set ((unsigned) ((const char *) record - base), hb::unique_ptr<hb_set_t> {s})))
        return hb_set_get_empty ();

      return s;
    }
    return cached_unicodes.get ((unsigned) ((const char *) record - base));
  }

};

static inline uint_fast16_t
_hb_symbol_pua_map (unsigned codepoint)
{
  if (codepoint <= 0x00FFu)
  {
    /* For symbol-encoded OpenType fonts, we duplicate the
     * U+F000..F0FF range at U+0000..U+00FF.  That's what
     * Windows seems to do, and that's hinted about at:
     * https://docs.microsoft.com/en-us/typography/opentype/spec/recom
     * under "Non-Standard (Symbol) Fonts". */
    return 0xF000u + codepoint;
  }
  return 0;
}

struct cmap
{
  static constexpr hb_tag_t tableTag = HB_OT_TAG_cmap;


  static SubtableUnicodesCache* create_filled_cache(hb_blob_ptr_t<cmap> source_table) {
    const cmap* cmap = source_table.get();
    auto it =
    + hb_iter (cmap->encodingRecord)
    | hb_filter ([&](const EncodingRecord& _) {
      return cmap::filter_encoding_records_for_subset (cmap, _);
    })
    ;

    SubtableUnicodesCache* cache = SubtableUnicodesCache::create(source_table);
    for (const EncodingRecord& _ : it)
      cache->set_for(&_); // populate the cache for this encoding record.

    return cache;
  }

  template<typename Iterator, typename EncodingRecIter,
	   hb_requires (hb_is_iterator (EncodingRecIter))>
  bool serialize (hb_serialize_context_t *c,
		  Iterator it,
		  EncodingRecIter encodingrec_iter,
		  const void *base,
		  hb_subset_plan_t *plan,
                  bool drop_format_4 = false)
  {
    if (unlikely (!c->extend_min ((*this))))  return false;
    this->version = 0;

    unsigned format4objidx = 0, format12objidx = 0, format14objidx = 0;
    auto snap = c->snapshot ();

    SubtableUnicodesCache local_unicodes_cache (base);
    const SubtableUnicodesCache* unicodes_cache = &local_unicodes_cache;

    if (plan->accelerator &&
        plan->accelerator->cmap_cache &&
        plan->accelerator->cmap_cache->same_base (base))
      unicodes_cache = plan->accelerator->cmap_cache;

    for (const EncodingRecord& _ : encodingrec_iter)
    {
      if (c->in_error ())
        return false;

      unsigned format = (base+_.subtable).u.format;
      if (format != 4 && format != 12 && format != 14) continue;

      const hb_set_t* unicodes_set = unicodes_cache->set_for (&_, local_unicodes_cache);

      if (!drop_format_4 && format == 4)
      {
        c->copy (_, + it | hb_filter (*unicodes_set, hb_first), 4u, base, plan, &format4objidx);
        if (c->in_error () && c->only_overflow ())
        {
          // cmap4 overflowed, reset and retry serialization without format 4 subtables.
          c->revert (snap);
          return serialize (c, it,
                            encodingrec_iter,
                            base,
                            plan,
                            true);
        }
      }

      else if (format == 12)
      {
        if (_can_drop (_,
                       *unicodes_set,
                       base,
                       *unicodes_cache,
                       local_unicodes_cache,
                       + it | hb_map (hb_first), encodingrec_iter))
          continue;
        c->copy (_, + it | hb_filter (*unicodes_set, hb_first), 12u, base, plan, &format12objidx);
      }
      else if (format == 14) c->copy (_, it, 14u, base, plan, &format14objidx);
    }
    c->check_assign(this->encodingRecord.len,
                    (c->length () - cmap::min_size)/EncodingRecord::static_size,
                    HB_SERIALIZE_ERROR_INT_OVERFLOW);

