diff options
Diffstat (limited to '')
27 files changed, 4690 insertions, 0 deletions
diff --git a/modules/woff2/src/buffer.h b/modules/woff2/src/buffer.h new file mode 100644 index 0000000000..7240e5181c --- /dev/null +++ b/modules/woff2/src/buffer.h @@ -0,0 +1,164 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* The parts of ots.h & opentype-sanitiser.h that we need, taken from the + https://code.google.com/p/ots/ project. */ + +#ifndef WOFF2_BUFFER_H_ +#define WOFF2_BUFFER_H_ + +#if defined(_WIN32) +#include <stdlib.h> +typedef signed char int8_t; +typedef unsigned char uint8_t; +typedef short int16_t; +typedef unsigned short uint16_t; +typedef int int32_t; +typedef unsigned int uint32_t; +typedef __int64 int64_t; +typedef unsigned __int64 uint64_t; +#define ntohl(x) _byteswap_ulong (x) +#define ntohs(x) _byteswap_ushort (x) +#define htonl(x) _byteswap_ulong (x) +#define htons(x) _byteswap_ushort (x) +#else +#include <arpa/inet.h> +#include <stdint.h> +#endif + +#include <cstdio> +#include <cstdlib> +#include <cstring> +#include <limits> + +namespace woff2 { + +#if defined(_MSC_VER) || !defined(FONT_COMPRESSION_DEBUG) +#define FONT_COMPRESSION_FAILURE() false +#else +#define FONT_COMPRESSION_FAILURE() \ + woff2::Failure(__FILE__, __LINE__, __PRETTY_FUNCTION__) +inline bool Failure(const char *f, int l, const char *fn) { + fprintf(stderr, "ERROR at %s:%d (%s)\n", f, l, fn); + fflush(stderr); + return false; +} +#endif + +// ----------------------------------------------------------------------------- +// Buffer helper class +// +// This class perform some trival buffer operations while checking for +// out-of-bounds errors. As a family they return false if anything is amiss, +// updating the current offset otherwise. +// ----------------------------------------------------------------------------- +class Buffer { + public: + Buffer(const uint8_t *data, size_t len) + : buffer_(data), + length_(len), + offset_(0) { } + + bool Skip(size_t n_bytes) { + return Read(NULL, n_bytes); + } + + bool Read(uint8_t *data, size_t n_bytes) { + if (n_bytes > 1024 * 1024 * 1024) { + return FONT_COMPRESSION_FAILURE(); + } + if ((offset_ + n_bytes > length_) || + (offset_ > length_ - n_bytes)) { + return FONT_COMPRESSION_FAILURE(); + } + if (data) { + std::memcpy(data, buffer_ + offset_, n_bytes); + } + offset_ += n_bytes; + return true; + } + + inline bool ReadU8(uint8_t *value) { + if (offset_ + 1 > length_) { + return FONT_COMPRESSION_FAILURE(); + } + *value = buffer_[offset_]; + ++offset_; + return true; + } + + bool ReadU16(uint16_t *value) { + if (offset_ + 2 > length_) { + return FONT_COMPRESSION_FAILURE(); + } + std::memcpy(value, buffer_ + offset_, sizeof(uint16_t)); + *value = ntohs(*value); + offset_ += 2; + return true; + } + + bool ReadS16(int16_t *value) { + return ReadU16(reinterpret_cast<uint16_t*>(value)); + } + + bool ReadU24(uint32_t *value) { + if (offset_ + 3 > length_) { + return FONT_COMPRESSION_FAILURE(); + } + *value = static_cast<uint32_t>(buffer_[offset_]) << 16 | + static_cast<uint32_t>(buffer_[offset_ + 1]) << 8 | + static_cast<uint32_t>(buffer_[offset_ + 2]); + offset_ += 3; + return true; + } + + bool ReadU32(uint32_t *value) { + if (offset_ + 4 > length_) { + return FONT_COMPRESSION_FAILURE(); + } + std::memcpy(value, buffer_ + offset_, sizeof(uint32_t)); + *value = ntohl(*value); + offset_ += 4; + return true; + } + + bool ReadS32(int32_t *value) { + return ReadU32(reinterpret_cast<uint32_t*>(value)); + } + + bool ReadTag(uint32_t *value) { + if (offset_ + 4 > length_) { + return FONT_COMPRESSION_FAILURE(); + } + std::memcpy(value, buffer_ + offset_, sizeof(uint32_t)); + offset_ += 4; + return true; + } + + bool ReadR64(uint64_t *value) { + if (offset_ + 8 > length_) { + return FONT_COMPRESSION_FAILURE(); + } + std::memcpy(value, buffer_ + offset_, sizeof(uint64_t)); + offset_ += 8; + return true; + } + + const uint8_t *buffer() const { return buffer_; } + size_t offset() const { return offset_; } + size_t length() const { return length_; } + + void set_offset(size_t newoffset) { offset_ = newoffset; } + + private: + const uint8_t * const buffer_; + const size_t length_; + size_t offset_; +}; + +} // namespace woff2 + +#endif // WOFF2_BUFFER_H_ diff --git a/modules/woff2/src/convert_woff2ttf_fuzzer.cc b/modules/woff2/src/convert_woff2ttf_fuzzer.cc new file mode 100644 index 0000000000..2d977a3ef2 --- /dev/null +++ b/modules/woff2/src/convert_woff2ttf_fuzzer.cc @@ -0,0 +1,13 @@ +#include <stddef.h> +#include <stdint.h> + +#include <woff2/decode.h> + +// Entry point for LibFuzzer. +extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) { + std::string buf; + woff2::WOFF2StringOut out(&buf); + out.SetMaxSize(30 * 1024 * 1024); + woff2::ConvertWOFF2ToTTF(data, size, &out); + return 0; +} diff --git a/modules/woff2/src/convert_woff2ttf_fuzzer_new_entry.cc b/modules/woff2/src/convert_woff2ttf_fuzzer_new_entry.cc new file mode 100644 index 0000000000..5ad93364bf --- /dev/null +++ b/modules/woff2/src/convert_woff2ttf_fuzzer_new_entry.cc @@ -0,0 +1,12 @@ +#include <string> +#include <woff2/decode.h> + +extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t data_size) { + // Decode using newer entry pattern. + // Same pattern as woff2_decompress. + std::string output(std::min(woff2::ComputeWOFF2FinalSize(data, data_size), + woff2::kDefaultMaxSize), 0); + woff2::WOFF2StringOut out(&output); + woff2::ConvertWOFF2ToTTF(data, data_size, &out); + return 0; +} diff --git a/modules/woff2/src/file.h b/modules/woff2/src/file.h new file mode 100644 index 0000000000..70ea7a7fb1 --- /dev/null +++ b/modules/woff2/src/file.h @@ -0,0 +1,30 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* File IO helpers. */ + +#ifndef WOFF2_FILE_H_ +#define WOFF2_FILE_H_ + +#include <fstream> +#include <iterator> + +namespace woff2 { + +inline std::string GetFileContent(std::string filename) { + std::ifstream ifs(filename.c_str(), std::ios::binary); + return std::string(std::istreambuf_iterator<char>(ifs.rdbuf()), + std::istreambuf_iterator<char>()); +} + +inline void SetFileContents(std::string filename, std::string::iterator start, + std::string::iterator end) { + std::ofstream ofs(filename.c_str(), std::ios::binary); + std::copy(start, end, std::ostream_iterator<char>(ofs)); +} + +} // namespace woff2 +#endif // WOFF2_FILE_H_ diff --git a/modules/woff2/src/font.cc b/modules/woff2/src/font.cc new file mode 100644 index 0000000000..a45153e9cb --- /dev/null +++ b/modules/woff2/src/font.cc @@ -0,0 +1,400 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Font management utilities */ + +#include "./font.h" + +#include <algorithm> + +#include "./buffer.h" +#include "./port.h" +#include "./store_bytes.h" +#include "./table_tags.h" +#include "./woff2_common.h" + +namespace woff2 { + +Font::Table* Font::FindTable(uint32_t tag) { + std::map<uint32_t, Font::Table>::iterator it = tables.find(tag); + return it == tables.end() ? 0 : &it->second; +} + +const Font::Table* Font::FindTable(uint32_t tag) const { + std::map<uint32_t, Font::Table>::const_iterator it = tables.find(tag); + return it == tables.end() ? 0 : &it->second; +} + +std::vector<uint32_t> Font::OutputOrderedTags() const { + std::vector<uint32_t> output_order; + + for (const auto& i : tables) { + const Font::Table& table = i.second; + // This is a transformed table, we will write it together with the + // original version. + if (table.tag & 0x80808080) { + continue; + } + output_order.push_back(table.tag); + } + + // Alphabetize then put loca immediately after glyf + auto glyf_loc = std::find(output_order.begin(), output_order.end(), + kGlyfTableTag); + auto loca_loc = std::find(output_order.begin(), output_order.end(), + kLocaTableTag); + if (glyf_loc != output_order.end() && loca_loc != output_order.end()) { + output_order.erase(loca_loc); + output_order.insert(std::find(output_order.begin(), output_order.end(), + kGlyfTableTag) + 1, kLocaTableTag); + } + + return output_order; +} + +bool ReadTrueTypeFont(Buffer* file, const uint8_t* data, size_t len, + Font* font) { + // We don't care about the search_range, entry_selector and range_shift + // fields, they will always be computed upon writing the font. + if (!file->ReadU16(&font->num_tables) || + !file->Skip(6)) { + return FONT_COMPRESSION_FAILURE(); + } + + std::map<uint32_t, uint32_t> intervals; + for (uint16_t i = 0; i < font->num_tables; ++i) { + Font::Table table; + table.flag_byte = 0; + table.reuse_of = NULL; + if (!file->ReadU32(&table.tag) || + !file->ReadU32(&table.checksum) || + !file->ReadU32(&table.offset) || + !file->ReadU32(&table.length)) { + return FONT_COMPRESSION_FAILURE(); + } + if ((table.offset & 3) != 0 || + table.length > len || + len - table.length < table.offset) { + return FONT_COMPRESSION_FAILURE(); + } + intervals[table.offset] = table.length; + table.data = data + table.offset; + if (font->tables.find(table.tag) != font->tables.end()) { + return FONT_COMPRESSION_FAILURE(); + } + font->tables[table.tag] = table; + } + + // Check that tables are non-overlapping. + uint32_t last_offset = 12UL + 16UL * font->num_tables; + for (const auto& i : intervals) { + if (i.first < last_offset || i.first + i.second < i.first) { + return FONT_COMPRESSION_FAILURE(); + } + last_offset = i.first + i.second; + } + + // Sanity check key tables + const Font::Table* head_table = font->FindTable(kHeadTableTag); + if (head_table != NULL && head_table->length < 52) { + return FONT_COMPRESSION_FAILURE(); + } + + return true; +} + +bool ReadCollectionFont(Buffer* file, const uint8_t* data, size_t len, + Font* font, + std::map<uint32_t, Font::Table*>* all_tables) { + if (!file->ReadU32(&font->flavor)) { + return FONT_COMPRESSION_FAILURE(); + } + if (!ReadTrueTypeFont(file, data, len, font)) { + return FONT_COMPRESSION_FAILURE(); + } + + for (auto& entry : font->tables) { + Font::Table& table = entry.second; + + if (all_tables->find(table.offset) == all_tables->end()) { + (*all_tables)[table.offset] = font->FindTable(table.tag); + } else { + table.reuse_of = (*all_tables)[table.offset]; + if (table.tag != table.reuse_of->tag) { + return FONT_COMPRESSION_FAILURE(); + } + } + + } + return true; +} + +bool ReadTrueTypeCollection(Buffer* file, const uint8_t* data, size_t len, + FontCollection* font_collection) { + uint32_t num_fonts; + + if (!file->ReadU32(&font_collection->header_version) || + !file->ReadU32(&num_fonts)) { + return FONT_COMPRESSION_FAILURE(); + } + + std::vector<uint32_t> offsets; + for (size_t i = 0; i < num_fonts; i++) { + uint32_t offset; + if (!file->ReadU32(&offset)) { + return FONT_COMPRESSION_FAILURE(); + } + offsets.push_back(offset); + } + + font_collection->fonts.resize(offsets.size()); + std::vector<Font>::iterator font_it = font_collection->fonts.begin(); + + std::map<uint32_t, Font::Table*> all_tables; + for (const auto offset : offsets) { + file->set_offset(offset); + Font& font = *font_it++; + if (!ReadCollectionFont(file, data, len, &font, &all_tables)) { + return FONT_COMPRESSION_FAILURE(); + } + } + + return true; +} + +bool ReadFont(const uint8_t* data, size_t len, Font* font) { + Buffer file(data, len); + + if (!file.ReadU32(&font->flavor)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (font->flavor == kTtcFontFlavor) { + return FONT_COMPRESSION_FAILURE(); + } + return ReadTrueTypeFont(&file, data, len, font); +} + +bool ReadFontCollection(const uint8_t* data, size_t len, + FontCollection* font_collection) { + Buffer file(data, len); + + if (!file.ReadU32(&font_collection->flavor)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (font_collection->flavor != kTtcFontFlavor) { + font_collection->fonts.resize(1); + Font& font = font_collection->fonts[0]; + font.flavor = font_collection->flavor; + return ReadTrueTypeFont(&file, data, len, &font); + } + return ReadTrueTypeCollection(&file, data, len, font_collection); +} + +size_t FontFileSize(const Font& font) { + size_t max_offset = 12ULL + 16ULL * font.num_tables; + for (const auto& i : font.tables) { + const Font::Table& table = i.second; + size_t padding_size = (4 - (table.length & 3)) & 3; + size_t end_offset = (padding_size + table.offset) + table.length; + max_offset = std::max(max_offset, end_offset); + } + return max_offset; +} + +size_t FontCollectionFileSize(const FontCollection& font_collection) { + size_t max_offset = 0; + for (auto& font : font_collection.fonts) { + // font file size actually just finds max offset + max_offset = std::max(max_offset, FontFileSize(font)); + } + return max_offset; +} + +bool WriteFont(const Font& font, uint8_t* dst, size_t dst_size) { + size_t offset = 0; + return WriteFont(font, &offset, dst, dst_size); +} + +bool WriteTableRecord(const Font::Table* table, size_t* offset, uint8_t* dst, + size_t dst_size) { + if (dst_size < *offset + kSfntEntrySize) { + return FONT_COMPRESSION_FAILURE(); + } + if (table->IsReused()) { + table = table->reuse_of; + } + StoreU32(table->tag, offset, dst); + StoreU32(table->checksum, offset, dst); + StoreU32(table->offset, offset, dst); + StoreU32(table->length, offset, dst); + return true; +} + +bool WriteTable(const Font::Table& table, size_t* offset, uint8_t* dst, + size_t dst_size) { + if (!WriteTableRecord(&table, offset, dst, dst_size)) { + return false; + } + + // Write the actual table data if it's the first time we've seen it + if (!table.IsReused()) { + if (table.offset + table.length < table.offset || + dst_size < table.offset + table.length) { + return FONT_COMPRESSION_FAILURE(); + } + memcpy(dst + table.offset, table.data, table.length); + size_t padding_size = (4 - (table.length & 3)) & 3; + if (table.offset + table.length + padding_size < padding_size || + dst_size < table.offset + table.length + padding_size) { + return FONT_COMPRESSION_FAILURE(); + } + memset(dst + table.offset + table.length, 0, padding_size); + } + return true; +} + +bool WriteFont(const Font& font, size_t* offset, uint8_t* dst, + size_t dst_size) { + if (dst_size < 12ULL + 16ULL * font.num_tables) { + return FONT_COMPRESSION_FAILURE(); + } + StoreU32(font.flavor, offset, dst); + Store16(font.num_tables, offset, dst); + uint16_t max_pow2 = font.num_tables ? Log2Floor(font.num_tables) : 0; + uint16_t search_range = max_pow2 ? 