[OTS] Integrate WOFF 2.0 algorithm into OTS

src/woff2.cc

+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// This is the implementation of decompression of the proposed WOFF Ultra
+// Condensed file format.
+
+#include <cassert>
+#include <cstdlib>
+#include <vector>
+
+#include <zlib.h>
+
+#include "third_party/lzma_sdk/LzmaLib.h"
+
+#include "opentype-sanitiser.h"
+#include "ots-memory-stream.h"
+#include "ots.h"
+#include "woff2.h"
+
+namespace {
+
+// simple glyph flags
+const int kGlyfOnCurve = 1 << 0;
+const int kGlyfXShort = 1 << 1;
+const int kGlyfYShort = 1 << 2;
+const int kGlyfRepeat = 1 << 3;
+const int kGlyfThisXIsSame = 1 << 4;
+const int kGlyfThisYIsSame = 1 << 5;
+
+// composite glyph flags
+const int FLAG_ARG_1_AND_2_ARE_WORDS = 1 << 0;
+const int FLAG_ARGS_ARE_XY_VALUES = 1 << 1;
+const int FLAG_ROUND_XY_TO_GRID = 1 << 2;
+const int FLAG_WE_HAVE_A_SCALE = 1 << 3;
+const int FLAG_RESERVED = 1 << 4;
+const int FLAG_MORE_COMPONENTS = 1 << 5;
+const int FLAG_WE_HAVE_AN_X_AND_Y_SCALE = 1 << 6;
+const int FLAG_WE_HAVE_A_TWO_BY_TWO = 1 << 7;
+const int FLAG_WE_HAVE_INSTRUCTIONS = 1 << 8;
+const int FLAG_USE_MY_METRICS = 1 << 9;
+const int FLAG_OVERLAP_COMPOUND = 1 << 10;
+const int FLAG_SCALED_COMPONENT_OFFSET = 1 << 11;
+const int FLAG_UNSCALED_COMPONENT_OFFSET = 1 << 12;
+
+const size_t kSfntHeaderSize = 12;
+const size_t kSfntEntrySize = 16;
+const size_t kCheckSumAdjustmentOffset = 8;
+
+const size_t kEndPtsOfContoursOffset = 10;
+const size_t kCompositeGlyphBegin = 10;
+
+// Note that the byte order is big-endian, not the same as ots.cc
+#define TAG(a, b, c, d) ((a << 24) | (b << 16) | (c << 8) | d)
+
+const unsigned int kWoff2FlagsContinueStream = 1 << 4;
+const unsigned int kWoff2FlagsTransform = 1 << 5;
+
+const size_t kWoff2HeaderSize = 44;
+const size_t kWoff2EntrySize = 20;
+
+const size_t kLzmaHeaderSize = 13;
+
+// Compression type values common to both short and long formats
+const uint32_t kCompressionTypeMask = 0xf;
+const uint32_t kCompressionTypeNone = 0;
+const uint32_t kCompressionTypeGzip = 1;
+const uint32_t kCompressionTypeLzma = 2;
+
+// This is a special value for the short format only, as described in
+// "Design for compressed header format" in draft doc.
+const uint32_t kShortFlagsContinue = 3;
+
+const uint32_t kKnownTags[] = {
+  TAG('c', 'm', 'a', 'p'),  // 0
+  TAG('h', 'e', 'a', 'd'),  // 1
+  TAG('h', 'h', 'e', 'a'),  // 2
+  TAG('h', 'm', 't', 'x'),  // 3
+  TAG('m', 'a', 'x', 'p'),  // 4
+  TAG('n', 'a', 'm', 'e'),  // 5
+  TAG('O', 'S', '/', '2'),  // 6
+  TAG('p', 'o', 's', 't'),  // 7
+  TAG('c', 'v', 't', ' '),  // 8
+  TAG('f', 'p', 'g', 'm'),  // 9
+  TAG('g', 'l', 'y', 'f'),  // 10
+  TAG('l', 'o', 'c', 'a'),  // 11
+  TAG('p', 'r', 'e', 'p'),  // 12
+  TAG('C', 'F', 'F', ' '),  // 13
+  TAG('V', 'O', 'R', 'G'),  // 14
+  TAG('E', 'B', 'D', 'T'),  // 15
+  TAG('E', 'B', 'L', 'C'),  // 16
+  TAG('g', 'a', 's', 'p'),  // 17
+  TAG('h', 'd', 'm', 'x'),  // 18
+  TAG('k', 'e', 'r', 'n'),  // 19
+  TAG('L', 'T', 'S', 'H'),  // 20
+  TAG('P', 'C', 'L', 'T'),  // 21
+  TAG('V', 'D', 'M', 'X'),  // 22
+  TAG('v', 'h', 'e', 'a'),  // 23
+  TAG('v', 'm', 't', 'x'),  // 24
+  TAG('B', 'A', 'S', 'E'),  // 25
+  TAG('G', 'D', 'E', 'F'),  // 26
+  TAG('G', 'P', 'O', 'S'),  // 27
+  TAG('G', 'S', 'U', 'B'),  // 28
+};
+
+struct Point {
+  int x;
+  int y;
+  bool on_curve;
+};
+
+struct Table {
+  uint32_t tag;
+  uint32_t flags;
+  uint32_t src_offset;
+  uint32_t src_length;
+
+  uint32_t transform_length;
+
+  uint32_t dst_offset;
+  uint32_t dst_length;
+
+  Table()
+      : tag(0),
+        flags(0),
+        src_offset(0),
+        src_length(0),
+        transform_length(0),
+        dst_offset(0),
+        dst_length(0) {}
+};
+
+// Based on section 6.1.1 of MicroType Express draft spec
+bool Read255UShort(ots::Buffer* buf, unsigned int* value) {
+  static const int kWordCode = 253;
+  static const int kOneMoreByteCode2 = 254;
+  static const int kOneMoreByteCode1 = 255;
+  static const int kLowestUCode = 253;
+  uint8_t code = 0;
+  if (!buf->ReadU8(&code)) {
+    return OTS_FAILURE();
+  }
+  if (code == kWordCode) {
+    uint16_t result = 0;