    // Fail if format 4 was dropped and there is no cmap12.
    return !drop_format_4 || format12objidx;
  }

  template<typename Iterator, typename EncodingRecordIterator,
      hb_requires (hb_is_iterator (Iterator)),
      hb_requires (hb_is_iterator (EncodingRecordIterator))>
  bool _can_drop (const EncodingRecord& cmap12,
                  const hb_set_t& cmap12_unicodes,
                  const void* base,
                  const SubtableUnicodesCache& unicodes_cache,
                  SubtableUnicodesCache& local_unicodes_cache,
                  Iterator subset_unicodes,
                  EncodingRecordIterator encoding_records)
  {
    for (auto cp : + subset_unicodes | hb_filter (cmap12_unicodes))
    {
      if (cp >= 0x10000) return false;
    }

    unsigned target_platform;
    unsigned target_encoding;
    unsigned target_language = (base+cmap12.subtable).get_language ();

    if (cmap12.platformID == 0 && cmap12.encodingID == 4)
    {
      target_platform = 0;
      target_encoding = 3;
    } else if (cmap12.platformID == 3 && cmap12.encodingID == 10) {
      target_platform = 3;
      target_encoding = 1;
    } else {
      return false;
    }

    for (const auto& _ : encoding_records)
    {
      if (_.platformID != target_platform
          || _.encodingID != target_encoding
          || (base+_.subtable).get_language() != target_language)
        continue;

      const hb_set_t* sibling_unicodes = unicodes_cache.set_for (&_, local_unicodes_cache);

      auto cmap12 = + subset_unicodes | hb_filter (cmap12_unicodes);
      auto sibling = + subset_unicodes | hb_filter (*sibling_unicodes);
      for (; cmap12 && sibling; cmap12++, sibling++)
      {
        unsigned a = *cmap12;
        unsigned b = *sibling;
        if (a != b) return false;
      }

      return !cmap12 && !sibling;
    }

    return false;
  }

  void closure_glyphs (const hb_set_t      *unicodes,
		       hb_set_t            *glyphset) const
  {
    + hb_iter (encodingRecord)
    | hb_map (&EncodingRecord::subtable)
    | hb_map (hb_add (this))
    | hb_filter ([&] (const CmapSubtable& _) { return _.u.format == 14; })
    | hb_apply ([=] (const CmapSubtable& _) { _.u.format14.closure_glyphs (unicodes, glyphset); })
    ;
  }

  bool subset (hb_subset_context_t *c) const
  {
    TRACE_SUBSET (this);

    cmap *cmap_prime = c->serializer->start_embed<cmap> ();
    if (unlikely (!c->serializer->check_success (cmap_prime))) return_trace (false);

    auto encodingrec_iter =
    + hb_iter (encodingRecord)
    | hb_filter ([&](const EncodingRecord& _) {
      return cmap::filter_encoding_records_for_subset (this, _);
    })
    ;

    if (unlikely (!encodingrec_iter.len ())) return_trace (false);

    const EncodingRecord *unicode_bmp= nullptr, *unicode_ucs4 = nullptr, *ms_bmp = nullptr, *ms_ucs4 = nullptr;
    bool has_format12 = false;

    for (const EncodingRecord& _ : encodingrec_iter)
    {
      unsigned format = (this + _.subtable).u.format;
      if (format == 12) has_format12 = true;

      const EncodingRecord *table = std::addressof (_);
      if      (_.platformID == 0 && _.encodingID ==  3) unicode_bmp = table;
      else if (_.platformID == 0 && _.encodingID ==  4) unicode_ucs4 = table;
      else if (_.platformID == 3 && _.encodingID ==  1) ms_bmp = table;
      else if (_.platformID == 3 && _.encodingID == 10) ms_ucs4 = table;
    }

    if (unlikely (!has_format12 && !unicode_bmp && !ms_bmp)) return_trace (false);
    if (unlikely (has_format12 && (!unicode_ucs4 && !ms_ucs4))) return_trace (false);

    auto it =
    + c->plan->unicode_to_new_gid_list.iter ()
    | hb_filter ([&] (const hb_pair_t<hb_codepoint_t, hb_codepoint_t> _)
		 { return (_.second != HB_MAP_VALUE_INVALID); })
    ;

    return_trace (cmap_prime->serialize (c->serializer,
                                         it,
                                         encodingrec_iter,
                                         this,
                                         c->plan));
  }

  const CmapSubtable *find_best_subtable (bool *symbol = nullptr) const
  {
    if (symbol) *symbol = false;

    const CmapSubtable *subtable;

    /* Symbol subtable.
     * Prefer symbol if available.
     * https://github.com/harfbuzz/harfbuzz/issues/1918 */
    if ((subtable = this->find_subtable (3, 0)))
    {
      if (symbol) *symbol = true;
      return subtable;
    }

    /* 32-bit subtables. */
    if ((subtable = this->find_subtable (3, 10))) return subtable;
    if ((subtable = this->find_subtable (0, 6))) return subtable;
    if ((subtable = this->find_subtable (0, 4))) return subtable;