1 << (max_pow2 + 4) : 0; + uint16_t range_shift = (font.num_tables << 4) - search_range; + Store16(search_range, offset, dst); + Store16(max_pow2, offset, dst); + Store16(range_shift, offset, dst); + + for (const auto& i : font.tables) { + if (!WriteTable(i.second, offset, dst, dst_size)) { + return false; + } + } + + return true; +} + +bool WriteFontCollection(const FontCollection& font_collection, uint8_t* dst, + size_t dst_size) { + size_t offset = 0; + + // It's simpler if this just a simple sfnt + if (font_collection.flavor != kTtcFontFlavor) { + return WriteFont(font_collection.fonts[0], &offset, dst, dst_size); + } + + // Write TTC header + StoreU32(kTtcFontFlavor, &offset, dst); + StoreU32(font_collection.header_version, &offset, dst); + StoreU32(font_collection.fonts.size(), &offset, dst); + + // Offset Table, zeroed for now + size_t offset_table = offset; // where to write offsets later + for (size_t i = 0; i < font_collection.fonts.size(); i++) { + StoreU32(0, &offset, dst); + } + + if (font_collection.header_version == 0x00020000) { + StoreU32(0, &offset, dst); // ulDsigTag + StoreU32(0, &offset, dst); // ulDsigLength + StoreU32(0, &offset, dst); // ulDsigOffset + } + + // Write fonts and their offsets. + for (size_t i = 0; i < font_collection.fonts.size(); i++) { + const auto& font = font_collection.fonts[i]; + StoreU32(offset, &offset_table, dst); + if (!WriteFont(font, &offset, dst, dst_size)) { + return false; + } + } + + return true; +} + +int NumGlyphs(const Font& font) { + const Font::Table* head_table = font.FindTable(kHeadTableTag); + const Font::Table* loca_table = font.FindTable(kLocaTableTag); + if (head_table == NULL || loca_table == NULL || head_table->length < 52) { + return 0; + } + int index_fmt = IndexFormat(font); + int loca_record_size = (index_fmt == 0 ? 2 : 4); + if (loca_table->length < loca_record_size) { + return 0; + } + return (loca_table->length / loca_record_size) - 1; +} + +int IndexFormat(const Font& font) { + const Font::Table* head_table = font.FindTable(kHeadTableTag); + if (head_table == NULL) { + return 0; + } + return head_table->data[51]; +} + +bool Font::Table::IsReused() const { + return this->reuse_of != NULL; +} + +bool GetGlyphData(const Font& font, int glyph_index, + const uint8_t** glyph_data, size_t* glyph_size) { + if (glyph_index < 0) { + return FONT_COMPRESSION_FAILURE(); + } + const Font::Table* head_table = font.FindTable(kHeadTableTag); + const Font::Table* loca_table = font.FindTable(kLocaTableTag); + const Font::Table* glyf_table = font.FindTable(kGlyfTableTag); + if (head_table == NULL || loca_table == NULL || glyf_table == NULL || + head_table->length < 52) { + return FONT_COMPRESSION_FAILURE(); + } + + int index_fmt = IndexFormat(font); + + Buffer loca_buf(loca_table->data, loca_table->length); + if (index_fmt == 0) { + uint16_t offset1, offset2; + if (!loca_buf.Skip(2 * glyph_index) || + !loca_buf.ReadU16(&offset1) || + !loca_buf.ReadU16(&offset2) || + offset2 < offset1 || + 2 * offset2 > glyf_table->length) { + return FONT_COMPRESSION_FAILURE(); + } + *glyph_data = glyf_table->data + 2 * offset1; + *glyph_size = 2 * (offset2 - offset1); + } else { + uint32_t offset1, offset2; + if (!loca_buf.Skip(4 * glyph_index) || + !loca_buf.ReadU32(&offset1) || + !loca_buf.ReadU32(&offset2) || + offset2 < offset1 || + offset2 > glyf_table->length) { + return FONT_COMPRESSION_FAILURE(); + } + *glyph_data = glyf_table->data + offset1; + *glyph_size = offset2 - offset1; + } + return true; +} + +bool RemoveDigitalSignature(Font* font) { + std::map<uint32_t, Font::Table>::iterator it = + font->tables.find(kDsigTableTag); + if (it != font->tables.end()) { + font->tables.erase(it); + font->num_tables = font->tables.size(); + } + return true; +} + +} // namespace woff2 diff --git a/modules/woff2/src/font.h b/modules/woff2/src/font.h new file mode 100644 index 0000000000..7b8ddddc5a --- /dev/null +++ b/modules/woff2/src/font.h @@ -0,0 +1,105 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Data model for a font file in sfnt format, reading and writing functions and + accessors for the glyph data. */ + +#ifndef WOFF2_FONT_H_ +#define WOFF2_FONT_H_ + +#include <stddef.h> +#include <inttypes.h> +#include <map> +#include <vector> + +namespace woff2 { + +// Represents an sfnt font file. Only the table directory is parsed, for the +// table data we only store a raw pointer, therefore a font object is valid only +// as long the data from which it was parsed is around. +struct Font { + uint32_t flavor; + uint16_t num_tables; + + struct Table { + uint32_t tag; + uint32_t checksum; + uint32_t offset; + uint32_t length; + const uint8_t* data; + + // Buffer used to mutate the data before writing out. + std::vector<uint8_t> buffer; + + // If we've seen this tag/offset before, pointer to the first time we saw it + // If this is the first time we've seen this table, NULL + // Intended use is to bypass re-processing tables + Font::Table* reuse_of; + + uint8_t flag_byte; + + // Is this table reused by a TTC + bool IsReused() const; + }; + std::map<uint32_t, Table> tables; + std::vector<uint32_t> OutputOrderedTags() const; + + Table* FindTable(uint32_t tag); + const Table* FindTable(uint32_t tag) const; +}; + +// Accomodates both singular (OTF, TTF) and collection (TTC) fonts +struct FontCollection { + uint32_t flavor; + uint32_t header_version; + // (offset, first use of table*) pairs + std::map<uint32_t, Font::Table*> tables; + std::vector<Font> fonts; +}; + +// Parses the font from the given data. Returns false on parsing failure or +// buffer overflow. The font is valid only so long the input data pointer is +// valid. Does NOT support collections. +bool ReadFont(const uint8_t* data, size_t len, Font* font); + +// Parses the font from the given data. Returns false on parsing failure or +// buffer overflow. The font is valid only so long the input data pointer is +// valid. Supports collections. +bool ReadFontCollection(const uint8_t* data, size_t len, FontCollection* fonts); + +// Returns the file size of the font. +size_t FontFileSize(const Font& font); +size_t FontCollectionFileSize(const FontCollection& font); + +// Writes the font into the specified dst buffer. The dst_size should be the +// same as returned by FontFileSize(). Returns false upon buffer overflow (which +// should not happen if dst_size was computed by FontFileSize()). +bool WriteFont(const Font& font, uint8_t* dst, size_t dst_size); +// Write the font at a specific offset +bool WriteFont(const Font& font, size_t* offset, uint8_t* dst, size_t dst_size); + +bool WriteFontCollection(const FontCollection& font_collection, uint8_t* dst, + size_t dst_size); + +// Returns the number of glyphs in the font. +// NOTE: Currently this works only for TrueType-flavored fonts, will return +// zero for CFF-flavored fonts. +int NumGlyphs(const Font& font); + +// Returns the index format of the font +int IndexFormat(const Font& font); + +// Sets *glyph_data and *glyph_size to point to the location of the glyph data +// with the given index. Returns false if the glyph is not found. +bool GetGlyphData(const Font& font, int glyph_index, + const uint8_t** glyph_data, size_t* glyph_size); + +// Removes the digital signature (DSIG) table +bool RemoveDigitalSignature(Font* font); + +} // namespace woff2 + +#endif // WOFF2_FONT_H_ diff --git a/modules/woff2/src/glyph.cc b/modules/woff2/src/glyph.cc new file mode 100644 index 0000000000..5b4948679c --- /dev/null +++ b/modules/woff2/src/glyph.cc @@ -0,0 +1,383 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Glyph manipulation */ + +#include "./glyph.h" + +#include <stdlib.h> +#include <limits> +#include "./buffer.h" +#include "./store_bytes.h" + +namespace woff2 { + +static const int32_t kFLAG_ONCURVE = 1; +static const int32_t kFLAG_XSHORT = 1 << 1; +static const int32_t kFLAG_YSHORT = 1 << 2; +static const int32_t kFLAG_REPEAT = 1 << 3; +static const int32_t kFLAG_XREPEATSIGN = 1 << 4; +static const int32_t kFLAG_YREPEATSIGN = 1 << 5; +static const int32_t kFLAG_OVERLAP_SIMPLE = 1 << 6; +static const int32_t kFLAG_ARG_1_AND_2_ARE_WORDS = 1 << 0; +static const int32_t kFLAG_WE_HAVE_A_SCALE = 1 << 3; +static const int32_t kFLAG_MORE_COMPONENTS = 1 << 5; +static const int32_t kFLAG_WE_HAVE_AN_X_AND_Y_SCALE = 1 << 6; +static const int32_t kFLAG_WE_HAVE_A_TWO_BY_TWO = 1 << 7; +static const int32_t kFLAG_WE_HAVE_INSTRUCTIONS = 1 << 8; + +bool ReadCompositeGlyphData(Buffer* buffer, Glyph* glyph) { + glyph->have_instructions = false; + glyph->composite_data = buffer->buffer() + buffer->offset(); + size_t start_offset = buffer->offset(); + uint16_t flags = kFLAG_MORE_COMPONENTS; + while (flags & kFLAG_MORE_COMPONENTS) { + if (!buffer->ReadU16(&flags)) { + return FONT_COMPRESSION_FAILURE(); + } + glyph->have_instructions |= (flags & kFLAG_WE_HAVE_INSTRUCTIONS) != 0; + size_t arg_size = 2; // glyph index + if (flags & kFLAG_ARG_1_AND_2_ARE_WORDS) { + arg_size += 4; + } else { + arg_size += 2; + } + if (flags & kFLAG_WE_HAVE_A_SCALE) { + arg_size += 2; + } else if (flags & kFLAG_WE_HAVE_AN_X_AND_Y_SCALE) { + arg_size += 4; + } else if (flags & kFLAG_WE_HAVE_A_TWO_BY_TWO) { + arg_size += 8; + } + if (!buffer->Skip(arg_size)) { + return FONT_COMPRESSION_FAILURE(); + } + } + if (buffer->offset() - start_offset > std::numeric_limits<uint32_t>::max()) { + return FONT_COMPRESSION_FAILURE(); + } + glyph->composite_data_size = buffer->offset() - start_offset; + return true; +} + +bool ReadGlyph(const uint8_t* data, size_t len, Glyph* glyph) { + Buffer buffer(data, len); + + int16_t num_contours; + if (!buffer.ReadS16(&num_contours)) { + return FONT_COMPRESSION_FAILURE(); + } + + // Read the bounding box. + if (!buffer.ReadS16(&glyph->x_min) || + !buffer.ReadS16(&glyph->y_min) || + !buffer.ReadS16(&glyph->x_max) || + !buffer.ReadS16(&glyph->y_max)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (num_contours == 0) { + // Empty glyph. + return true; + } + + if (num_contours > 0) { + // Simple glyph. + glyph->contours.resize(num_contours); + + // Read the number of points per contour. + uint16_t last_point_index = 0; + for (int i = 0; i < num_contours; ++i) { + uint16_t point_index; + if (!buffer.ReadU16(&point_index)) { + return FONT_COMPRESSION_FAILURE(); + } + uint16_t num_points = point_index - last_point_index + (i == 0 ? 1 : 0); + glyph->contours[i].resize(num_points); + last_point_index = point_index; + } + + // Read the instructions. + if (!buffer.ReadU16(&glyph->instructions_size)) { + return FONT_COMPRESSION_FAILURE(); + } + glyph->instructions_data = data + buffer.offset(); + if (!buffer.Skip(glyph->instructions_size)) { + return FONT_COMPRESSION_FAILURE(); + } + + // Read the run-length coded flags. + std::vector<std::vector<uint8_t> > flags(num_contours); + { + uint8_t flag = 0; + uint8_t flag_repeat = 0; + for (int i = 0; i < num_contours; ++i) { + flags[i].resize(glyph->contours[i].size()); + for (size_t j = 0; j < glyph->contours[i].size(); ++j) { + if (flag_repeat == 0) { + if (!buffer.ReadU8(&flag)) { + return FONT_COMPRESSION_FAILURE(); + } + if (flag & kFLAG_REPEAT) { + if (!buffer.ReadU8(&flag_repeat)) { + return FONT_COMPRESSION_FAILURE(); + } + } + } else { + flag_repeat--; + } + flags[i][j] = flag; + glyph->contours[i][j].on_curve = flag & kFLAG_ONCURVE; + } + } + } + + if (!flags.empty() && !flags[0].empty()) { + glyph->overlap_simple_flag_set = (flags[0][0] & kFLAG_OVERLAP_SIMPLE); + } + + // Read the x coordinates. + int prev_x = 0; + for (int i = 0; i < num_contours; ++i) { + for (size_t j = 0; j < glyph->contours[i].size(); ++j) { + uint8_t flag = flags[i][j]; + if (flag & kFLAG_XSHORT) { + // single byte x-delta coord value + uint8_t x_delta; + if (!buffer.ReadU8(&x_delta)) { + return FONT_COMPRESSION_FAILURE(); + } + int sign = (flag & kFLAG_XREPEATSIGN) ? 1 : -1; + glyph->contours[i][j].x = prev_x + sign * x_delta; + } else { + // double byte x-delta coord value + int16_t x_delta = 0; + if (!(flag & kFLAG_XREPEATSIGN)) { + if (!buffer.ReadS16(&x_delta)) { + return FONT_COMPRESSION_FAILURE(); + } + } + glyph->contours[i][j].x = prev_x + x_delta; + } + prev_x = glyph->contours[i][j].x; + } + } + + // Read the y coordinates. + int prev_y = 0; + for (int i = 0; i < num_contours; ++i) { + for (size_t j = 0; j < glyph->contours[i].size(); ++j) { + uint8_t flag = flags[i][j]; + if (flag & kFLAG_YSHORT) { + // single byte y-delta coord value + uint8_t y_delta; + if (!buffer.ReadU8(&y_delta)) { + return FONT_COMPRESSION_FAILURE(); + } + int sign = (flag & kFLAG_YREPEATSIGN) ? 