+    if (!buf->ReadU16(&result)) {
+      return OTS_FAILURE();
+    }
+    *value = result;
+    return true;
+  } else if (code == kOneMoreByteCode1) {
+    uint8_t result = 0;
+    if (!buf->ReadU8(&result)) {
+      return OTS_FAILURE();
+    }
+    *value = result + kLowestUCode;
+    return true;
+  } else if (code == kOneMoreByteCode2) {
+    uint8_t result = 0;
+    if (!buf->ReadU8(&result)) {
+      return OTS_FAILURE();
+    }
+    *value = result + kLowestUCode * 2;
+    return true;
+  } else {
+    *value = code;
+    return true;
+  }
+}
+
+bool ReadBase128(ots::Buffer* buf, uint32_t* value) {
+  uint32_t result = 0;
+  for (size_t i = 0; i < 5; ++i) {
+    uint8_t code = 0;
+    if (!buf->ReadU8(&code)) {
+      return OTS_FAILURE();
+    }
+    // If any of the top seven bits are set then we're about to overflow.
+    if (result & 0xe0000000U) {
+      return OTS_FAILURE();
+    }
+    result = (result << 7) | (code & 0x7f);
+    if ((code & 0x80) == 0) {
+      *value = result;
+      return true;
+    }
+  }
+  // Make sure not to exceed the size bound
+  return OTS_FAILURE();
+}
+
+// Caller must ensure that buffer overrun won't happen.
+// TODO(ksakamaoto): Consider creating 'writer' version of the Buffer class
+// and use it across the code.
+size_t StoreU32(uint8_t* dst, size_t offset, uint32_t x) {
+  dst[offset] = x >> 24;
+  dst[offset + 1] = x >> 16;
+  dst[offset + 2] = x >> 8;
+  dst[offset + 3] = x;
+  return offset + 4;
+}
+
+size_t Store16(uint8_t* dst, size_t offset, int x) {
+  dst[offset] = x >> 8;
+  dst[offset + 1] = x;
+  return offset + 2;
+}
+
+int WithSign(int flag, int baseval) {
+  assert(0 <= baseval && baseval < 65536);
+  return (flag & 1) ? baseval : -baseval;
+}
+
+bool TripletDecode(const uint8_t* flags_in, const uint8_t* in, size_t in_size,
+    unsigned int n_points, std::vector<Point>* result,
+    size_t* in_bytes_consumed) {
+  int x = 0;
+  int y = 0;
+
+  // Early return if |in| buffer is too small. Each point consumes 1-4 bytes.
+  if (n_points > in_size) {
+    return OTS_FAILURE();
+  }
+  unsigned int triplet_index = 0;
+
+  for (unsigned int i = 0; i < n_points; ++i) {
+    uint8_t flag = flags_in[i];
+    bool on_curve = !(flag >> 7);
+    flag &= 0x7f;
+    unsigned int n_data_bytes;
+    if (flag < 84) {
+      n_data_bytes = 1;
+    } else if (flag < 120) {
+      n_data_bytes = 2;
+    } else if (flag < 124) {
+      n_data_bytes = 3;
+    } else {
+      n_data_bytes = 4;
+    }
+    if (triplet_index + n_data_bytes > in_size ||
+        triplet_index + n_data_bytes < triplet_index) {
+      return OTS_FAILURE();
+    }
+    int dx, dy;
+    if (flag < 10) {
+      dx = 0;
+      dy = WithSign(flag, ((flag & 14) << 7) + in[triplet_index]);
+    } else if (flag < 20) {
+      dx = WithSign(flag, (((flag - 10) & 14) << 7) + in[triplet_index]);
+      dy = 0;
+    } else if (flag < 84) {
+      int b0 = flag - 20;
+      int b1 = in[triplet_index];
+      dx = WithSign(flag, 1 + (b0 & 0x30) + (b1 >> 4));
+      dy = WithSign(flag >> 1, 1 + ((b0 & 0x0c) << 2) + (b1 & 0x0f));
+    } else if (flag < 120) {
+      int b0 = flag - 84;
+      dx = WithSign(flag, 1 + ((b0 / 12) << 8) + in[triplet_index]);
+      dy = WithSign(flag >> 1,
+                    1 + (((b0 % 12) >> 2) << 8) + in[triplet_index + 1]);
+    } else if (flag < 124) {
+      int b2 = in[triplet_index + 1];
+      dx = WithSign(flag, (in[triplet_index] << 4) + (b2 >> 4));
+      dy = WithSign(flag >> 1, ((b2 & 0x0f) << 8) + in[triplet_index + 2]);
+    } else {
+      dx = WithSign(flag, (in[triplet_index] << 8) + in[triplet_index + 1]);
+      dy = WithSign(flag >> 1,
+          (in[triplet_index + 2] << 8) + in[triplet_index + 3]);
+    }
+    triplet_index += n_data_bytes;
+    // Possible overflow but coordinate values are not security sensitive
+    x += dx;
+    y += dy;
+    result->push_back(Point());
+    Point& back = result->back();
+    back.x = x;
+    back.y = y;
+    back.on_curve = on_curve;
+  }
+  *in_bytes_consumed = triplet_index;
+  return true;
+}
+
+// This function stores just the point data. On entry, dst points to the
+// beginning of a simple glyph. Returns true on success.