    /* 16-bit subtables. */
    if ((subtable = this->find_subtable (3, 1))) return subtable;
    if ((subtable = this->find_subtable (0, 3))) return subtable;
    if ((subtable = this->find_subtable (0, 2))) return subtable;
    if ((subtable = this->find_subtable (0, 1))) return subtable;
    if ((subtable = this->find_subtable (0, 0))) return subtable;

    /* Meh. */
    return &Null (CmapSubtable);
  }

  struct accelerator_t
  {
    accelerator_t (hb_face_t *face)
    {
      this->table = hb_sanitize_context_t ().reference_table<cmap> (face);
      bool symbol;
      this->subtable = table->find_best_subtable (&symbol);
      this->subtable_uvs = &Null (CmapSubtableFormat14);
      {
	const CmapSubtable *st = table->find_subtable (0, 5);
	if (st && st->u.format == 14)
	  subtable_uvs = &st->u.format14;
      }

      this->get_glyph_data = subtable;
      if (unlikely (symbol))
      {
	switch ((unsigned) face->table.OS2->get_font_page ()) {
	case OS2::font_page_t::FONT_PAGE_NONE:
	  this->get_glyph_funcZ = get_glyph_from_symbol<CmapSubtable, _hb_symbol_pua_map>;
	  break;
#ifndef HB_NO_OT_SHAPER_ARABIC_FALLBACK
	case OS2::font_page_t::FONT_PAGE_SIMP_ARABIC:
	  this->get_glyph_funcZ = get_glyph_from_symbol<CmapSubtable, _hb_arabic_pua_simp_map>;
	  break;
	case OS2::font_page_t::FONT_PAGE_TRAD_ARABIC:
	  this->get_glyph_funcZ = get_glyph_from_symbol<CmapSubtable, _hb_arabic_pua_trad_map>;
	  break;
#endif
	default:
	  this->get_glyph_funcZ = get_glyph_from<CmapSubtable>;
	  break;
	}
      }
      else
      {
	switch (subtable->u.format) {
	/* Accelerate format 4 and format 12. */
	default:
	  this->get_glyph_funcZ = get_glyph_from<CmapSubtable>;
	  break;
	case 12:
	  this->get_glyph_funcZ = get_glyph_from<CmapSubtableFormat12>;
	  break;
	case  4:
	{
	  this->format4_accel.init (&subtable->u.format4);
	  this->get_glyph_data = &this->format4_accel;
	  this->get_glyph_funcZ = this->format4_accel.get_glyph_func;
	  break;
	}
	}
      }
    }
    ~accelerator_t () { this->table.destroy (); }

    bool get_nominal_glyph (hb_codepoint_t  unicode,
			    hb_codepoint_t *glyph) const
    {
      if (unlikely (!this->get_glyph_funcZ)) return false;
      return this->get_glyph_funcZ (this->get_glyph_data, unicode, glyph);
    }
    unsigned int get_nominal_glyphs (unsigned int count,
				     const hb_codepoint_t *first_unicode,
				     unsigned int unicode_stride,
				     hb_codepoint_t *first_glyph,
				     unsigned int glyph_stride) const
    {
      if (unlikely (!this->get_glyph_funcZ)) return 0;

      hb_cmap_get_glyph_func_t get_glyph_funcZ = this->get_glyph_funcZ;
      const void *get_glyph_data = this->get_glyph_data;

      unsigned int done;
      for (done = 0;
	   done < count && get_glyph_funcZ (get_glyph_data, *first_unicode, first_glyph);
	   done++)
      {
	first_unicode = &StructAtOffsetUnaligned<hb_codepoint_t> (first_unicode, unicode_stride);
	first_glyph = &StructAtOffsetUnaligned<hb_codepoint_t> (first_glyph, glyph_stride);
      }
      return done;
    }

    bool get_variation_glyph (hb_codepoint_t  unicode,
			      hb_codepoint_t  variation_selector,
			      hb_codepoint_t *glyph) const
    {
      switch (this->subtable_uvs->get_glyph_variant (unicode,
						     variation_selector,
						     glyph))
      {
	case GLYPH_VARIANT_NOT_FOUND:	return false;
	case GLYPH_VARIANT_FOUND:	return true;
	case GLYPH_VARIANT_USE_DEFAULT:	break;
      }