1 : -1; + glyph->contours[i][j].y = prev_y + sign * y_delta; + } else { + // double byte y-delta coord value + int16_t y_delta = 0; + if (!(flag & kFLAG_YREPEATSIGN)) { + if (!buffer.ReadS16(&y_delta)) { + return FONT_COMPRESSION_FAILURE(); + } + } + glyph->contours[i][j].y = prev_y + y_delta; + } + prev_y = glyph->contours[i][j].y; + } + } + } else if (num_contours == -1) { + // Composite glyph. + if (!ReadCompositeGlyphData(&buffer, glyph)) { + return FONT_COMPRESSION_FAILURE(); + } + // Read the instructions. + if (glyph->have_instructions) { + if (!buffer.ReadU16(&glyph->instructions_size)) { + return FONT_COMPRESSION_FAILURE(); + } + glyph->instructions_data = data + buffer.offset(); + if (!buffer.Skip(glyph->instructions_size)) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + glyph->instructions_size = 0; + } + } else { + return FONT_COMPRESSION_FAILURE(); + } + return true; +} + +namespace { + +void StoreBbox(const Glyph& glyph, size_t* offset, uint8_t* dst) { + Store16(glyph.x_min, offset, dst); + Store16(glyph.y_min, offset, dst); + Store16(glyph.x_max, offset, dst); + Store16(glyph.y_max, offset, dst); +} + +void StoreInstructions(const Glyph& glyph, size_t* offset, uint8_t* dst) { + Store16(glyph.instructions_size, offset, dst); + StoreBytes(glyph.instructions_data, glyph.instructions_size, offset, dst); +} + +bool StoreEndPtsOfContours(const Glyph& glyph, size_t* offset, uint8_t* dst) { + int end_point = -1; + for (const auto& contour : glyph.contours) { + end_point += contour.size(); + if (contour.size() > std::numeric_limits<uint16_t>::max() || + end_point > std::numeric_limits<uint16_t>::max()) { + return FONT_COMPRESSION_FAILURE(); + } + Store16(end_point, offset, dst); + } + return true; +} + +bool StorePoints(const Glyph& glyph, size_t* offset, + uint8_t* dst, size_t dst_size) { + int previous_flag = -1; + int repeat_count = 0; + int last_x = 0; + int last_y = 0; + size_t x_bytes = 0; + size_t y_bytes = 0; + + // Store the flags and calculate the total size of the x and y coordinates. + for (const auto& contour : glyph.contours) { + for (const auto& point : contour) { + int flag = point.on_curve ? kFLAG_ONCURVE : 0; + if (previous_flag == -1 && glyph.overlap_simple_flag_set) { + // First flag needs to have overlap simple bit set. + flag = flag | kFLAG_OVERLAP_SIMPLE; + } + int dx = point.x - last_x; + int dy = point.y - last_y; + if (dx == 0) { + flag |= kFLAG_XREPEATSIGN; + } else if (dx > -256 && dx < 256) { + flag |= kFLAG_XSHORT | (dx > 0 ? kFLAG_XREPEATSIGN : 0); + x_bytes += 1; + } else { + x_bytes += 2; + } + if (dy == 0) { + flag |= kFLAG_YREPEATSIGN; + } else if (dy > -256 && dy < 256) { + flag |= kFLAG_YSHORT | (dy > 0 ? kFLAG_YREPEATSIGN : 0); + y_bytes += 1; + } else { + y_bytes += 2; + } + if (flag == previous_flag && repeat_count != 255) { + dst[*offset - 1] |= kFLAG_REPEAT; + repeat_count++; + } else { + if (repeat_count != 0) { + if (*offset >= dst_size) { + return FONT_COMPRESSION_FAILURE(); + } + dst[(*offset)++] = repeat_count; + } + if (*offset >= dst_size) { + return FONT_COMPRESSION_FAILURE(); + } + dst[(*offset)++] = flag; + repeat_count = 0; + } + last_x = point.x; + last_y = point.y; + previous_flag = flag; + } + } + if (repeat_count != 0) { + if (*offset >= dst_size) { + return FONT_COMPRESSION_FAILURE(); + } + dst[(*offset)++] = repeat_count; + } + + if (*offset + x_bytes + y_bytes > dst_size) { + return FONT_COMPRESSION_FAILURE(); + } + + // Store the x and y coordinates. + size_t x_offset = *offset; + size_t y_offset = *offset + x_bytes; + last_x = 0; + last_y = 0; + for (const auto& contour : glyph.contours) { + for (const auto& point : contour) { + int dx = point.x - last_x; + int dy = point.y - last_y; + if (dx == 0) { + // pass + } else if (dx > -256 && dx < 256) { + dst[x_offset++] = std::abs(dx); + } else { + Store16(dx, &x_offset, dst); + } + if (dy == 0) { + // pass + } else if (dy > -256 && dy < 256) { + dst[y_offset++] = std::abs(dy); + } else { + Store16(dy, &y_offset, dst); + } + last_x += dx; + last_y += dy; + } + } + *offset = y_offset; + return true; +} + +} // namespace + +bool StoreGlyph(const Glyph& glyph, uint8_t* dst, size_t* dst_size) { + size_t offset = 0; + if (glyph.composite_data_size > 0) { + // Composite glyph. + if (*dst_size < ((10ULL + glyph.composite_data_size) + + ((glyph.have_instructions ? 2ULL : 0) + + glyph.instructions_size))) { + return FONT_COMPRESSION_FAILURE(); + } + Store16(-1, &offset, dst); + StoreBbox(glyph, &offset, dst); + StoreBytes(glyph.composite_data, glyph.composite_data_size, &offset, dst); + if (glyph.have_instructions) { + StoreInstructions(glyph, &offset, dst); + } + } else if (glyph.contours.size() > 0) { + // Simple glyph. + if (glyph.contours.size() > std::numeric_limits<int16_t>::max()) { + return FONT_COMPRESSION_FAILURE(); + } + if (*dst_size < ((12ULL + 2 * glyph.contours.size()) + + glyph.instructions_size)) { + return FONT_COMPRESSION_FAILURE(); + } + Store16(glyph.contours.size(), &offset, dst); + StoreBbox(glyph, &offset, dst); + if (!StoreEndPtsOfContours(glyph, &offset, dst)) { + return FONT_COMPRESSION_FAILURE(); + } + StoreInstructions(glyph, &offset, dst); + if (!StorePoints(glyph, &offset, dst, *dst_size)) { + return FONT_COMPRESSION_FAILURE(); + } + } + *dst_size = offset; + return true; +} + +} // namespace woff2 diff --git a/modules/woff2/src/glyph.h b/modules/woff2/src/glyph.h new file mode 100644 index 0000000000..c38eb5f2bc --- /dev/null +++ b/modules/woff2/src/glyph.h @@ -0,0 +1,71 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Data model and I/O for glyph data within sfnt format files for the purpose of + performing the preprocessing step of the WOFF 2.0 conversion. */ + +#ifndef WOFF2_GLYPH_H_ +#define WOFF2_GLYPH_H_ + +#include <inttypes.h> +#include <stddef.h> + +#include <cstdint> +#include <vector> + +namespace woff2 { + +// Represents a parsed simple or composite glyph. The composite glyph data and +// instructions are un-parsed and we keep only pointers to the raw data, +// therefore the glyph is valid only so long the data from which it was parsed +// is around. +class Glyph { + public: + Glyph() + : instructions_size(0), + overlap_simple_flag_set(false), + composite_data_size(0) {} + + // Bounding box. + int16_t x_min; + int16_t x_max; + int16_t y_min; + int16_t y_max; + + // Instructions. + uint16_t instructions_size; + const uint8_t* instructions_data; + + // Flags. + bool overlap_simple_flag_set; + + // Data model for simple glyphs. + struct Point { + int x; + int y; + bool on_curve; + }; + std::vector<std::vector<Point> > contours; + + // Data for composite glyphs. + const uint8_t* composite_data; + uint32_t composite_data_size; + bool have_instructions; +}; + +// Parses the glyph from the given data. Returns false on parsing failure or +// buffer overflow. The glyph is valid only so long the input data pointer is +// valid. +bool ReadGlyph(const uint8_t* data, size_t len, Glyph* glyph); + +// Stores the glyph into the specified dst buffer. The *dst_size is the buffer +// size on entry and is set to the actual (unpadded) stored size on exit. +// Returns false on buffer overflow. +bool StoreGlyph(const Glyph& glyph, uint8_t* dst, size_t* dst_size); + +} // namespace woff2 + +#endif // WOFF2_GLYPH_H_ diff --git a/modules/woff2/src/normalize.cc b/modules/woff2/src/normalize.cc new file mode 100644 index 0000000000..6685e08752 --- /dev/null +++ b/modules/woff2/src/normalize.cc @@ -0,0 +1,314 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Glyph normalization */ + +#include "./normalize.h" + +#include <inttypes.h> +#include <stddef.h> + +#include "./buffer.h" +#include "./port.h" +#include "./font.h" +#include "./glyph.h" +#include "./round.h" +#include "./store_bytes.h" +#include "./table_tags.h" +#include "./woff2_common.h" + +namespace woff2 { + +namespace { + +void StoreLoca(int index_fmt, uint32_t value, size_t* offset, uint8_t* dst) { + if (index_fmt == 0) { + Store16(value >> 1, offset, dst); + } else { + StoreU32(value, offset, dst); + } +} + +} // namespace + +namespace { + +bool WriteNormalizedLoca(int index_fmt, int num_glyphs, Font* font) { + Font::Table* glyf_table = font->FindTable(kGlyfTableTag); + Font::Table* loca_table = font->FindTable(kLocaTableTag); + + int glyph_sz = index_fmt == 0 ? 2 : 4; + loca_table->buffer.resize(Round4(num_glyphs + 1) * glyph_sz); + loca_table->length = (num_glyphs + 1) * glyph_sz; + + uint8_t* glyf_dst = num_glyphs ? &glyf_table->buffer[0] : NULL; + uint8_t* loca_dst = &loca_table->buffer[0]; + uint32_t glyf_offset = 0; + size_t loca_offset = 0; + + for (int i = 0; i < num_glyphs; ++i) { + StoreLoca(index_fmt, glyf_offset, &loca_offset, loca_dst); + Glyph glyph; + const uint8_t* glyph_data; + size_t glyph_size; + if (!GetGlyphData(*font, i, &glyph_data, &glyph_size) || + (glyph_size > 0 && !ReadGlyph(glyph_data, glyph_size, &glyph))) { + return FONT_COMPRESSION_FAILURE(); + } + size_t glyf_dst_size = glyf_table->buffer.size() - glyf_offset; + if (!StoreGlyph(glyph, glyf_dst + glyf_offset, &glyf_dst_size)) { + return FONT_COMPRESSION_FAILURE(); + } + glyf_dst_size = Round4(glyf_dst_size); + if (glyf_dst_size > std::numeric_limits<uint32_t>::max() || + glyf_offset + static_cast<uint32_t>(glyf_dst_size) < glyf_offset || + (index_fmt == 0 && glyf_offset + glyf_dst_size >= (1UL << 17))) { + return FONT_COMPRESSION_FAILURE(); + } + glyf_offset += glyf_dst_size; + } + + StoreLoca(index_fmt, glyf_offset, &loca_offset, loca_dst); + + glyf_table->buffer.resize(glyf_offset); + glyf_table->data = glyf_offset ? &glyf_table->buffer[0] : NULL; + glyf_table->length = glyf_offset; + loca_table->data = loca_offset ? &loca_table->buffer[0] : NULL; + + return true; +} + +} // namespace + +namespace { + +bool MakeEditableBuffer(Font* font, int tableTag) { + Font::Table* table = font->FindTable(tableTag); + if (table == NULL) { + return FONT_COMPRESSION_FAILURE(); + } + if (table->IsReused()) { + return true; + } + int sz = Round4(table->length); + table->buffer.resize(sz); + uint8_t* buf = &table->buffer[0]; + memcpy(buf, table->data, table->length); + if (PREDICT_FALSE(sz > table->length)) { + memset(buf + table->length, 0, sz - table->length); + } + table->data = buf; + return true; +} + +} // namespace + +bool NormalizeGlyphs(Font* font) { + Font::Table* head_table = font->FindTable(kHeadTableTag); + Font::Table* glyf_table = font->FindTable(kGlyfTableTag); + Font::Table* loca_table = font->FindTable(kLocaTableTag); + if (head_table == NULL) { + return FONT_COMPRESSION_FAILURE(); + } + // If you don't have glyf/loca this transform isn't very interesting + if (loca_table == NULL && glyf_table == NULL) { + return true; + } + // It would be best if you didn't have just one of glyf/loca + if ((glyf_table == NULL) != (loca_table == NULL)) { + return FONT_COMPRESSION_FAILURE(); + } + // Must share neither or both loca & glyf + if (loca_table->IsReused() != glyf_table->IsReused()) { + return FONT_COMPRESSION_FAILURE(); + } + if (loca_table->IsReused()) { + return true; + } + + int index_fmt = head_table->data[51]; + int num_glyphs = NumGlyphs(*font); + + // We need to allocate a bit more than its original length for the normalized + // glyf table, since it can happen that the glyphs in the original table are + // 2-byte aligned, while in the normalized table they are 4-byte aligned. + // That gives a maximum of 2 bytes increase per glyph. However, there is no + // theoretical guarantee that the total size of the flags plus the coordinates + // is the smallest possible in the normalized version, so we have to allow + // some general overhead. + // TODO(user) Figure out some more precise upper bound on the size of + // the overhead. + size_t max_normalized_glyf_size = 1.1 * glyf_table->length + 2 * num_glyphs; + + glyf_table->buffer.resize(max_normalized_glyf_size); + + // if we can't write a loca using short's (index_fmt 0) + // try again using longs (index_fmt 1) + if (!WriteNormalizedLoca(index_fmt, num_glyphs, font)) { + if (index_fmt != 0) { + return FONT_COMPRESSION_FAILURE(); + } + + // Rewrite loca with 4-byte entries & update head to match + index_fmt = 1; + if (!WriteNormalizedLoca(index_fmt, num_glyphs, font)) { + return FONT_COMPRESSION_FAILURE(); + } + head_table->buffer[51] = 1; + } + + return true; +} + +bool NormalizeOffsets(Font* font) { + uint32_t offset = 12 + 16 * font->num_tables; + for (auto tag : font->OutputOrderedTags()) { + auto& table = font->tables[tag]; + table.offset = offset; + offset += Round4(table.length); + } + return true; +} + +namespace { + +uint32_t ComputeHeaderChecksum(const Font& font) { + uint32_t checksum = font.flavor; + uint16_t max_pow2 = font.num_tables ? Log2Floor(font.num_tables) : 0; + uint16_t search_range = max_pow2 ? 