+bool StorePoints(const std::vector<Point>& points,
+    unsigned int n_contours, unsigned int instruction_length,
+    uint8_t* dst, size_t dst_size, size_t* glyph_size) {
+  // I believe that n_contours < 65536, in which case this is safe. However, a
+  // comment and/or an assert would be good.
+  unsigned int flag_offset = kEndPtsOfContoursOffset + 2 * n_contours + 2 +
+    instruction_length;
+  int last_flag = -1;
+  int repeat_count = 0;
+  int last_x = 0;
+  int last_y = 0;
+  unsigned int x_bytes = 0;
+  unsigned int y_bytes = 0;
+
+  for (size_t i = 0; i < points.size(); ++i) {
+    const Point& point = points.at(i);
+    int flag = point.on_curve ? kGlyfOnCurve : 0;
+    int dx = point.x - last_x;
+    int dy = point.y - last_y;
+    if (dx == 0) {
+      flag |= kGlyfThisXIsSame;
+    } else if (dx > -256 && dx < 256) {
+      flag |= kGlyfXShort | (dx > 0 ? kGlyfThisXIsSame : 0);
+      x_bytes += 1;
+    } else {
+      x_bytes += 2;
+    }
+    if (dy == 0) {
+      flag |= kGlyfThisYIsSame;
+    } else if (dy > -256 && dy < 256) {
+      flag |= kGlyfYShort | (dy > 0 ? kGlyfThisYIsSame : 0);
+      y_bytes += 1;
+    } else {
+      y_bytes += 2;
+    }
+
+    if (flag == last_flag && repeat_count != 255) {
+      dst[flag_offset - 1] |= kGlyfRepeat;
+      repeat_count++;
+    } else {
+      if (repeat_count != 0) {
+        if (flag_offset >= dst_size) {
+          return OTS_FAILURE();
+        }
+        dst[flag_offset++] = repeat_count;
+      }
+      if (flag_offset >= dst_size) {
+        return OTS_FAILURE();
+      }
+      dst[flag_offset++] = flag;
+      repeat_count = 0;
+    }
+    last_x = point.x;
+    last_y = point.y;
+    last_flag = flag;
+  }
+
+  if (repeat_count != 0) {
+    if (flag_offset >= dst_size) {
+      return OTS_FAILURE();
+    }
+    dst[flag_offset++] = repeat_count;
+  }
+  unsigned int xy_bytes = x_bytes + y_bytes;
+  if (xy_bytes < x_bytes ||
+      flag_offset + xy_bytes < flag_offset ||
+      flag_offset + xy_bytes > dst_size) {
+    return OTS_FAILURE();
+  }
+
+  int x_offset = flag_offset;
+  int y_offset = flag_offset + x_bytes;
+  last_x = 0;
+  last_y = 0;
+  for (size_t i = 0; i < points.size(); ++i) {
+    int dx = points.at(i).x - last_x;
+    if (dx == 0) {
+      // pass
+    } else if (dx > -256 && dx < 256) {
+      dst[x_offset++] = std::abs(dx);
+    } else {
+      // will always fit for valid input, but overflow is harmless
+      x_offset = Store16(dst, x_offset, dx);
+    }
+    last_x += dx;
+    int dy = points.at(i).y - last_y;
+    if (dy == 0) {
+      // pass
+    } else if (dy > -256 && dy < 256) {
+      dst[y_offset++] = std::abs(dy);
+    } else {
+      y_offset = Store16(dst, y_offset, dy);
+    }
+    last_y += dy;
+  }
+  *glyph_size = y_offset;
+  return true;
+}
+
+// Compute the bounding box of the coordinates, and store into a glyf buffer.
+// A precondition is that there are at least 10 bytes available.
+void ComputeBbox(const std::vector<Point>& points, uint8_t* dst) {
+  int x_min = 0;
+  int y_min = 0;
+  int x_max = 0;
+  int y_max = 0;
+
+  for (size_t i = 0; i < points.size(); ++i) {
+    int x = points.at(i).x;
+    int y = points.at(i).y;
+    if (i == 0 || x < x_min) x_min = x;
+    if (i == 0 || x > x_max) x_max = x;
+    if (i == 0 || y < y_min) y_min = y;
+    if (i == 0 || y > y_max) y_max = y;
+  }
+  size_t offset = 2;
+  offset = Store16(dst, offset, x_min);
+  offset = Store16(dst, offset, y_min);
+  offset = Store16(dst, offset, x_max);
+  offset = Store16(dst, offset, y_max);
+}
+
+// Process entire bbox stream. This is done as a separate pass to allow for
+// composite bbox computations (an optional more aggressive transform).