      return get_nominal_glyph (unicode, glyph);
    }

    void collect_unicodes (hb_set_t *out, unsigned int num_glyphs) const
    { subtable->collect_unicodes (out, num_glyphs); }
    void collect_mapping (hb_set_t *unicodes, hb_map_t *mapping,
			  unsigned num_glyphs = UINT_MAX) const
    { subtable->collect_mapping (unicodes, mapping, num_glyphs); }
    void collect_variation_selectors (hb_set_t *out) const
    { subtable_uvs->collect_variation_selectors (out); }
    void collect_variation_unicodes (hb_codepoint_t variation_selector,
				     hb_set_t *out) const
    { subtable_uvs->collect_variation_unicodes (variation_selector, out); }

    protected:
    typedef bool (*hb_cmap_get_glyph_func_t) (const void *obj,
					      hb_codepoint_t codepoint,
					      hb_codepoint_t *glyph);
    typedef uint_fast16_t (*hb_pua_remap_func_t) (unsigned);

    template <typename Type>
    HB_INTERNAL static bool get_glyph_from (const void *obj,
					    hb_codepoint_t codepoint,
					    hb_codepoint_t *glyph)
    {
      const Type *typed_obj = (const Type *) obj;
      return typed_obj->get_glyph (codepoint, glyph);
    }

    template <typename Type, hb_pua_remap_func_t remap>
    HB_INTERNAL static bool get_glyph_from_symbol (const void *obj,
						   hb_codepoint_t codepoint,
						   hb_codepoint_t *glyph)
    {
      const Type *typed_obj = (const Type *) obj;
      if (likely (typed_obj->get_glyph (codepoint, glyph)))
	return true;

      if (hb_codepoint_t c = remap (codepoint))
	return typed_obj->get_glyph (c, glyph);

      return false;
    }

    private:
    hb_nonnull_ptr_t<const CmapSubtable> subtable;
    hb_nonnull_ptr_t<const CmapSubtableFormat14> subtable_uvs;

    hb_cmap_get_glyph_func_t get_glyph_funcZ;
    const void *get_glyph_data;

    CmapSubtableFormat4::accelerator_t format4_accel;

    public:
    hb_blob_ptr_t<cmap> table;
  };

  protected:

  const CmapSubtable *find_subtable (unsigned int platform_id,
				     unsigned int encoding_id) const
  {
    EncodingRecord key;
    key.platformID = platform_id;
    key.encodingID = encoding_id;

    const EncodingRecord &result = encodingRecord.bsearch (key);
    if (!result.subtable)
      return nullptr;

    return &(this+result.subtable);
  }

  const EncodingRecord *find_encodingrec (unsigned int platform_id,
					  unsigned int encoding_id) const
  {
    EncodingRecord key;
    key.platformID = platform_id;
    key.encodingID = encoding_id;

    return encodingRecord.as_array ().bsearch (key);
  }

  bool find_subtable (unsigned format) const
  {
    auto it =
    + hb_iter (encodingRecord)
    | hb_map (&EncodingRecord::subtable)
    | hb_map (hb_add (this))
    | hb_filter ([&] (const CmapSubtable& _) { return _.u.format == format; })
    ;

    return it.len ();
  }

  public:

  bool sanitize (hb_sanitize_context_t *c) const
  {
    TRACE_SANITIZE (this);
    return_trace (c->check_struct (this) &&
		  likely (version == 0) &&
		  encodingRecord.sanitize (c, this));
  }

 private:

  static bool filter_encoding_records_for_subset(const cmap* cmap,
                                                 const EncodingRecord& _)
  {
    return
        (_.platformID == 0 && _.encodingID == 3) ||
        (_.platformID == 0 && _.encodingID == 4) ||
        (_.platformID == 3 && _.encodingID == 1) ||
        (_.platformID == 3 && _.encodingID == 10) ||
        (cmap + _.subtable).u.format == 14;
  }

  protected:
  HBUINT16	version;	/* Table version number (0). */
  SortedArray16Of<EncodingRecord>
		encodingRecord;	/* Encoding tables. */
  public:
  DEFINE_SIZE_ARRAY (4, encodingRecord);
};

struct cmap_accelerator_t : cmap::accelerator_t {
  cmap_accelerator_t (hb_face_t *face) : cmap::accelerator_t (face) {}
};

} /* namespace OT */


#endif /* HB_OT_CMAP_TABLE_HH */