1 << (max_pow2 + 4) : 0; + uint16_t range_shift = (font.num_tables << 4) - search_range; + checksum += (font.num_tables << 16 | search_range); + checksum += (max_pow2 << 16 | range_shift); + for (const auto& i : font.tables) { + const Font::Table* table = &i.second; + if (table->IsReused()) { + table = table->reuse_of; + } + checksum += table->tag; + checksum += table->checksum; + checksum += table->offset; + checksum += table->length; + } + return checksum; +} + +} // namespace + +bool FixChecksums(Font* font) { + Font::Table* head_table = font->FindTable(kHeadTableTag); + if (head_table == NULL) { + return FONT_COMPRESSION_FAILURE(); + } + if (head_table->reuse_of != NULL) { + head_table = head_table->reuse_of; + } + if (head_table->length < 12) { + return FONT_COMPRESSION_FAILURE(); + } + + uint8_t* head_buf = &head_table->buffer[0]; + size_t offset = 8; + StoreU32(0, &offset, head_buf); + uint32_t file_checksum = 0; + uint32_t head_checksum = 0; + for (auto& i : font->tables) { + Font::Table* table = &i.second; + if (table->IsReused()) { + table = table->reuse_of; + } + table->checksum = ComputeULongSum(table->data, table->length); + file_checksum += table->checksum; + + if (table->tag == kHeadTableTag) { + head_checksum = table->checksum; + } + } + + file_checksum += ComputeHeaderChecksum(*font); + offset = 8; + StoreU32(0xb1b0afba - file_checksum, &offset, head_buf); + + return true; +} + +namespace { +bool MarkTransformed(Font* font) { + Font::Table* head_table = font->FindTable(kHeadTableTag); + if (head_table == NULL) { + return FONT_COMPRESSION_FAILURE(); + } + if (head_table->reuse_of != NULL) { + head_table = head_table->reuse_of; + } + if (head_table->length < 17) { + return FONT_COMPRESSION_FAILURE(); + } + // set bit 11 of head table 'flags' to indicate that font has undergone + // lossless modifying transform + int head_flags = head_table->data[16]; + head_table->buffer[16] = head_flags | 0x08; + return true; +} +} // namespace + + +bool NormalizeWithoutFixingChecksums(Font* font) { + return (MakeEditableBuffer(font, kHeadTableTag) && + RemoveDigitalSignature(font) && + MarkTransformed(font) && + NormalizeGlyphs(font) && + NormalizeOffsets(font)); +} + +bool NormalizeFont(Font* font) { + return (NormalizeWithoutFixingChecksums(font) && + FixChecksums(font)); +} + +bool NormalizeFontCollection(FontCollection* font_collection) { + if (font_collection->fonts.size() == 1) { + return NormalizeFont(&font_collection->fonts[0]); + } + + uint32_t offset = CollectionHeaderSize(font_collection->header_version, + font_collection->fonts.size()); + for (auto& font : font_collection->fonts) { + if (!NormalizeWithoutFixingChecksums(&font)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Font normalization failed.\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + offset += kSfntHeaderSize + kSfntEntrySize * font.num_tables; + } + + // Start table offsets after TTC Header and Sfnt Headers + for (auto& font : font_collection->fonts) { + for (auto tag : font.OutputOrderedTags()) { + Font::Table& table = font.tables[tag]; + if (table.IsReused()) { + table.offset = table.reuse_of->offset; + } else { + table.offset = offset; + offset += Round4(table.length); + } + } + } + + // Now we can fix the checksums + for (auto& font : font_collection->fonts) { + if (!FixChecksums(&font)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Failed to fix checksums\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + } + + return true; +} + +} // namespace woff2 diff --git a/modules/woff2/src/normalize.h b/modules/woff2/src/normalize.h new file mode 100644 index 0000000000..c6fee74db6 --- /dev/null +++ b/modules/woff2/src/normalize.h @@ -0,0 +1,39 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Functions for normalizing fonts. Since the WOFF 2.0 decoder creates font + files in normalized form, the WOFF 2.0 conversion is guaranteed to be + lossless (in a bitwise sense) only for normalized font files. */ + +#ifndef WOFF2_NORMALIZE_H_ +#define WOFF2_NORMALIZE_H_ + +namespace woff2 { + +struct Font; +struct FontCollection; + +// Changes the offset fields of the table headers so that the data for the +// tables will be written in order of increasing tag values, without any gaps +// other than the 4-byte padding. +bool NormalizeOffsets(Font* font); + +// Changes the checksum fields of the table headers and the checksum field of +// the head table so that it matches the current data. +bool FixChecksums(Font* font); + +// Parses each of the glyphs in the font and writes them again to the glyf +// table in normalized form, as defined by the StoreGlyph() function. Changes +// the loca table accordigly. +bool NormalizeGlyphs(Font* font); + +// Performs all of the normalization steps above. +bool NormalizeFont(Font* font); +bool NormalizeFontCollection(FontCollection* font_collection); + +} // namespace woff2 + +#endif // WOFF2_NORMALIZE_H_ diff --git a/modules/woff2/src/port.h b/modules/woff2/src/port.h new file mode 100644 index 0000000000..8b60fee8bc --- /dev/null +++ b/modules/woff2/src/port.h @@ -0,0 +1,66 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Helper function for bit twiddling and macros for branch prediction. */ + +#ifndef WOFF2_PORT_H_ +#define WOFF2_PORT_H_ + +#include <assert.h> + +namespace woff2 { + +typedef unsigned int uint32; + +inline int Log2Floor(uint32 n) { +#if defined(__GNUC__) + return n == 0 ? -1 : 31 ^ __builtin_clz(n); +#else + if (n == 0) + return -1; + int log = 0; + uint32 value = n; + for (int i = 4; i >= 0; --i) { + int shift = (1 << i); + uint32 x = value >> shift; + if (x != 0) { + value = x; + log += shift; + } + } + assert(value == 1); + return log; +#endif +} + +} // namespace woff2 + +/* Compatibility with non-clang compilers. */ +#ifndef __has_builtin +#define __has_builtin(x) 0 +#endif + +#if (__GNUC__ > 2) || (__GNUC__ == 2 && __GNUC_MINOR__ > 95) || \ + (defined(__llvm__) && __has_builtin(__builtin_expect)) +#define PREDICT_FALSE(x) (__builtin_expect(x, 0)) +#define PREDICT_TRUE(x) (__builtin_expect(!!(x), 1)) +#else +#define PREDICT_FALSE(x) (x) +#define PREDICT_TRUE(x) (x) +#endif + +#if (defined(__ARM_ARCH) && (__ARM_ARCH == 7)) || \ + (defined(M_ARM) && (M_ARM == 7)) || \ + defined(__aarch64__) || defined(__ARM64_ARCH_8__) || defined(__i386) || \ + defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64) +#if defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) +#define WOFF_LITTLE_ENDIAN +#elif defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) +#define WOFF_BIG_ENDIAN +#endif /* endianness */ +#endif /* CPU whitelist */ + +#endif // WOFF2_PORT_H_ diff --git a/modules/woff2/src/round.h b/modules/woff2/src/round.h new file mode 100644 index 0000000000..e5b2cb1220 --- /dev/null +++ b/modules/woff2/src/round.h @@ -0,0 +1,27 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Helper for rounding */ + +#ifndef WOFF2_ROUND_H_ +#define WOFF2_ROUND_H_ + +#include <limits> + +namespace woff2 { + +// Round a value up to the nearest multiple of 4. Don't round the value in the +// case that rounding up overflows. +template<typename T> T Round4(T value) { + if (std::numeric_limits<T>::max() - value < 3) { + return value; + } + return (value + 3) & ~3; +} + +} // namespace woff2 + +#endif // WOFF2_ROUND_H_ diff --git a/modules/woff2/src/store_bytes.h b/modules/woff2/src/store_bytes.h new file mode 100644 index 0000000000..fff3c62f84 --- /dev/null +++ b/modules/woff2/src/store_bytes.h @@ -0,0 +1,71 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Helper functions for storing integer values into byte streams. + No bounds checking is performed, that is the responsibility of the caller. */ + +#ifndef WOFF2_STORE_BYTES_H_ +#define WOFF2_STORE_BYTES_H_ + +#include <inttypes.h> +#include <stddef.h> +#include <string.h> + +#include "./port.h" + +namespace woff2 { + +inline 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; +} + +inline size_t Store16(uint8_t* dst, size_t offset, int x) { +#if defined(WOFF_LITTLE_ENDIAN) + *reinterpret_cast<uint16_t*>(dst + offset) = + ((x & 0xFF) << 8) | ((x & 0xFF00) >> 8); +#elif defined(WOFF_BIG_ENDIAN) + *reinterpret_cast<uint16_t*>(dst + offset) = static_cast<uint16_t>(x); +#else + dst[offset] = x >> 8; + dst[offset + 1] = x; +#endif + return offset + 2; +} + +inline void StoreU32(uint32_t val, size_t* offset, uint8_t* dst) { + dst[(*offset)++] = val >> 24; + dst[(*offset)++] = val >> 16; + dst[(*offset)++] = val >> 8; + dst[(*offset)++] = val; +} + +inline void Store16(int val, size_t* offset, uint8_t* dst) { +#if defined(WOFF_LITTLE_ENDIAN) + *reinterpret_cast<uint16_t*>(dst + *offset) = + ((val & 0xFF) << 8) | ((val & 0xFF00) >> 8); + *offset += 2; +#elif defined(WOFF_BIG_ENDIAN) + *reinterpret_cast<uint16_t*>(dst + *offset) = static_cast<uint16_t>(val); + *offset += 2; +#else + dst[(*offset)++] = val >> 8; + dst[(*offset)++] = val; +#endif +} + +inline void StoreBytes(const uint8_t* data, size_t len, + size_t* offset, uint8_t* dst) { + memcpy(&dst[*offset], data, len); + *offset += len; +} + +} // namespace woff2 + +#endif // WOFF2_STORE_BYTES_H_ diff --git a/modules/woff2/src/table_tags.cc b/modules/woff2/src/table_tags.cc new file mode 100644 index 0000000000..dca3ec8d28 --- /dev/null +++ b/modules/woff2/src/table_tags.cc @@ -0,0 +1,82 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Font table tags */ + +#include "./table_tags.h" + +namespace woff2 { + +// 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 uint32_t kKnownTags[63] = { + 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 + TAG('E', 'B', 'S', 'C'), // 29 + TAG('J', 'S', 'T', 'F'), // 30 + TAG('M', 'A', 'T', 'H'), // 31 + TAG('C', 'B', 'D', 'T'), // 32 + TAG('C', 'B', 'L', 'C'), // 33 + TAG('C', 'O', 'L', 'R'), // 34 + TAG('C', 'P', 'A', 'L'), // 35 + TAG('S', 'V', 'G', ' '), // 36 + TAG('s', 'b', 'i', 'x'), // 37 + TAG('a', 'c', 'n', 't'), // 38 + TAG('a', 'v', 'a', 'r'), // 39 + TAG('b', 'd', 'a', 't'), // 40 + TAG('b', 'l', 'o', 'c'), // 41 + TAG('b', 's', 'l', 'n'), // 42 + TAG('c', 'v', 'a', 'r'), // 43 + TAG('f', 'd', 's', 'c'), // 44 + TAG('f', 'e', 'a', 't'), // 45 + TAG('f', 'm', 't', 'x'), // 46 + TAG('f', 'v', 'a', 'r'), // 47 + TAG('g', 'v', 'a', 'r'), // 48 + TAG('h', 's', 't', 'y'), // 49 + TAG('j', 'u', 's', 't'), // 50 + TAG('l', 'c', 'a', 'r'), // 51 + TAG('m', 'o', 'r', 't'), // 52 + TAG('m', 'o', 'r', 'x'), // 53 + TAG('o', 'p', 'b', 'd'), // 54 + TAG('p', 'r', 'o', 'p'), // 55 + TAG('t', 'r', 'a', 'k'), // 56 + TAG('Z', 'a', 'p', 'f'), // 57 + TAG('S', 'i', 'l', 'f'), // 58 + TAG('G', 'l', 'a', 't'), // 59 + TAG('G', 'l', 'o', 'c'), // 60 + TAG('F', 'e', 'a', 't'), // 61 + TAG('S', 'i', 'l', 'l'), // 62 +}; + +} // namespace woff2 diff --git a/modules/woff2/src/table_tags.h b/modules/woff2/src/table_tags.h new file mode 100644 index 0000000000..42dc0ae1c4 --- /dev/null +++ b/modules/woff2/src/table_tags.h @@ -0,0 +1,30 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Font table tags */ + +#ifndef WOFF2_TABLE_TAGS_H_ +#define WOFF2_TABLE_TAGS_H_ + +#include <inttypes.h> + +namespace woff2 { + +// Tags of popular tables. +static const uint32_t kGlyfTableTag = 0x676c7966; +static const uint32_t kHeadTableTag = 0x68656164; +static const uint32_t kLocaTableTag = 0x6c6f6361; +static const uint32_t kDsigTableTag = 0x44534947; +static const uint32_t kCffTableTag = 0x43464620; +static const uint32_t kHmtxTableTag = 0x686d7478; +static const uint32_t kHheaTableTag = 0x68686561; +static const uint32_t kMaxpTableTag = 0x6d617870; + +extern const uint32_t kKnownTags[]; + +} // namespace woff2 + +#endif // WOFF2_TABLE_TAGS_H_ diff --git a/modules/woff2/src/transform.cc b/modules/woff2/src/transform.cc new file mode 100644 index 0000000000..1016efcd2b --- /dev/null +++ b/modules/woff2/src/transform.cc @@ -0,0 +1,430 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Library for preprocessing fonts as part of the WOFF 2.0 conversion. */ + +#include "./transform.h" + +#include <complex> // for std::abs + +#include "./buffer.h" +#include "./font.h" +#include "./glyph.h" +#include "./table_tags.h" +#include "./variable_length.h" + +namespace woff2 { + +namespace { + +const int FLAG_ARG_1_AND_2_ARE_WORDS = 1 << 0; +const int FLAG_WE_HAVE_INSTRUCTIONS = 1 << 8; +const int FLAG_OVERLAP_SIMPLE_BITMAP = 1 << 0; + +void WriteBytes(std::vector<uint8_t>* out, const uint8_t* data, size_t len) { + if (len == 0) return; + size_t offset = out->size(); + out->resize(offset + len); + memcpy(&(*out)[offset], data, len); +} + +void WriteBytes(std::vector<uint8_t>* out, const std::vector<uint8_t>& in) { + for (size_t i = 0; i < in.size(); ++i) { + out->push_back(in[i]); + } +} + +void WriteUShort(std::vector<uint8_t>* out, int value) { + out->push_back(value >> 8); + out->push_back(value & 255); +} + +void WriteLong(std::vector<uint8_t>* out, int value) { + out->push_back((value >> 24) & 255); + out->push_back((value >> 16) & 255); + out->push_back((value >> 8) & 255); + out->push_back(value & 255); +} + +// Glyf table preprocessing, based on +// GlyfEncoder.java +class GlyfEncoder { + public: + explicit GlyfEncoder(int num_glyphs) + : n_glyphs_(num_glyphs) { + bbox_bitmap_.resize(((num_glyphs + 31) >> 5) << 2); + } + + bool Encode(int glyph_id, const Glyph& glyph) { + if (glyph.composite_data_size > 0) { + WriteCompositeGlyph(glyph_id, glyph); + } else if (glyph.contours.size() > 0) { + WriteSimpleGlyph(glyph_id, glyph); + } else { + WriteUShort(&n_contour_stream_, 0); + } + return true; + } + + void GetTransformedGlyfBytes(std::vector<uint8_t>* result) { + WriteUShort(result, 0); // Version + WriteUShort(result, overlap_bitmap_.empty() + ? 