+bool ProcessBboxStream(ots::Buffer* bbox_stream, unsigned int n_glyphs,
+    const std::vector<uint32_t>& loca_values, uint8_t* glyf_buf,
+    size_t glyf_buf_length) {
+  const uint8_t* buf = bbox_stream->buffer();
+  if (n_glyphs >= 65536 || loca_values.size() != n_glyphs + 1) {
+    return OTS_FAILURE();
+  }
+  // Safe because n_glyphs is bounded
+  unsigned int bitmap_length = ((n_glyphs + 31) >> 5) << 2;
+  if (!bbox_stream->Skip(bitmap_length)) {
+    return OTS_FAILURE();
+  }
+  for (unsigned int i = 0; i < n_glyphs; ++i) {
+    if (buf[i >> 3] & (0x80 >> (i & 7))) {
+      uint32_t loca_offset = loca_values.at(i);
+      if (loca_values.at(i + 1) - loca_offset < kEndPtsOfContoursOffset) {
+        return OTS_FAILURE();
+      }
+      if (glyf_buf_length < 2 + 10 ||
+          loca_offset > glyf_buf_length - 2 - 10) {
+        return OTS_FAILURE();
+      }
+      if (!bbox_stream->Read(glyf_buf + loca_offset + 2, 8)) {
+        return OTS_FAILURE();
+      }
+    }
+  }
+  return true;
+}
+
+bool ProcessComposite(ots::Buffer* composite_stream, uint8_t* dst,
+    size_t dst_size, size_t* glyph_size, bool* have_instructions) {
+  size_t start_offset = composite_stream->offset();
+  bool we_have_instructions = false;
+
+  uint16_t flags = FLAG_MORE_COMPONENTS;
+  while (flags & FLAG_MORE_COMPONENTS) {
+    if (!composite_stream->ReadU16(&flags)) {
+      return OTS_FAILURE();
+    }
+    we_have_instructions |= (flags & FLAG_WE_HAVE_INSTRUCTIONS) != 0;
+    size_t arg_size = 2;  // glyph index
+    if (flags & FLAG_ARG_1_AND_2_ARE_WORDS) {
+      arg_size += 4;
+    } else {
+      arg_size += 2;
+    }
+    if (flags & FLAG_WE_HAVE_A_SCALE) {
+      arg_size += 2;
+    } else if (flags & FLAG_WE_HAVE_AN_X_AND_Y_SCALE) {
+      arg_size += 4;
+    } else if (flags & FLAG_WE_HAVE_A_TWO_BY_TWO) {
+      arg_size += 8;
+    }
+    if (!composite_stream->Skip(arg_size)) {
+      return OTS_FAILURE();
+    }
+  }
+  size_t composite_glyph_size = composite_stream->offset() - start_offset;
+  if (composite_glyph_size + kCompositeGlyphBegin > dst_size) {
+    return OTS_FAILURE();
+  }
+  Store16(dst, 0, 0xffff);  // nContours = -1 for composite glyph
+  std::memcpy(dst + kCompositeGlyphBegin,
+      composite_stream->buffer() + start_offset,
+      composite_glyph_size);
+  *glyph_size = kCompositeGlyphBegin + composite_glyph_size;
+  *have_instructions = we_have_instructions;
+  return true;
+}
+
+// Build TrueType loca table
+bool StoreLoca(const std::vector<uint32_t>& loca_values, int index_format,
+    uint8_t* dst, size_t dst_size) {
+  const uint64_t loca_size = loca_values.size();
+  const uint64_t offset_size = index_format ? 4 : 2;
+  if ((loca_size << 2) >> 2 != loca_size) {
+    return OTS_FAILURE();
+  }
+  // No integer overflow here (loca_size <= 2^16).