0x00 + : FLAG_OVERLAP_SIMPLE_BITMAP); // Flags + WriteUShort(result, n_glyphs_); + WriteUShort(result, 0); // index_format, will be set later + WriteLong(result, n_contour_stream_.size()); + WriteLong(result, n_points_stream_.size()); + WriteLong(result, flag_byte_stream_.size()); + WriteLong(result, glyph_stream_.size()); + WriteLong(result, composite_stream_.size()); + WriteLong(result, bbox_bitmap_.size() + bbox_stream_.size()); + WriteLong(result, instruction_stream_.size()); + WriteBytes(result, n_contour_stream_); + WriteBytes(result, n_points_stream_); + WriteBytes(result, flag_byte_stream_); + WriteBytes(result, glyph_stream_); + WriteBytes(result, composite_stream_); + WriteBytes(result, bbox_bitmap_); + WriteBytes(result, bbox_stream_); + WriteBytes(result, instruction_stream_); + if (!overlap_bitmap_.empty()) { + WriteBytes(result, overlap_bitmap_); + } + } + + private: + void WriteInstructions(const Glyph& glyph) { + Write255UShort(&glyph_stream_, glyph.instructions_size); + WriteBytes(&instruction_stream_, + glyph.instructions_data, glyph.instructions_size); + } + + bool ShouldWriteSimpleGlyphBbox(const Glyph& glyph) { + if (glyph.contours.empty() || glyph.contours[0].empty()) { + return glyph.x_min || glyph.y_min || glyph.x_max || glyph.y_max; + } + + int16_t x_min = glyph.contours[0][0].x; + int16_t y_min = glyph.contours[0][0].y; + int16_t x_max = x_min; + int16_t y_max = y_min; + for (const auto& contour : glyph.contours) { + for (const auto& point : contour) { + if (point.x < x_min) x_min = point.x; + if (point.x > x_max) x_max = point.x; + if (point.y < y_min) y_min = point.y; + if (point.y > y_max) y_max = point.y; + } + } + + if (glyph.x_min != x_min) + return true; + if (glyph.y_min != y_min) + return true; + if (glyph.x_max != x_max) + return true; + if (glyph.y_max != y_max) + return true; + + return false; + } + + void WriteSimpleGlyph(int glyph_id, const Glyph& glyph) { + if (glyph.overlap_simple_flag_set) { + EnsureOverlapBitmap(); + overlap_bitmap_[glyph_id >> 3] |= 0x80 >> (glyph_id & 7); + } + int num_contours = glyph.contours.size(); + WriteUShort(&n_contour_stream_, num_contours); + if (ShouldWriteSimpleGlyphBbox(glyph)) { + WriteBbox(glyph_id, glyph); + } + for (int i = 0; i < num_contours; i++) { + Write255UShort(&n_points_stream_, glyph.contours[i].size()); + } + int lastX = 0; + int lastY = 0; + for (int i = 0; i < num_contours; i++) { + int num_points = glyph.contours[i].size(); + for (int j = 0; j < num_points; j++) { + int x = glyph.contours[i][j].x; + int y = glyph.contours[i][j].y; + int dx = x - lastX; + int dy = y - lastY; + WriteTriplet(glyph.contours[i][j].on_curve, dx, dy); + lastX = x; + lastY = y; + } + } + if (num_contours > 0) { + WriteInstructions(glyph); + } + } + + void WriteCompositeGlyph(int glyph_id, const Glyph& glyph) { + WriteUShort(&n_contour_stream_, -1); + WriteBbox(glyph_id, glyph); + WriteBytes(&composite_stream_, + glyph.composite_data, + glyph.composite_data_size); + if (glyph.have_instructions) { + WriteInstructions(glyph); + } + } + + void WriteBbox(int glyph_id, const Glyph& glyph) { + bbox_bitmap_[glyph_id >> 3] |= 0x80 >> (glyph_id & 7); + WriteUShort(&bbox_stream_, glyph.x_min); + WriteUShort(&bbox_stream_, glyph.y_min); + WriteUShort(&bbox_stream_, glyph.x_max); + WriteUShort(&bbox_stream_, glyph.y_max); + } + + void WriteTriplet(bool on_curve, int x, int y) { + int abs_x = std::abs(x); + int abs_y = std::abs(y); + int on_curve_bit = on_curve ? 0 : 128; + int x_sign_bit = (x < 0) ? 0 : 1; + int y_sign_bit = (y < 0) ? 0 : 1; + int xy_sign_bits = x_sign_bit + 2 * y_sign_bit; + if (x == 0 && abs_y < 1280) { + flag_byte_stream_.push_back(on_curve_bit + + ((abs_y & 0xf00) >> 7) + y_sign_bit); + glyph_stream_.push_back(abs_y & 0xff); + } else if (y == 0 && abs_x < 1280) { + flag_byte_stream_.push_back(on_curve_bit + 10 + + ((abs_x & 0xf00) >> 7) + x_sign_bit); + glyph_stream_.push_back(abs_x & 0xff); + } else if (abs_x < 65 && abs_y < 65) { + flag_byte_stream_.push_back(on_curve_bit + 20 + + ((abs_x - 1) & 0x30) + + (((abs_y - 1) & 0x30) >> 2) + + xy_sign_bits); + glyph_stream_.push_back((((abs_x - 1) & 0xf) << 4) | ((abs_y - 1) & 0xf)); + } else if (abs_x < 769 && abs_y < 769) { + flag_byte_stream_.push_back(on_curve_bit + 84 + + 12 * (((abs_x - 1) & 0x300) >> 8) + + (((abs_y - 1) & 0x300) >> 6) + xy_sign_bits); + glyph_stream_.push_back((abs_x - 1) & 0xff); + glyph_stream_.push_back((abs_y - 1) & 0xff); + } else if (abs_x < 4096 && abs_y < 4096) { + flag_byte_stream_.push_back(on_curve_bit + 120 + xy_sign_bits); + glyph_stream_.push_back(abs_x >> 4); + glyph_stream_.push_back(((abs_x & 0xf) << 4) | (abs_y >> 8)); + glyph_stream_.push_back(abs_y & 0xff); + } else { + flag_byte_stream_.push_back(on_curve_bit + 124 + xy_sign_bits); + glyph_stream_.push_back(abs_x >> 8); + glyph_stream_.push_back(abs_x & 0xff); + glyph_stream_.push_back(abs_y >> 8); + glyph_stream_.push_back(abs_y & 0xff); + } + } + + void EnsureOverlapBitmap() { + if (overlap_bitmap_.empty()) { + overlap_bitmap_.resize((n_glyphs_ + 7) >> 3); + } + } + + std::vector<uint8_t> n_contour_stream_; + std::vector<uint8_t> n_points_stream_; + std::vector<uint8_t> flag_byte_stream_; + std::vector<uint8_t> composite_stream_; + std::vector<uint8_t> bbox_bitmap_; + std::vector<uint8_t> bbox_stream_; + std::vector<uint8_t> glyph_stream_; + std::vector<uint8_t> instruction_stream_; + std::vector<uint8_t> overlap_bitmap_; + int n_glyphs_; +}; + +} // namespace + +bool TransformGlyfAndLocaTables(Font* font) { + // no transform for CFF + const Font::Table* glyf_table = font->FindTable(kGlyfTableTag); + const Font::Table* loca_table = font->FindTable(kLocaTableTag); + + // If you don't have glyf/loca this transform isn't very interesting + if (loca_table == NULL && glyf_table == NULL) { + return true; + } + // It would be best if you didn't have just one of glyf/loca + if ((glyf_table == NULL) != (loca_table == NULL)) { + return FONT_COMPRESSION_FAILURE(); + } + // Must share neither or both loca & glyf + if (loca_table->IsReused() != glyf_table->IsReused()) { + return FONT_COMPRESSION_FAILURE(); + } + if (loca_table->IsReused()) { + return true; + } + + Font::Table* transformed_glyf = &font->tables[kGlyfTableTag ^ 0x80808080]; + Font::Table* transformed_loca = &font->tables[kLocaTableTag ^ 0x80808080]; + + int num_glyphs = NumGlyphs(*font); + GlyfEncoder encoder(num_glyphs); + for (int i = 0; i < num_glyphs; ++i) { + Glyph glyph; + const uint8_t* glyph_data; + size_t glyph_size; + if (!GetGlyphData(*font, i, &glyph_data, &glyph_size) || + (glyph_size > 0 && !ReadGlyph(glyph_data, glyph_size, &glyph))) { + return FONT_COMPRESSION_FAILURE(); + } + encoder.Encode(i, glyph); + } + encoder.GetTransformedGlyfBytes(&transformed_glyf->buffer); + + const Font::Table* head_table = font->FindTable(kHeadTableTag); + if (head_table == NULL || head_table->length < 52) { + return FONT_COMPRESSION_FAILURE(); + } + transformed_glyf->buffer[7] = head_table->data[51]; // index_format + + transformed_glyf->tag = kGlyfTableTag ^ 0x80808080; + transformed_glyf->length = transformed_glyf->buffer.size(); + transformed_glyf->data = transformed_glyf->buffer.data(); + + transformed_loca->tag = kLocaTableTag ^ 0x80808080; + transformed_loca->length = 0; + transformed_loca->data = NULL; + + return true; +} + +// See https://www.microsoft.com/typography/otspec/hmtx.htm +// See WOFF2 spec, 5.4. Transformed hmtx table format +bool TransformHmtxTable(Font* font) { + const Font::Table* glyf_table = font->FindTable(kGlyfTableTag); + const Font::Table* hmtx_table = font->FindTable(kHmtxTableTag); + const Font::Table* hhea_table = font->FindTable(kHheaTableTag); + + // If you don't have hmtx or a glyf not much is going to happen here + if (hmtx_table == NULL || glyf_table == NULL) { + return true; + } + + // hmtx without hhea doesn't make sense + if (hhea_table == NULL) { + return FONT_COMPRESSION_FAILURE(); + } + + // Skip 34 to reach 'hhea' numberOfHMetrics + Buffer hhea_buf(hhea_table->data, hhea_table->length); + uint16_t num_hmetrics; + if (!hhea_buf.Skip(34) || !hhea_buf.ReadU16(&num_hmetrics)) { + return FONT_COMPRESSION_FAILURE(); + } + + // Must have at least one hMetric + if (num_hmetrics < 1) { + return FONT_COMPRESSION_FAILURE(); + } + + int num_glyphs = NumGlyphs(*font); + + // Most fonts can be transformed; assume it's a go until proven otherwise + std::vector<uint16_t> advance_widths; + std::vector<int16_t> proportional_lsbs; + std::vector<int16_t> monospace_lsbs; + + bool remove_proportional_lsb = true; + bool remove_monospace_lsb = (num_glyphs - num_hmetrics) > 0; + + Buffer hmtx_buf(hmtx_table->data, hmtx_table->length); + for (int i = 0; i < num_glyphs; i++) { + Glyph glyph; + const uint8_t* glyph_data; + size_t glyph_size; + if (!GetGlyphData(*font, i, &glyph_data, &glyph_size) || + (glyph_size > 0 && !ReadGlyph(glyph_data, glyph_size, &glyph))) { + return FONT_COMPRESSION_FAILURE(); + } + + uint16_t advance_width = 0; + int16_t lsb = 0; + + if (i < num_hmetrics) { + // [0, num_hmetrics) are proportional hMetrics + if (!hmtx_buf.ReadU16(&advance_width)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (!hmtx_buf.ReadS16(&lsb)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (glyph_size > 0 && glyph.x_min != lsb) { + remove_proportional_lsb = false; + } + + advance_widths.push_back(advance_width); + proportional_lsbs.push_back(lsb); + } else { + // [num_hmetrics, num_glyphs) are monospace leftSideBearing's + if (!hmtx_buf.ReadS16(&lsb)) { + return FONT_COMPRESSION_FAILURE(); + } + if (glyph_size > 0 && glyph.x_min != lsb) { + remove_monospace_lsb = false; + } + monospace_lsbs.push_back(lsb); + } + + // If we know we can't optimize, bail out completely + if (!remove_proportional_lsb && !remove_monospace_lsb) { + return true; + } + } + + Font::Table* transformed_hmtx = &font->tables[kHmtxTableTag ^ 0x80808080]; + + uint8_t flags = 0; + size_t transformed_size = 1 + 2 * advance_widths.size(); + if (remove_proportional_lsb) { + flags |= 1; + } else { + transformed_size += 2 * proportional_lsbs.size(); + } + if (remove_monospace_lsb) { + flags |= 1 << 1; + } else { + transformed_size += 2 * monospace_lsbs.size(); + } + + transformed_hmtx->buffer.reserve(transformed_size); + std::vector<uint8_t>* out = &transformed_hmtx->buffer; + WriteBytes(out, &flags, 1); + for (uint16_t advance_width : advance_widths) { + WriteUShort(out, advance_width); + } + + if (!remove_proportional_lsb) { + for (int16_t lsb : proportional_lsbs) { + WriteUShort(out, lsb); + } + } + if (!remove_monospace_lsb) { + for (int16_t lsb : monospace_lsbs) { + WriteUShort(out, lsb); + } + } + + transformed_hmtx->tag = kHmtxTableTag ^ 0x80808080; + transformed_hmtx->flag_byte = 1 << 6; + transformed_hmtx->length = transformed_hmtx->buffer.size(); + transformed_hmtx->data = transformed_hmtx->buffer.data(); + + + return true; +} + +} // namespace woff2 diff --git a/modules/woff2/src/transform.h b/modules/woff2/src/transform.h new file mode 100644 index 0000000000..a4583d16cb --- /dev/null +++ b/modules/woff2/src/transform.h @@ -0,0 +1,26 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Library for preprocessing fonts as part of the WOFF 2.0 conversion. */ + +#ifndef WOFF2_TRANSFORM_H_ +#define WOFF2_TRANSFORM_H_ + +#include "./font.h" + +namespace woff2 { + +// Adds the transformed versions of the glyf and loca tables to the font. The +// transformed loca table has zero length. The tag of the transformed tables is +// derived from the original tag by flipping the MSBs of every byte. +bool TransformGlyfAndLocaTables(Font* font); + +// Apply transformation to hmtx table if applicable for this font. +bool TransformHmtxTable(Font* font); + +} // namespace woff2 + +#endif // WOFF2_TRANSFORM_H_ diff --git a/modules/woff2/src/variable_length.cc b/modules/woff2/src/variable_length.cc new file mode 100644 index 0000000000..4f348d5e50 --- /dev/null +++ b/modules/woff2/src/variable_length.cc @@ -0,0 +1,129 @@ +/* Copyright 2015 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Helper functions for woff2 variable length types: 255UInt16 and UIntBase128 */ + +#include "./variable_length.h" + +namespace woff2 { + +size_t Size255UShort(uint16_t value) { + size_t result = 3; + if (value < 253) { + result = 1; + } else if (value < 762) { + result = 2; + } else { + result = 3; + } + return result; +} + +void Write255UShort(std::vector<uint8_t>* out, int value) { + if (value < 253) { + out->push_back(value); + } else if (value < 506) { + out->push_back(255); + out->push_back(value - 253); + } else if (value < 762) { + out->push_back(254); + out->push_back(value - 506); + } else { + out->push_back(253); + out->push_back(value >> 8); + out->push_back(value & 0xff); + } +} + +void Store255UShort(int val, size_t* offset, uint8_t* dst) { + std::vector<uint8_t> packed; + Write255UShort(&packed, val); + for (uint8_t packed_byte : packed) { + dst[(*offset)++] = packed_byte; + } +} + +// Based on section 6.1.1 of MicroType Express draft spec +bool Read255UShort(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 FONT_COMPRESSION_FAILURE(); + } + if (code == kWordCode) { + uint16_t result = 0; + if (!buf->ReadU16(&result)) { + return FONT_COMPRESSION_FAILURE(); + } + *value = result; + return true; + } else if (code == kOneMoreByteCode1) { + uint8_t result = 0; + if (!buf->ReadU8(&result)) { + return FONT_COMPRESSION_FAILURE(); + } + *value = result + kLowestUCode; + return true; + } else if (code == kOneMoreByteCode2) { + uint8_t result = 0; + if (!buf->ReadU8(&result)) { + return FONT_COMPRESSION_FAILURE(); + } + *value = result + kLowestUCode * 2; + return true; + } else { + *value = code; + return true; + } +} + +bool ReadBase128(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 FONT_COMPRESSION_FAILURE(); + } + // Leading zeros are invalid. + if (i == 0 && code == 0x80) { + return FONT_COMPRESSION_FAILURE(); + } + // If any of the top seven bits are set then we're about to overflow. + if (result & 0xfe000000) { + return FONT_COMPRESSION_FAILURE(); + } + result = (result << 7) | (code & 0x7f); + if ((code & 0x80) == 0) { + *value = result; + return true; + } + } + // Make sure not to exceed the size bound + return FONT_COMPRESSION_FAILURE(); +} + +size_t Base128Size(size_t n) { + size_t size = 1; + for (; n >= 128; n >>= 7) ++size; + return size; +} + +void StoreBase128(size_t len, size_t* offset, uint8_t* dst) { + size_t size = Base128Size(len); + for (size_t i = 0; i < size; ++i) { + int b = static_cast<int>((len >> (7 * (size - i - 1))) & 0x7f); + if (i < size - 1) { + b |= 0x80; + } + dst[(*offset)++] = b; + } +} + +} // namespace woff2 + diff --git a/modules/woff2/src/variable_length.h b/modules/woff2/src/variable_length.h new file mode 100644 index 0000000000..2080106535 --- /dev/null +++ b/modules/woff2/src/variable_length.h @@ -0,0 +1,30 @@ +/* Copyright 2015 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Helper functions for woff2 variable length types: 255UInt16 and UIntBase128 */ + +#ifndef WOFF2_VARIABLE_LENGTH_H_ +#define WOFF2_VARIABLE_LENGTH_H_ + +#include <inttypes.h> +#include <vector> +#include "./buffer.h" + +namespace woff2 { + +size_t Size255UShort(uint16_t value); +bool Read255UShort(Buffer* buf, unsigned int* value); +void Write255UShort(std::vector<uint8_t>* out, int value); +void Store255UShort(int val, size_t* offset, uint8_t* dst); + +size_t Base128Size(size_t n); +bool ReadBase128(Buffer* buf, uint32_t* value); +void StoreBase128(size_t len, size_t* offset, uint8_t* dst); + +} // namespace woff2 + +#endif // WOFF2_VARIABLE_LENGTH_H_ + diff --git a/modules/woff2/src/woff2_common.cc b/modules/woff2/src/woff2_common.cc new file mode 100644 index 0000000000..fe0a3beda7 --- /dev/null +++ b/modules/woff2/src/woff2_common.cc @@ -0,0 +1,58 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Helpers common across multiple parts of woff2 */ + +#include <algorithm> + +#include "./woff2_common.h" + +#include "./port.h" + +namespace woff2 { + + +uint32_t ComputeULongSum(const uint8_t* buf, size_t size) { + uint32_t checksum = 0; + size_t aligned_size = size & ~3; + for (size_t i = 0; i < aligned_size; i += 4) { +#if defined(WOFF_LITTLE_ENDIAN) + uint32_t v = *reinterpret_cast<const uint32_t*>(buf + i); + checksum += (((v & 0xFF) << 24) | ((v & 0xFF00) << 8) | + ((v & 0xFF0000) >> 8) | ((v & 0xFF000000) >> 24)); +#elif defined(WOFF_BIG_ENDIAN) + checksum += *reinterpret_cast<const uint32_t*>(buf + i); +#else + checksum += (buf[i] << 24) | (buf[i + 1] << 16) | + (buf[i + 2] << 8) | buf[i + 3]; +#endif + } + + // treat size not aligned on 4 as if it were padded to 4 with 0's + if (size != aligned_size) { + uint32_t v = 0; + for (size_t i = aligned_size; i < size; ++i) { + v |= buf[i] << (24 - 8 * (i & 3)); + } + checksum += v; + } + + return checksum; +} + +size_t CollectionHeaderSize(uint32_t header_version, uint32_t num_fonts) { + size_t size = 0; + if (header_version == 0x00020000) { + size += 12; // ulDsig{Tag,Length,Offset} + } + if (header_version == 0x00010000 || header_version == 0x00020000) { + size += 12 // TTCTag, Version, numFonts + + 4 * num_fonts; // OffsetTable[numFonts] + } + return size; +} + +} // namespace woff2 diff --git a/modules/woff2/src/woff2_common.h b/modules/woff2/src/woff2_common.h new file mode 100644 index 0000000000..51fd4a7bf8 --- /dev/null +++ b/modules/woff2/src/woff2_common.h @@ -0,0 +1,64 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Common definition for WOFF2 encoding/decoding */ + +#ifndef WOFF2_WOFF2_COMMON_H_ +#define WOFF2_WOFF2_COMMON_H_ + +#include <stddef.h> +#include <inttypes.h> + +#include <string> + +namespace woff2 { + +static const uint32_t kWoff2Signature = 0x774f4632; // "wOF2" + +// Leave the first byte open to store flag_byte +const unsigned int kWoff2FlagsTransform = 1 << 8; + +// TrueType Collection ID string: 'ttcf' +static const uint32_t kTtcFontFlavor = 0x74746366; + +static const size_t kSfntHeaderSize = 12; +static const size_t kSfntEntrySize = 16; + +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; + const uint8_t* dst_data; + + bool operator<(const Table& other) const { + return tag < other.tag; + } +}; + + +// Size of the collection header. 0 if version indicates this isn't a +// collection. Ref http://www.microsoft.com/typography/otspec/otff.htm, +// True Type Collections +size_t CollectionHeaderSize(uint32_t header_version, uint32_t num_fonts); + +// Compute checksum over size bytes of buf +uint32_t ComputeULongSum(const uint8_t* buf, size_t size); + +} // namespace woff2 + +#endif // WOFF2_WOFF2_COMMON_H_ diff --git a/modules/woff2/src/woff2_compress.cc b/modules/woff2/src/woff2_compress.cc new file mode 100644 index 0000000000..36e8fcafe9 --- /dev/null +++ b/modules/woff2/src/woff2_compress.cc @@ -0,0 +1,43 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* A commandline tool for compressing ttf format files to woff2. */ + +#include <string> + +#include "file.h" +#include <woff2/encode.h> + + +int main(int argc, char **argv) { + if (argc != 2) { + fprintf(stderr, "One argument, the input filename, must be provided.\n"); + return 1; + } + + std::string filename(argv[1]); + std::string outfilename = filename.substr(0, filename.find_last_of(".")) + ".woff2"; + fprintf(stdout, "Processing %s => %s\n", + filename.c_str(), outfilename.c_str()); + std::string input = woff2::GetFileContent(filename); + + const uint8_t* input_data = reinterpret_cast<const uint8_t*>(input.data()); + size_t output_size = woff2::MaxWOFF2CompressedSize(input_data, input.size()); + std::string output(output_size, 0); + uint8_t* output_data = reinterpret_cast<uint8_t*>(&output[0]); + + woff2::WOFF2Params params; + if (!woff2::ConvertTTFToWOFF2(input_data, input.size(), + output_data, &output_size, params)) { + fprintf(stderr, "Compression failed.\n"); + return 1; + } + output.resize(output_size); + + woff2::SetFileContents(outfilename, output.begin(), output.end()); + + return 0; +} diff --git a/modules/woff2/src/woff2_dec.cc b/modules/woff2/src/woff2_dec.cc new file mode 100644 index 0000000000..f1f98cb51a --- /dev/null +++ b/modules/woff2/src/woff2_dec.cc @@ -0,0 +1,1399 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Library for converting WOFF2 format font files to their TTF versions. */ + +#include <woff2/decode.h> + +#include <stdlib.h> +#include <algorithm> +#include <complex> +#include <cstring> +#include <limits> +#include <string> +#include <vector> +#include <map> +#include <memory> +#include <utility> + +#include "./buffer.h" +#include "./port.h" +#include "./round.h" +#include "./store_bytes.h" +#include "./table_tags.h" +#include "./variable_length.h" +#include "./woff2_common.h" + +#include "../RLBoxWOFF2Sandbox.h" + +namespace woff2 { + +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; +const int kOverlapSimple = 1 << 6; + +// composite glyph flags +// See CompositeGlyph.java in sfntly for full definitions +const int FLAG_ARG_1_AND_2_ARE_WORDS = 1 << 0; +const int FLAG_WE_HAVE_A_SCALE = 1 << 3; +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; + +// glyf flags +const int FLAG_OVERLAP_SIMPLE_BITMAP = 1 << 0; + +const size_t kCheckSumAdjustmentOffset = 8; + +const size_t kEndPtsOfContoursOffset = 10; +const size_t kCompositeGlyphBegin = 10; + +// 98% of Google Fonts have no glyph above 5k bytes +// Largest glyph ever observed was 72k bytes +const size_t kDefaultGlyphBuf = 5120; + +// Over 14k test fonts the max compression ratio seen to date was ~20. +// >100 suggests you wrote a bad uncompressed size. +const float kMaxPlausibleCompressionRatio = 100.0; + +// metadata for a TTC font entry +struct TtcFont { + uint32_t flavor; + uint32_t dst_offset; + uint32_t header_checksum; + std::vector<uint16_t> table_indices; +}; + +struct WOFF2Header { + uint32_t flavor; + uint32_t header_version; + uint16_t num_tables; + uint64_t compressed_offset; + uint32_t compressed_length; + uint32_t uncompressed_size; + std::vector<Table> tables; // num_tables unique tables + std::vector<TtcFont> ttc_fonts; // metadata to help rebuild font +}; + +/** + * Accumulates data we may need to reconstruct a single font. One per font + * created for a TTC. + */ +struct WOFF2FontInfo { + uint16_t num_glyphs; + uint16_t index_format; + uint16_t num_hmetrics; + std::vector<int16_t> x_mins; + std::map<uint32_t, uint32_t> table_entry_by_tag; +}; + +// Accumulates metadata as we rebuild the font +struct RebuildMetadata { + uint32_t header_checksum; // set by WriteHeaders + std::vector<WOFF2FontInfo> font_infos; + // checksums for tables that have been written. + // (tag, src_offset) => checksum. Need both because 0-length loca. + std::map<std::pair<uint32_t, uint32_t>, uint32_t> checksums; +}; + +int WithSign(int flag, int baseval) { + // Precondition: 0 <= baseval < 65536 (to avoid integer overflow) + return (flag & 1) ? baseval : -baseval; +} + +bool _SafeIntAddition(int a, int b, int* result) { + if (PREDICT_FALSE( + ((a > 0) && (b > std::numeric_limits<int>::max() - a)) || + ((a < 0) && (b < std::numeric_limits<int>::min() - a)))) { + return false; + } + *result = a + b; + return true; +} + +bool TripletDecode(const uint8_t* flags_in, const uint8_t* in, size_t in_size, + unsigned int n_points, Point* result, size_t* in_bytes_consumed) { + int x = 0; + int y = 0; + + if (PREDICT_FALSE(n_points > in_size)) { + return FONT_COMPRESSION_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 (PREDICT_FALSE(triplet_index + n_data_bytes > in_size || + triplet_index + n_data_bytes < triplet_index)) { + return FONT_COMPRESSION_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; + if (!_SafeIntAddition(x, dx, &x)) { + return false; + } + if (!_SafeIntAddition(y, dy, &y)) { + return false; + } + *result++ = {x, y, 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(unsigned int n_points, const Point* points, + unsigned int n_contours, unsigned int instruction_length, + bool has_overlap_bit, 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 (unsigned int i = 0; i < n_points; ++i) { + const Point& point = points[i]; + int flag = point.on_curve ? kGlyfOnCurve : 0; + if (has_overlap_bit && i == 0) { + flag |= kOverlapSimple; + } + + 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 (PREDICT_FALSE(flag_offset >= dst_size)) { + return FONT_COMPRESSION_FAILURE(); + } + dst[flag_offset++] = repeat_count; + } + if (PREDICT_FALSE(flag_offset >= dst_size)) { + return FONT_COMPRESSION_FAILURE(); + } + dst[flag_offset++] = flag; + repeat_count = 0; + } + last_x = point.x; + last_y = point.y; + last_flag = flag; + } + + if (repeat_count != 0) { + if (PREDICT_FALSE(flag_offset >= dst_size)) { + return FONT_COMPRESSION_FAILURE(); + } + dst[flag_offset++] = repeat_count; + } + unsigned int xy_bytes = x_bytes + y_bytes; + if (PREDICT_FALSE(xy_bytes < x_bytes || + flag_offset + xy_bytes < flag_offset || + flag_offset + xy_bytes > dst_size)) { + return FONT_COMPRESSION_FAILURE(); + } + + int x_offset = flag_offset; + int y_offset = flag_offset + x_bytes; + last_x = 0; + last_y = 0; + for (unsigned int i = 0; i < n_points; ++i) { + int dx = points[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[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. +// dst should point to the beginning of a 'glyf' record. +void ComputeBbox(unsigned int n_points, const Point* points, uint8_t* dst) { + int x_min = 0; + int y_min = 0; + int x_max = 0; + int y_max = 0; + + if (n_points > 0) { + x_min = points[0].x; + x_max = points[0].x; + y_min = points[0].y; + y_max = points[0].y; + } + for (unsigned int i = 1; i < n_points; ++i) { + int x = points[i].x; + int y = points[i].y; + x_min = std::min(x, x_min); + x_max = std::max(x, x_max); + y_min = std::min(y, y_min); + y_max = std::max(y, y_max); + } + 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); +} + + +bool SizeOfComposite(Buffer composite_stream, size_t* 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 (PREDICT_FALSE(!composite_stream.ReadU16(&flags))) { + return FONT_COMPRESSION_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 (PREDICT_FALSE(!composite_stream.Skip(arg_size))) { + return FONT_COMPRESSION_FAILURE(); + } + } + + *size = composite_stream.offset() - start_offset; + *have_instructions = we_have_instructions; + + return true; +} + +bool Pad4(WOFF2Out* out) { + uint8_t zeroes[] = {0, 0, 0}; + if (PREDICT_FALSE(out->Size() + 3 < out->Size())) { + return FONT_COMPRESSION_FAILURE(); + } + uint32_t pad_bytes = Round4(out->Size()) - out->Size(); + if (pad_bytes > 0) { + if (PREDICT_FALSE(!out->Write(&zeroes, pad_bytes))) { + return FONT_COMPRESSION_FAILURE(); + } + } + return true; +} + +// Build TrueType loca table +bool StoreLoca(const std::vector<uint32_t>& loca_values, int index_format, + uint32_t* checksum, WOFF2Out* out) { + // TODO(user) figure out what index format to use based on whether max + // offset fits into uint16_t or not + const uint64_t loca_size = loca_values.size(); + const uint64_t offset_size = index_format ? 