+  if (offset_size * loca_size > dst_size) {
+    return OTS_FAILURE();
+  }
+  size_t offset = 0;
+  for (size_t i = 0; i < loca_values.size(); ++i) {
+    uint32_t value = loca_values.at(i);
+    if (index_format) {
+      offset = StoreU32(dst, offset, value);
+    } else {
+      offset = Store16(dst, offset, value >> 1);
+    }
+  }
+  return true;
+}
+
+// Reconstruct entire glyf table based on transformed original
+bool ReconstructGlyf(const uint8_t* data, size_t data_size,
+    uint8_t* dst, size_t dst_size,
+    uint8_t* loca_buf, size_t loca_size) {
+  static const int kNumSubStreams = 7;
+  ots::Buffer file(data, data_size);
+  uint32_t version;
+  std::vector<std::pair<const uint8_t*, size_t> > substreams(kNumSubStreams);
+
+  if (!file.ReadU32(&version)) {
+    return OTS_FAILURE();
+  }
+  uint16_t num_glyphs;
+  uint16_t index_format;
+  if (!file.ReadU16(&num_glyphs) ||
+      !file.ReadU16(&index_format)) {
+    return OTS_FAILURE();
+  }
+  unsigned int offset = (2 + kNumSubStreams) * 4;
+  if (offset > data_size) {
+    return OTS_FAILURE();
+  }
+  // Invariant from here on: data_size >= offset
+  for (int i = 0; i < kNumSubStreams; ++i) {
+    uint32_t substream_size;
+    if (!file.ReadU32(&substream_size)) {
+      return OTS_FAILURE();
+    }
+    if (substream_size > data_size - offset) {
+      return OTS_FAILURE();
+    }
+    substreams.at(i) = std::make_pair(data + offset, substream_size);
+    offset += substream_size;
+  }
+  ots::Buffer n_contour_stream(substreams.at(0).first, substreams.at(0).second);
+  ots::Buffer n_points_stream(substreams.at(1).first, substreams.at(1).second);
+  ots::Buffer flag_stream(substreams.at(2).first, substreams.at(2).second);
+  ots::Buffer glyph_stream(substreams.at(3).first, substreams.at(3).second);
+  ots::Buffer composite_stream(substreams.at(4).first, substreams.at(4).second);
+  ots::Buffer bbox_stream(substreams.at(5).first, substreams.at(5).second);
+  ots::Buffer instruction_stream(substreams.at(6).first,
+                                 substreams.at(6).second);
+
+  std::vector<uint32_t> loca_values;
+  loca_values.reserve(num_glyphs + 1);
+  std::vector<unsigned int> n_points_vec;
+  std::vector<Point> points;
+  uint32_t loca_offset = 0;
+  for (unsigned int i = 0; i < num_glyphs; ++i) {
+    size_t glyph_size = 0;
+    uint16_t n_contours = 0;
+    if (!n_contour_stream.ReadU16(&n_contours)) {
+      return OTS_FAILURE();
+    }
+    uint8_t* glyf_dst = dst + loca_offset;
+    size_t glyf_dst_size = dst_size - loca_offset;
+    if (n_contours == 0xffff) {
+      // composite glyph
+      bool have_instructions = false;
+      unsigned int instruction_size = 0;
+      if (!ProcessComposite(&composite_stream, glyf_dst, glyf_dst_size,
+            &glyph_size, &have_instructions)) {
+        return OTS_FAILURE();
+      }
+      if (have_instructions) {
+        if (!Read255UShort(&glyph_stream, &instruction_size)) {
+          return OTS_FAILURE();
+        }
+        // No integer overflow here (instruction_size < 2^16).
+        if (instruction_size + 2 > glyf_dst_size - glyph_size) {
+          return OTS_FAILURE();
+        }
+        Store16(glyf_dst, glyph_size, instruction_size);
+        if (!instruction_stream.Read(glyf_dst + glyph_size + 2,
+              instruction_size)) {
+          return OTS_FAILURE();
+        }
+        glyph_size += instruction_size + 2;
+      }
+    } else if (n_contours > 0) {
+      // simple glyph
+      n_points_vec.clear();
+      points.clear();
+      unsigned int total_n_points = 0;
+      unsigned int n_points_contour;
+      for (unsigned int j = 0; j < n_contours; ++j) {
+        if (!Read255UShort(&n_points_stream, &n_points_contour)) {
+          return OTS_FAILURE();
+        }
+        n_points_vec.push_back(n_points_contour);
+        if (total_n_points + n_points_contour < total_n_points) {
+          return OTS_FAILURE();
+        }
+        total_n_points += n_points_contour;
+      }
+      unsigned int flag_size = total_n_points;
+      if (flag_size > flag_stream.length() - flag_stream.offset()) {
+        return OTS_FAILURE();
+      }
+      const uint8_t* flags_buf = flag_stream.buffer() + flag_stream.offset();
+      const uint8_t* triplet_buf = glyph_stream.buffer() +
+        glyph_stream.offset();
+      size_t triplet_size = glyph_stream.length() - glyph_stream.offset();
+      size_t triplet_bytes_consumed = 0;
+      if (!TripletDecode(flags_buf, triplet_buf, triplet_size, total_n_points,
+            &points, &triplet_bytes_consumed)) {
+        return OTS_FAILURE();
+      }
+      const uint32_t header_and_endpts_contours_size =
+          kEndPtsOfContoursOffset + 2 * n_contours;
+      if (glyf_dst_size < header_and_endpts_contours_size) {
+        return OTS_FAILURE();
+      }
+      Store16(glyf_dst, 0, n_contours);
+      ComputeBbox(points, glyf_dst);
+      size_t offset = kEndPtsOfContoursOffset;
+      int end_point = -1;
+      for (unsigned int contour_ix = 0; contour_ix < n_contours; ++contour_ix) {
+        end_point += n_points_vec.at(contour_ix);
+        if (end_point >= 65536) {
+          return OTS_FAILURE();
+        }
+        offset = Store16(glyf_dst, offset, end_point);
+      }
+      if (!flag_stream.Skip(flag_size)) {
+        return OTS_FAILURE();
+      }
+      if (!glyph_stream.Skip(triplet_bytes_consumed)) {
+        return OTS_FAILURE();
+      }
+      unsigned int instruction_size;
+      if (!Read255UShort(&glyph_stream, &instruction_size)) {
+        return OTS_FAILURE();
+      }
+      // No integer overflow here (instruction_size < 2^16).