4 : 2; + if (PREDICT_FALSE((loca_size << 2) >> 2 != loca_size)) { + return FONT_COMPRESSION_FAILURE(); + } + std::vector<uint8_t> loca_content(loca_size * offset_size); + uint8_t* dst = &loca_content[0]; + size_t offset = 0; + for (size_t i = 0; i < loca_values.size(); ++i) { + uint32_t value = loca_values[i]; + if (index_format) { + offset = StoreU32(dst, offset, value); + } else { + offset = Store16(dst, offset, value >> 1); + } + } + *checksum = ComputeULongSum(&loca_content[0], loca_content.size()); + if (PREDICT_FALSE(!out->Write(&loca_content[0], loca_content.size()))) { + return FONT_COMPRESSION_FAILURE(); + } + return true; +} + +// Reconstruct entire glyf table based on transformed original +bool ReconstructGlyf(const uint8_t* data, Table* glyf_table, + uint32_t* glyf_checksum, Table * loca_table, + uint32_t* loca_checksum, WOFF2FontInfo* info, + WOFF2Out* out) { + static const int kNumSubStreams = 7; + Buffer file(data, glyf_table->transform_length); + uint16_t version; + std::vector<std::pair<const uint8_t*, size_t> > substreams(kNumSubStreams); + const size_t glyf_start = out->Size(); + + if (PREDICT_FALSE(!file.ReadU16(&version))) { + return FONT_COMPRESSION_FAILURE(); + } + + uint16_t flags; + if (PREDICT_FALSE(!file.ReadU16(&flags))) { + return FONT_COMPRESSION_FAILURE(); + } + bool has_overlap_bitmap = (flags & FLAG_OVERLAP_SIMPLE_BITMAP); + + if (PREDICT_FALSE(!file.ReadU16(&info->num_glyphs) || + !file.ReadU16(&info->index_format))) { + return FONT_COMPRESSION_FAILURE(); + } + + // https://dev.w3.org/webfonts/WOFF2/spec/#conform-mustRejectLoca + // dst_length here is origLength in the spec + uint32_t expected_loca_dst_length = (info->index_format ? 4 : 2) + * (static_cast<uint32_t>(info->num_glyphs) + 1); + if (PREDICT_FALSE(loca_table->dst_length != expected_loca_dst_length)) { + return FONT_COMPRESSION_FAILURE(); + } + + unsigned int offset = (2 + kNumSubStreams) * 4; + if (PREDICT_FALSE(offset > glyf_table->transform_length)) { + return FONT_COMPRESSION_FAILURE(); + } + // Invariant from here on: data_size >= offset + for (int i = 0; i < kNumSubStreams; ++i) { + uint32_t substream_size; + if (PREDICT_FALSE(!file.ReadU32(&substream_size))) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(substream_size > glyf_table->transform_length - offset)) { + return FONT_COMPRESSION_FAILURE(); + } + substreams[i] = std::make_pair(data + offset, substream_size); + offset += substream_size; + } + Buffer n_contour_stream(substreams[0].first, substreams[0].second); + Buffer n_points_stream(substreams[1].first, substreams[1].second); + Buffer flag_stream(substreams[2].first, substreams[2].second); + Buffer glyph_stream(substreams[3].first, substreams[3].second); + Buffer composite_stream(substreams[4].first, substreams[4].second); + Buffer bbox_stream(substreams[5].first, substreams[5].second); + Buffer instruction_stream(substreams[6].first, substreams[6].second); + + const uint8_t* overlap_bitmap = nullptr; + unsigned int overlap_bitmap_length = 0; + if (has_overlap_bitmap) { + overlap_bitmap_length = (info->num_glyphs + 7) >> 3; + overlap_bitmap = data + offset; + if (PREDICT_FALSE(overlap_bitmap_length > + glyf_table->transform_length - offset)) { + return FONT_COMPRESSION_FAILURE(); + } + } + + std::vector<uint32_t> loca_values(info->num_glyphs + 1); + std::vector<unsigned int> n_points_vec; + std::unique_ptr<Point[]> points; + size_t points_size = 0; + const uint8_t* bbox_bitmap = bbox_stream.buffer(); + // Safe because num_glyphs is bounded + unsigned int bitmap_length = ((info->num_glyphs + 31) >> 5) << 2; + if (!bbox_stream.Skip(bitmap_length)) { + return FONT_COMPRESSION_FAILURE(); + } + + // Temp buffer for glyph's. + size_t glyph_buf_size = kDefaultGlyphBuf; + std::unique_ptr<uint8_t[]> glyph_buf(new uint8_t[glyph_buf_size]); + + info->x_mins.resize(info->num_glyphs); + for (unsigned int i = 0; i < info->num_glyphs; ++i) { + size_t glyph_size = 0; + uint16_t n_contours = 0; + bool have_bbox = false; + if (bbox_bitmap[i >> 3] & (0x80 >> (i & 7))) { + have_bbox = true; + } + if (PREDICT_FALSE(!n_contour_stream.ReadU16(&n_contours))) { + return FONT_COMPRESSION_FAILURE(); + } + + if (n_contours == 0xffff) { + // composite glyph + bool have_instructions = false; + unsigned int instruction_size = 0; + if (PREDICT_FALSE(!have_bbox)) { + // composite glyphs must have an explicit bbox + return FONT_COMPRESSION_FAILURE(); + } + + size_t composite_size; + if (PREDICT_FALSE(!SizeOfComposite(composite_stream, &composite_size, + &have_instructions))) { + return FONT_COMPRESSION_FAILURE(); + } + if (have_instructions) { + if (PREDICT_FALSE(!Read255UShort(&glyph_stream, &instruction_size))) { + return FONT_COMPRESSION_FAILURE(); + } + } + + size_t size_needed = 12 + composite_size + instruction_size; + if (PREDICT_FALSE(glyph_buf_size < size_needed)) { + glyph_buf.reset(new uint8_t[size_needed]); + glyph_buf_size = size_needed; + } + + glyph_size = Store16(glyph_buf.get(), glyph_size, n_contours); + if (PREDICT_FALSE(!bbox_stream.Read(glyph_buf.get() + glyph_size, 8))) { + return FONT_COMPRESSION_FAILURE(); + } + glyph_size += 8; + + if (PREDICT_FALSE(!composite_stream.Read(glyph_buf.get() + glyph_size, + composite_size))) { + return FONT_COMPRESSION_FAILURE(); + } + glyph_size += composite_size; + if (have_instructions) { + glyph_size = Store16(glyph_buf.get(), glyph_size, instruction_size); + if (PREDICT_FALSE(!instruction_stream.Read(glyph_buf.get() + glyph_size, + instruction_size))) { + return FONT_COMPRESSION_FAILURE(); + } + glyph_size += instruction_size; + } + } else if (n_contours > 0) { + // simple glyph + n_points_vec.clear(); + unsigned int total_n_points = 0; + unsigned int n_points_contour; + for (unsigned int j = 0; j < n_contours; ++j) { + if (PREDICT_FALSE( + !Read255UShort(&n_points_stream, &n_points_contour))) { + return FONT_COMPRESSION_FAILURE(); + } + n_points_vec.push_back(n_points_contour); + if (PREDICT_FALSE(total_n_points + n_points_contour < total_n_points)) { + return FONT_COMPRESSION_FAILURE(); + } + total_n_points += n_points_contour; + } + unsigned int flag_size = total_n_points; + if (PREDICT_FALSE( + flag_size > flag_stream.length() - flag_stream.offset())) { + return FONT_COMPRESSION_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 (points_size < total_n_points) { + points_size = total_n_points; + points.reset(new Point[points_size]); + } + if (PREDICT_FALSE(!TripletDecode(flags_buf, triplet_buf, triplet_size, + total_n_points, points.get(), &triplet_bytes_consumed))) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(!flag_stream.Skip(flag_size))) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(!glyph_stream.Skip(triplet_bytes_consumed))) { + return FONT_COMPRESSION_FAILURE(); + } + unsigned int instruction_size; + if (PREDICT_FALSE(!Read255UShort(&glyph_stream, &instruction_size))) { + return FONT_COMPRESSION_FAILURE(); + } + + if (PREDICT_FALSE(total_n_points >= (1 << 27) + || instruction_size >= (1 << 30))) { + return FONT_COMPRESSION_FAILURE(); + } + size_t size_needed = 12 + 2 * n_contours + 5 * total_n_points + + instruction_size; + if (PREDICT_FALSE(glyph_buf_size < size_needed)) { + glyph_buf.reset(new uint8_t[size_needed]); + glyph_buf_size = size_needed; + } + + glyph_size = Store16(glyph_buf.get(), glyph_size, n_contours); + if (have_bbox) { + if (PREDICT_FALSE(!bbox_stream.Read(glyph_buf.get() + glyph_size, 8))) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + ComputeBbox(total_n_points, points.get(), glyph_buf.get()); + } + glyph_size = kEndPtsOfContoursOffset; + int end_point = -1; + for (unsigned int contour_ix = 0; contour_ix < n_contours; ++contour_ix) { + end_point += n_points_vec[contour_ix]; + if (PREDICT_FALSE(end_point >= 65536)) { + return FONT_COMPRESSION_FAILURE(); + } + glyph_size = Store16(glyph_buf.get(), glyph_size, end_point); + } + + glyph_size = Store16(glyph_buf.get(), glyph_size, instruction_size); + if (PREDICT_FALSE(!instruction_stream.Read(glyph_buf.get() + glyph_size, + instruction_size))) { + return FONT_COMPRESSION_FAILURE(); + } + glyph_size += instruction_size; + + bool has_overlap_bit = + has_overlap_bitmap && overlap_bitmap[i >> 3] & (0x80 >> (i & 7)); + + if (PREDICT_FALSE(!StorePoints( + total_n_points, points.get(), n_contours, instruction_size, + has_overlap_bit, glyph_buf.get(), glyph_buf_size, &glyph_size))) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + // n_contours == 0; empty glyph. Must NOT have a bbox. + if (PREDICT_FALSE(have_bbox)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Empty glyph has a bbox\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + } + + loca_values[i] = out->Size() - glyf_start; + if (PREDICT_FALSE(!out->Write(glyph_buf.get(), glyph_size))) { + return FONT_COMPRESSION_FAILURE(); + } + + // TODO(user) Old code aligned glyphs ... but do we actually need to? + if (PREDICT_FALSE(!Pad4(out))) { + return FONT_COMPRESSION_FAILURE(); + } + + *glyf_checksum += ComputeULongSum(glyph_buf.get(), glyph_size); + + // We may need x_min to reconstruct 'hmtx' + if (n_contours > 0) { + Buffer x_min_buf(glyph_buf.get() + 2, 2); + if (PREDICT_FALSE(!x_min_buf.ReadS16(&info->x_mins[i]))) { + return FONT_COMPRESSION_FAILURE(); + } + } + } + + // glyf_table dst_offset was set by ReconstructFont + glyf_table->dst_length = out->Size() - glyf_table->dst_offset; + loca_table->dst_offset = out->Size(); + // loca[n] will be equal the length of the glyph data ('glyf') table + loca_values[info->num_glyphs] = glyf_table->dst_length; + if (PREDICT_FALSE(!StoreLoca(loca_values, info->index_format, loca_checksum, + out))) { + return FONT_COMPRESSION_FAILURE(); + } + loca_table->dst_length = out->Size() - loca_table->dst_offset; + + return true; +} + +Table* FindTable(std::vector<Table*>* tables, uint32_t tag) { + for (Table* table : *tables) { + if (table->tag == tag) { + return table; + } + } + return NULL; +} + +// Get numberOfHMetrics, https://www.microsoft.com/typography/otspec/hhea.htm +bool ReadNumHMetrics(const uint8_t* data, size_t data_size, + uint16_t* num_hmetrics) { + // Skip 34 to reach 'hhea' numberOfHMetrics + Buffer buffer(data, data_size); + if (PREDICT_FALSE(!buffer.Skip(34) || !buffer.ReadU16(num_hmetrics))) { + return FONT_COMPRESSION_FAILURE(); + } + return true; +} + +// http://dev.w3.org/webfonts/WOFF2/spec/Overview.html#hmtx_table_format +bool ReconstructTransformedHmtx(const uint8_t* transformed_buf, + size_t transformed_size, + uint16_t num_glyphs, + uint16_t num_hmetrics, + const std::vector<int16_t>& x_mins, + uint32_t* checksum, + WOFF2Out* out) { + Buffer hmtx_buff_in(transformed_buf, transformed_size); + + uint8_t hmtx_flags; + if (PREDICT_FALSE(!hmtx_buff_in.ReadU8(&hmtx_flags))) { + return FONT_COMPRESSION_FAILURE(); + } + + std::vector<uint16_t> advance_widths; + std::vector<int16_t> lsbs; + bool has_proportional_lsbs = (hmtx_flags & 1) == 0; + bool has_monospace_lsbs = (hmtx_flags & 2) == 0; + + // Bits 2-7 are reserved and MUST be zero. + if ((hmtx_flags & 0xFC) != 0) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Illegal hmtx flags; bits 2-7 must be 0\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + + // you say you transformed but there is little evidence of it + if (has_proportional_lsbs && has_monospace_lsbs) { + return FONT_COMPRESSION_FAILURE(); + } + + assert(x_mins.size() == num_glyphs); + + // num_glyphs 0 is OK if there is no 'glyf' but cannot then xform 'hmtx'. + if (PREDICT_FALSE(num_hmetrics > num_glyphs)) { + return FONT_COMPRESSION_FAILURE(); + } + + // https://www.microsoft.com/typography/otspec/hmtx.htm + // "...only one entry need be in the array, but that entry is required." + if (PREDICT_FALSE(num_hmetrics < 1)) { + return FONT_COMPRESSION_FAILURE(); + } + + for (uint16_t i = 0; i < num_hmetrics; i++) { + uint16_t advance_width; + if (PREDICT_FALSE(!hmtx_buff_in.ReadU16(&advance_width))) { + return FONT_COMPRESSION_FAILURE(); + } + advance_widths.push_back(advance_width); + } + + for (uint16_t i = 0; i < num_hmetrics; i++) { + int16_t lsb; + if (has_proportional_lsbs) { + if (PREDICT_FALSE(!hmtx_buff_in.ReadS16(&lsb))) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + lsb = x_mins[i]; + } + lsbs.push_back(lsb); + } + + for (uint16_t i = num_hmetrics; i < num_glyphs; i++) { + int16_t lsb; + if (has_monospace_lsbs) { + if (PREDICT_FALSE(!hmtx_buff_in.ReadS16(&lsb))) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + lsb = x_mins[i]; + } + lsbs.push_back(lsb); + } + + // bake me a shiny new hmtx table + uint32_t hmtx_output_size = 2 * num_glyphs + 2 * num_hmetrics; + std::vector<uint8_t> hmtx_table(hmtx_output_size); + uint8_t* dst = &hmtx_table[0]; + size_t dst_offset = 0; + for (uint32_t i = 0; i < num_glyphs; i++) { + if (i < num_hmetrics) { + Store16(advance_widths[i], &dst_offset, dst); + } + Store16(lsbs[i], &dst_offset, dst); + } + + *checksum = ComputeULongSum(&hmtx_table[0], hmtx_output_size); + if (PREDICT_FALSE(!out->Write(&hmtx_table[0], hmtx_output_size))) { + return FONT_COMPRESSION_FAILURE(); + } + + return true; +} + +bool Woff2Uncompress(uint8_t* dst_buf, size_t dst_size, + const uint8_t* src_buf, size_t src_size) { + size_t uncompressed_size = dst_size; + BrotliDecoderResult result = RLBoxBrotliDecoderDecompress( + src_size, src_buf, &uncompressed_size, dst_buf); + if (PREDICT_FALSE(result != BROTLI_DECODER_RESULT_SUCCESS || + uncompressed_size != dst_size)) { + return FONT_COMPRESSION_FAILURE(); + } + return true; +} + +bool ReadTableDirectory(Buffer* file, std::vector<Table>* tables, + size_t