+      if (glyf_dst_size - header_and_endpts_contours_size <
+          instruction_size + 2) {
+        return OTS_FAILURE();
+      }
+      uint8_t* instruction_dst = glyf_dst + header_and_endpts_contours_size;
+      Store16(instruction_dst, 0, instruction_size);
+      if (!instruction_stream.Read(instruction_dst + 2, instruction_size)) {
+        return OTS_FAILURE();
+      }
+      if (!StorePoints(points, n_contours, instruction_size,
+            glyf_dst, glyf_dst_size, &glyph_size)) {
+        return OTS_FAILURE();
+      }
+    } else {
+      glyph_size = 0;
+    }
+    loca_values.push_back(loca_offset);
+    if (glyph_size + 3 < glyph_size) {
+      return OTS_FAILURE();
+    }
+    glyph_size = ots::Round4(glyph_size);
+    if (glyph_size > dst_size - loca_offset) {
+      // This shouldn't happen, but this test defensively maintains the
+      // invariant that loca_offset <= dst_size.
+      return OTS_FAILURE();
+    }
+    loca_offset += glyph_size;
+  }
+  loca_values.push_back(loca_offset);
+  assert(loca_values.size() == static_cast<size_t>(num_glyphs + 1));
+  if (!ProcessBboxStream(&bbox_stream, num_glyphs, loca_values,
+          dst, dst_size)) {
+    return OTS_FAILURE();
+  }
+  return StoreLoca(loca_values, index_format, loca_buf, loca_size);
+}
+
+// This is linear search, but could be changed to binary because we
+// do have a guarantee that the tables are sorted by tag. But the total
+// cpu time is expected to be very small in any case.
+const Table* FindTable(const std::vector<Table>& tables, uint32_t tag) {
+  size_t n_tables = tables.size();
+  for (size_t i = 0; i < n_tables; ++i) {
+    if (tables.at(i).tag == tag) {
+      return &tables.at(i);
+    }
+  }
+  return NULL;
+}
+
+bool ReconstructTransformed(const std::vector<Table>& tables, uint32_t tag,
+    const uint8_t* transformed_buf, size_t transformed_size,
+    uint8_t* dst, size_t dst_length) {
+  if (tag == TAG('g', 'l', 'y', 'f')) {
+    const Table* glyf_table = FindTable(tables, tag);
+    const Table* loca_table = FindTable(tables, TAG('l', 'o', 'c', 'a'));
+    if (glyf_table == NULL || loca_table == NULL) {
+      return OTS_FAILURE();
+    }
+    if (static_cast<uint64_t>(glyf_table->dst_offset) + glyf_table->dst_length >
+        dst_length) {
+      return OTS_FAILURE();
+    }
+    if (static_cast<uint64_t>(loca_table->dst_offset) + loca_table->dst_length >
+        dst_length) {
+      return OTS_FAILURE();
+    }
+    return ReconstructGlyf(transformed_buf, transformed_size,
+        dst + glyf_table->dst_offset, glyf_table->dst_length,
+        dst + loca_table->dst_offset, loca_table->dst_length);
+  } else if (tag == TAG('l', 'o', 'c', 'a')) {
+    // processing was already done by glyf table, but validate
+    if (!FindTable(tables, TAG('g', 'l', 'y', 'f'))) {
+      return OTS_FAILURE();
+    }
+  } else {
+    // transform for the tag is not known
+    return OTS_FAILURE();
+  }
+  return true;
+}
+
+uint32_t ComputeChecksum(const uint8_t* buf, size_t size) {
+  uint32_t checksum = 0;
+  for (size_t i = 0; i < size; i += 4) {
+    // We assume the addition is mod 2^32, which is valid because unsigned
+    checksum += (buf[i] << 24) | (buf[i + 1] << 16) |
+      (buf[i + 2] << 8) | buf[i + 3];
+  }
+  return checksum;
+}
+
+bool FixChecksums(const std::vector<Table>& tables, uint8_t* dst) {
+  const Table* head_table = FindTable(tables, TAG('h', 'e', 'a', 'd'));
+  if (head_table == NULL ||
+      head_table->dst_length < kCheckSumAdjustmentOffset + 4) {
+    return OTS_FAILURE();
+  }
+  size_t adjustment_offset = head_table->dst_offset + kCheckSumAdjustmentOffset;
+  if (adjustment_offset < head_table->dst_offset) {
+    return OTS_FAILURE();
+  }
+  StoreU32(dst, adjustment_offset, 0);
+  size_t n_tables = tables.size();
+  uint32_t file_checksum = 0;
+  for (size_t i = 0; i < n_tables; ++i) {
+    const Table* table = &tables.at(i);
+    size_t table_length = table->dst_length;
+    uint8_t* table_data = dst + table->dst_offset;
+    uint32_t checksum = ComputeChecksum(table_data, table_length);
+    StoreU32(dst, kSfntHeaderSize + i * kSfntEntrySize + 4, checksum);
+    file_checksum += checksum;  // The addition is mod 2^32
+  }
+  file_checksum += ComputeChecksum(dst,
+      kSfntHeaderSize + kSfntEntrySize * n_tables);
+  uint32_t checksum_adjustment = 0xb1b0afba - file_checksum;
+  StoreU32(dst, adjustment_offset, checksum_adjustment);
+  return true;
+}
+
+bool Woff2Uncompress(uint8_t* dst_buf, size_t dst_size,
+    const uint8_t* src_buf, size_t src_size, uint32_t compression_type) {
+  if (compression_type == kCompressionTypeGzip) {
+    uLongf uncompressed_length = dst_size;
+    int r = uncompress(reinterpret_cast<Bytef *>(dst_buf), &uncompressed_length,
+        src_buf, src_size);
+    if (r != Z_OK || uncompressed_length != dst_size) {
+      return OTS_FAILURE();
+    }
+    return true;
+  } else if (compression_type == kCompressionTypeLzma) {
+    if (src_size < kLzmaHeaderSize) {
+      // Make sure we have at least a full Lzma header
+      return OTS_FAILURE();
+    }
+    // TODO: check that size matches (or elide size?)
+    size_t uncompressed_size = dst_size;
+    size_t compressed_size = src_size;
+    int result = LzmaUncompress(dst_buf, &dst_size,
+        src_buf + kLzmaHeaderSize, &compressed_size,
+        src_buf, LZMA_PROPS_SIZE);
+    if (result != SZ_OK || uncompressed_size != dst_size) {
+      return OTS_FAILURE();
+    }
+    return true;
+  }
+  // Unknown compression type
+  return OTS_FAILURE();
+}
+
+bool ReadShortDirectory(ots::Buffer* file, std::vector<Table>* tables,
+    size_t num_tables) {
+  uint32_t last_compression_type = 0;
+  for (size_t i = 0; i < num_tables; ++i) {
+    Table* table = &tables->at(i);
+    uint8_t flag_byte;
+    if (!file->ReadU8(&flag_byte)) {
+      return OTS_FAILURE();
+    }
+    uint32_t tag;
+    if ((flag_byte & 0x1f) == 0x1f) {
+      if (!file->ReadU32(&tag)) {
+        return OTS_FAILURE();
+      }
+    } else {
+      if ((flag_byte & 0x1f) >= arraysize(kKnownTags)) {
+        return OTS_FAILURE();
+      }
+      tag = kKnownTags[flag_byte & 0x1f];
+    }
+    uint32_t flags = flag_byte >> 6;
+    if (flags == kShortFlagsContinue) {
+      flags = last_compression_type | kWoff2FlagsContinueStream;
+    } else {
+      if (flags == kCompressionTypeNone ||
+          flags == kCompressionTypeGzip ||
+          flags == kCompressionTypeLzma) {
+        last_compression_type = flags;
+      } else {
+        return OTS_FAILURE();
+      }
+    }
+    if ((flag_byte & 0x20) != 0) {
+      flags |= kWoff2FlagsTransform;
+    }
+    uint32_t dst_length;
+    if (!ReadBase128(file, &dst_length)) {
+      return OTS_FAILURE();
+    }
+    uint32_t transform_length = dst_length;
+    if ((flags & kWoff2FlagsTransform) != 0) {
+      if (!ReadBase128(file, &transform_length)) {
+        return OTS_FAILURE();
+      }
+    }
+    uint32_t src_length = transform_length;
+    if ((flag_byte >> 6) == 1 || (flag_byte >> 6) == 2) {
+      if (!ReadBase128(file, &src_length)) {
+        return OTS_FAILURE();
+      }
+    }
+    // Disallow huge numbers (> 1GB) for sanity.
+    if (src_length > 1024 * 1024 * 1024 ||
+        transform_length > 1024 * 1024 * 1024 ||
+        dst_length > 1024 * 1024 * 1024) {
+      return OTS_FAILURE();
+    }
+
+    table->tag = tag;
+    table->flags = flags;
+    table->src_length = src_length;
+    table->transform_length = transform_length;
+    table->dst_length = dst_length;
+  }
+  return true;
+}
+
+}  // namespace
+
+namespace ots {
+
+size_t ComputeWOFF2FinalSize(const uint8_t* data, size_t length) {
+  ots::Buffer file(data, length);
+  uint32_t total_length;
+
+  if (!file.Skip(16) ||
+      !file.ReadU32(&total_length)) {
+    return 0;
+  }
+  return total_length;
+}
+
+bool ConvertWOFF2ToTTF(uint8_t* result, size_t result_length,
+                       const uint8_t* data, size_t length) {
+  static const uint32_t kWoff2Signature = 0x774f4632;  // "wOF2"
+  ots::Buffer file(data, length);
+
+  uint32_t signature;
+  uint32_t flavor;
+  if (!file.ReadU32(&signature) || signature != kWoff2Signature ||
+      !file.ReadU32(&flavor)) {
+    return OTS_FAILURE();
+  }
+
+  if (!IsValidVersionTag(ntohl(flavor))) {
+    return OTS_FAILURE();
+  }
+
+  uint32_t reported_length;
+  if (!file.ReadU32(&reported_length) || length != reported_length) {
+    return OTS_FAILURE();
+  }
+  uint16_t num_tables;
+  if (!file.ReadU16(&num_tables) || !num_tables) {
+    return OTS_FAILURE();
+  }
+  // We don't care about these fields of the header:
+  //   uint16_t reserved
+  //   uint32_t total_sfnt_size
+  //   uint16_t major_version, minor_version
+  //   uint32_t meta_offset, meta_length, meta_orig_length
+  //   uint32_t priv_offset, priv_length
+  if (!file.Skip(30)) {
+    return OTS_FAILURE();
+  }
+  std::vector<Table> tables(num_tables);
+  if (!ReadShortDirectory(&file, &tables, num_tables)) {
+    return OTS_FAILURE();
+  }
+  uint64_t src_offset = file.offset();
+  uint64_t dst_offset = kSfntHeaderSize +
+      kSfntEntrySize * static_cast<uint64_t>(num_tables);
+  uint64_t uncompressed_sum = 0;
+  for (uint16_t i = 0; i < num_tables; ++i) {
+    Table* table = &tables.at(i);
+    table->src_offset = src_offset;
+    src_offset += table->src_length;
+    if (src_offset > std::numeric_limits<uint32_t>::max()) {
+      return OTS_FAILURE();
+    }
+    src_offset = ots::Round4(src_offset);
+    table->dst_offset = dst_offset;
+    dst_offset += table->dst_length;
+    if (dst_offset > std::numeric_limits<uint32_t>::max()) {
+      return OTS_FAILURE();
+    }
+    dst_offset = ots::Round4(dst_offset);
+    if ((table->flags & kCompressionTypeMask) != kCompressionTypeNone) {
+      uncompressed_sum += table->src_length;
+      if (uncompressed_sum > std::numeric_limits<uint32_t>::max()) {
+        return OTS_FAILURE();
+      }
+    }
+  }
+  // Enforce same 30M limit on uncompressed tables as OTS
+  if (uncompressed_sum > 30 * 1024 * 1024) {
+    return OTS_FAILURE();
+  }
+  if (src_offset > length || dst_offset > result_length) {
+    return OTS_FAILURE();
+  }
+
+  const uint32_t sfnt_header_and_table_directory_size = 12 + 16 * num_tables;
+  if (sfnt_header_and_table_directory_size > result_length) {
+    return OTS_FAILURE();
+  }
+
+  // Start building the font
+  size_t offset = 0;
+  offset = StoreU32(result, offset, flavor);
+  offset = Store16(result, offset, num_tables);
+  unsigned max_pow2 = 0;
+  while (1u << (max_pow2 + 1) <= num_tables) {
+    max_pow2++;
+  }
+  const uint16_t output_search_range = (1u << max_pow2) << 4;
+  offset = Store16(result, offset, output_search_range);
+  offset = Store16(result, offset, max_pow2);
+  offset = Store16(result, offset, (num_tables << 4) - output_search_range);
+  for (uint16_t i = 0; i < num_tables; ++i) {
+    const Table* table = &tables.at(i);
+    offset = StoreU32(result, offset, table->tag);
+    offset = StoreU32(result, offset, 0);  // checksum, to fill in later
+    offset = StoreU32(result, offset, table->dst_offset);
+    offset = StoreU32(result, offset, table->dst_length);
+  }
+  std::vector<uint8_t> uncompressed_buf;
+  bool continue_valid = false;
+  for (uint16_t i = 0; i < num_tables; ++i) {
+    const Table* table = &tables.at(i);
+    uint32_t flags = table->flags;
+    const uint8_t* src_buf = data + table->src_offset;
+    uint32_t compression_type = flags & kCompressionTypeMask;
+    const uint8_t* transform_buf = NULL;
+    size_t transform_length = table->transform_length;
+    if ((flags & kWoff2FlagsContinueStream) != 0) {
+      if (!continue_valid) {
+        return OTS_FAILURE();
+      }
+    } else if (compression_type == kCompressionTypeNone) {
+      if (transform_length != table->src_length) {
+        return OTS_FAILURE();
+      }
+      transform_buf = src_buf;
+      continue_valid = false;
+    } else if ((flags & kWoff2FlagsContinueStream) == 0) {
+      uint64_t total_size = transform_length;
+      for (uint16_t j = i + 1; j < num_tables; ++j) {
+        if ((tables.at(j).flags & kWoff2FlagsContinueStream) == 0) {
+          break;
+        }
+        total_size += tables.at(j).transform_length;
+        if (total_size > std::numeric_limits<uint32_t>::max()) {
+          return OTS_FAILURE();
+        }
+      }
+      // Enforce same 30M limit on uncompressed tables as OTS
+      if (total_size > 30 * 1024 * 1024) {
+        return OTS_FAILURE();
+      }
+      uncompressed_buf.resize(total_size);
+      if (!Woff2Uncompress(&uncompressed_buf[0], total_size,
+          src_buf, table->src_length, compression_type)) {
+        return OTS_FAILURE();
+      }
+      transform_buf = &uncompressed_buf[0];
+      continue_valid = true;
+    } else {
+      return OTS_FAILURE();
+    }
+
+    if ((flags & kWoff2FlagsTransform) == 0) {
+      if (transform_length != table->dst_length) {
+        return OTS_FAILURE();
+      }
+      if (static_cast<uint64_t>(table->dst_offset) + transform_length >
+          result_length) {
+        return OTS_FAILURE();
+      }
+      std::memcpy(result + table->dst_offset, transform_buf,
+          transform_length);
+    } else {
+      if (!ReconstructTransformed(tables, table->tag,
+            transform_buf, transform_length, result, result_length)) {
+        return OTS_FAILURE();
+      }
+    }
+    if (continue_valid) {
+      transform_buf += transform_length;
+      if (transform_buf > &uncompressed_buf[uncompressed_buf.size()]) {
+        return OTS_FAILURE();
+      }
+    }
+  }
+
+  return FixChecksums(tables, result);
+}
+
+}  // namespace ots