num_tables) { + uint32_t src_offset = 0; + for (size_t i = 0; i < num_tables; ++i) { + Table* table = &(*tables)[i]; + uint8_t flag_byte; + if (PREDICT_FALSE(!file->ReadU8(&flag_byte))) { + return FONT_COMPRESSION_FAILURE(); + } + uint32_t tag; + if ((flag_byte & 0x3f) == 0x3f) { + if (PREDICT_FALSE(!file->ReadU32(&tag))) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + tag = kKnownTags[flag_byte & 0x3f]; + } + uint32_t flags = 0; + uint8_t xform_version = (flag_byte >> 6) & 0x03; + + // 0 means xform for glyph/loca, non-0 for others + if (tag == kGlyfTableTag || tag == kLocaTableTag) { + if (xform_version == 0) { + flags |= kWoff2FlagsTransform; + } + } else if (xform_version != 0) { + flags |= kWoff2FlagsTransform; + } + flags |= xform_version; + + uint32_t dst_length; + if (PREDICT_FALSE(!ReadBase128(file, &dst_length))) { + return FONT_COMPRESSION_FAILURE(); + } + uint32_t transform_length = dst_length; + if ((flags & kWoff2FlagsTransform) != 0) { + if (PREDICT_FALSE(!ReadBase128(file, &transform_length))) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(tag == kLocaTableTag && transform_length)) { + return FONT_COMPRESSION_FAILURE(); + } + } + if (PREDICT_FALSE(src_offset + transform_length < src_offset)) { + return FONT_COMPRESSION_FAILURE(); + } + table->src_offset = src_offset; + table->src_length = transform_length; + src_offset += transform_length; + + table->tag = tag; + table->flags = flags; + table->transform_length = transform_length; + table->dst_length = dst_length; + } + return true; +} + +// Writes a single Offset Table entry +size_t StoreOffsetTable(uint8_t* result, size_t offset, uint32_t flavor, + uint16_t num_tables) { + offset = StoreU32(result, offset, flavor); // sfnt version + offset = Store16(result, offset, num_tables); // 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); // searchRange + offset = Store16(result, offset, max_pow2); // entrySelector + // rangeShift + offset = Store16(result, offset, (num_tables << 4) - output_search_range); + return offset; +} + +size_t StoreTableEntry(uint8_t* result, uint32_t offset, uint32_t tag) { + offset = StoreU32(result, offset, tag); + offset = StoreU32(result, offset, 0); + offset = StoreU32(result, offset, 0); + offset = StoreU32(result, offset, 0); + return offset; +} + +// First table goes after all the headers, table directory, etc +uint64_t ComputeOffsetToFirstTable(const WOFF2Header& hdr) { + uint64_t offset = kSfntHeaderSize + + kSfntEntrySize * static_cast<uint64_t>(hdr.num_tables); + if (hdr.header_version) { + offset = CollectionHeaderSize(hdr.header_version, hdr.ttc_fonts.size()) + + kSfntHeaderSize * hdr.ttc_fonts.size(); + for (const auto& ttc_font : hdr.ttc_fonts) { + offset += kSfntEntrySize * ttc_font.table_indices.size(); + } + } + return offset; +} + +std::vector<Table*> Tables(WOFF2Header* hdr, size_t font_index) { + std::vector<Table*> tables; + if (PREDICT_FALSE(hdr->header_version)) { + for (auto index : hdr->ttc_fonts[font_index].table_indices) { + tables.push_back(&hdr->tables[index]); + } + } else { + for (auto& table : hdr->tables) { + tables.push_back(&table); + } + } + return tables; +} + +// Offset tables assumed to have been written in with 0's initially. +// WOFF2Header isn't const so we can use [] instead of at() (which upsets FF) +bool ReconstructFont(uint8_t* transformed_buf, + const uint32_t transformed_buf_size, + RebuildMetadata* metadata, + WOFF2Header* hdr, + size_t font_index, + WOFF2Out* out) { + size_t dest_offset = out->Size(); + uint8_t table_entry[12]; + WOFF2FontInfo* info = &metadata->font_infos[font_index]; + std::vector<Table*> tables = Tables(hdr, font_index); + + // 'glyf' without 'loca' doesn't make sense + const Table* glyf_table = FindTable(&tables, kGlyfTableTag); + const Table* loca_table = FindTable(&tables, kLocaTableTag); + if (PREDICT_FALSE(static_cast<bool>(glyf_table) != + static_cast<bool>(loca_table))) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Cannot have just one of glyf/loca\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + + if (glyf_table != NULL) { + if (PREDICT_FALSE((glyf_table->flags & kWoff2FlagsTransform) + != (loca_table->flags & kWoff2FlagsTransform))) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Cannot transform just one of glyf/loca\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + } + + uint32_t font_checksum = metadata->header_checksum; + if (hdr->header_version) { + font_checksum = hdr->ttc_fonts[font_index].header_checksum; + } + + uint32_t loca_checksum = 0; + for (size_t i = 0; i < tables.size(); i++) { + Table& table = *tables[i]; + + std::pair<uint32_t, uint32_t> checksum_key = {table.tag, table.src_offset}; + bool reused = metadata->checksums.find(checksum_key) + != metadata->checksums.end(); + if (PREDICT_FALSE(font_index == 0 && reused)) { + return FONT_COMPRESSION_FAILURE(); + } + + // TODO(user) a collection with optimized hmtx that reused glyf/loca + // would fail. We don't optimize hmtx for collections yet. + if (PREDICT_FALSE(static_cast<uint64_t>(table.src_offset) + table.src_length + > transformed_buf_size)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (table.tag == kHheaTableTag) { + if (!ReadNumHMetrics(transformed_buf + table.src_offset, + table.src_length, &info->num_hmetrics)) { + return FONT_COMPRESSION_FAILURE(); + } + } + + uint32_t checksum = 0; + if (!reused) { + if ((table.flags & kWoff2FlagsTransform) != kWoff2FlagsTransform) { + if (table.tag == kHeadTableTag) { + if (PREDICT_FALSE(table.src_length < 12)) { + return FONT_COMPRESSION_FAILURE(); + } + // checkSumAdjustment = 0 + StoreU32(transformed_buf + table.src_offset, 8, 0); + } + table.dst_offset = dest_offset; + checksum = ComputeULongSum(transformed_buf + table.src_offset, + table.src_length); + if (PREDICT_FALSE(!out->Write(transformed_buf + table.src_offset, + table.src_length))) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + if (table.tag == kGlyfTableTag) { + table.dst_offset = dest_offset; + + Table* loca_table = FindTable(&tables, kLocaTableTag); + if (PREDICT_FALSE(!ReconstructGlyf(transformed_buf + table.src_offset, + &table, &checksum, loca_table, &loca_checksum, info, out))) { + return FONT_COMPRESSION_FAILURE(); + } + } else if (table.tag == kLocaTableTag) { + // All the work was done by ReconstructGlyf. We already know checksum. + checksum = loca_checksum; + } else if (table.tag == kHmtxTableTag) { + table.dst_offset = dest_offset; + // Tables are sorted so all the info we need has been gathered. + if (PREDICT_FALSE(!ReconstructTransformedHmtx( + transformed_buf + table.src_offset, table.src_length, + info->num_glyphs, info->num_hmetrics, info->x_mins, &checksum, + out))) { + return FONT_COMPRESSION_FAILURE(); + } + } else { + return FONT_COMPRESSION_FAILURE(); // transform unknown + } + } + metadata->checksums[checksum_key] = checksum; + } else { + checksum = metadata->checksums[checksum_key]; + } + font_checksum += checksum; + + // update the table entry with real values. + StoreU32(table_entry, 0, checksum); + StoreU32(table_entry, 4, table.dst_offset); + StoreU32(table_entry, 8, table.dst_length); + if (PREDICT_FALSE(!out->Write(table_entry, + info->table_entry_by_tag[table.tag] + 4, 12))) { + return FONT_COMPRESSION_FAILURE(); + } + + // We replaced 0's. Update overall checksum. + font_checksum += ComputeULongSum(table_entry, 12); + + if (PREDICT_FALSE(!Pad4(out))) { + return FONT_COMPRESSION_FAILURE(); + } + + if (PREDICT_FALSE(static_cast<uint64_t>(table.dst_offset + table.dst_length) + > out->Size())) { + return FONT_COMPRESSION_FAILURE(); + } + dest_offset = out->Size(); + } + + // Update 'head' checkSumAdjustment. We already set it to 0 and summed font. + Table* head_table = FindTable(&tables, kHeadTableTag); + if (head_table) { + if (PREDICT_FALSE(head_table->dst_length < 12)) { + return FONT_COMPRESSION_FAILURE(); + } + uint8_t checksum_adjustment[4]; + StoreU32(checksum_adjustment, 0, 0xB1B0AFBA - font_checksum); + if (PREDICT_FALSE(!out->Write(checksum_adjustment, + head_table->dst_offset + 8, 4))) { + return FONT_COMPRESSION_FAILURE(); + } + } + + return true; +} + +bool ReadWOFF2Header(const uint8_t* data, size_t length, WOFF2Header* hdr) { + Buffer file(data, length); + + uint32_t signature; + if (PREDICT_FALSE(!file.ReadU32(&signature) || signature != kWoff2Signature || + !file.ReadU32(&hdr->flavor))) { + return FONT_COMPRESSION_FAILURE(); + } + + // TODO(user): Should call IsValidVersionTag() here. + + uint32_t reported_length; + if (PREDICT_FALSE( + !file.ReadU32(&reported_length) || length != reported_length)) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(!file.ReadU16(&hdr->num_tables) || !hdr->num_tables)) { + return FONT_COMPRESSION_FAILURE(); + } + + // We don't care about these fields of the header: + // uint16_t reserved + // uint32_t total_sfnt_size, we don't believe this, will compute later + if (PREDICT_FALSE(!file.Skip(6))) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(!file.ReadU32(&hdr->compressed_length))) { + return FONT_COMPRESSION_FAILURE(); + } + // We don't care about these fields of the header: + // uint16_t major_version, minor_version + if (PREDICT_FALSE(!file.Skip(2 * 2))) { + return FONT_COMPRESSION_FAILURE(); + } + uint32_t meta_offset; + uint32_t meta_length; + uint32_t meta_length_orig; + if (PREDICT_FALSE(!file.ReadU32(&meta_offset) || + !file.ReadU32(&meta_length) || + !file.ReadU32(&meta_length_orig))) { + return FONT_COMPRESSION_FAILURE(); + } + if (meta_offset) { + if (PREDICT_FALSE( + meta_offset >= length || length - meta_offset < meta_length)) { + return FONT_COMPRESSION_FAILURE(); + } + } + uint32_t priv_offset; + uint32_t priv_length; + if (PREDICT_FALSE(!file.ReadU32(&priv_offset) || + !file.ReadU32(&priv_length))) { + return FONT_COMPRESSION_FAILURE(); + } + if (priv_offset) { + if (PREDICT_FALSE( + priv_offset >= length || length - priv_offset < priv_length)) { + return FONT_COMPRESSION_FAILURE(); + } + } + hdr->tables.resize(hdr->num_tables); + if (PREDICT_FALSE(!ReadTableDirectory( + &file, &hdr->tables, hdr->num_tables))) { + return FONT_COMPRESSION_FAILURE(); + } + + // Before we sort for output the last table end is the uncompressed size. + Table& last_table = hdr->tables.back(); + hdr->uncompressed_size = last_table.src_offset + last_table.src_length; + if (PREDICT_FALSE(hdr->uncompressed_size < last_table.src_offset)) { + return FONT_COMPRESSION_FAILURE(); + } + + hdr->header_version = 0; + + if (hdr->flavor == kTtcFontFlavor) { + if (PREDICT_FALSE(!file.ReadU32(&hdr->header_version))) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(hdr->header_version != 0x00010000 + && hdr->header_version != 0x00020000)) { + return FONT_COMPRESSION_FAILURE(); + } + uint32_t num_fonts; + if (PREDICT_FALSE(!Read255UShort(&file, &num_fonts) || !num_fonts)) { + return FONT_COMPRESSION_FAILURE(); + } + hdr->ttc_fonts.resize(num_fonts); + + for (uint32_t i = 0; i < num_fonts; i++) { + TtcFont& ttc_font = hdr->ttc_fonts[i]; + uint32_t num_tables; + if (PREDICT_FALSE(!Read255UShort(&file, &num_tables) || !num_tables)) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(!file.ReadU32(&ttc_font.flavor))) { + return FONT_COMPRESSION_FAILURE(); + } + + ttc_font.table_indices.resize(num_tables); + + + unsigned int glyf_idx = 0; + unsigned int loca_idx = 0; + + for (uint32_t j = 0; j < num_tables; j++) { + unsigned int table_idx; + if (PREDICT_FALSE(!Read255UShort(&file, &table_idx)) || + table_idx >= hdr->tables.size()) { + return FONT_COMPRESSION_FAILURE(); + } + ttc_font.table_indices[j] = table_idx; + + const Table& table = hdr->tables[table_idx]; + if (table.tag == kLocaTableTag) { + loca_idx = table_idx; + } + if (table.tag == kGlyfTableTag) { + glyf_idx = table_idx; + } + + } + + // if we have both glyf and loca make sure they are consecutive + // if we have just one we'll reject the font elsewhere + if (glyf_idx > 0 || loca_idx > 0) { + if (PREDICT_FALSE(glyf_idx > loca_idx || loca_idx - glyf_idx != 1)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "TTC font %d has non-consecutive glyf/loca\n", i); +#endif + return FONT_COMPRESSION_FAILURE(); + } + } + } + } + + const uint64_t first_table_offset = ComputeOffsetToFirstTable(*hdr); + + hdr->compressed_offset = file.offset(); + if (PREDICT_FALSE(hdr->compressed_offset > + std::numeric_limits<uint32_t>::max())) { + return FONT_COMPRESSION_FAILURE(); + } + uint64_t src_offset = Round4(hdr->compressed_offset + hdr->compressed_length); + uint64_t dst_offset = first_table_offset; + + + if (PREDICT_FALSE(src_offset > length)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "offset fail; src_offset %" PRIu64 " length %lu " + "dst_offset %" PRIu64 "\n", + src_offset, length, dst_offset); +#endif + return FONT_COMPRESSION_FAILURE(); + } + if (meta_offset) { + if (PREDICT_FALSE(src_offset != meta_offset)) { + return FONT_COMPRESSION_FAILURE(); + } + src_offset = Round4(meta_offset + meta_length); + if (PREDICT_FALSE(src_offset > std::numeric_limits<uint32_t>::max())) { + return FONT_COMPRESSION_FAILURE(); + } + } + + if (priv_offset) { + if (PREDICT_FALSE(src_offset != priv_offset)) { + return FONT_COMPRESSION_FAILURE(); + } + src_offset = Round4(priv_offset + priv_length); + if (PREDICT_FALSE(src_offset > std::numeric_limits<uint32_t>::max())) { + return FONT_COMPRESSION_FAILURE(); + } + } + + if (PREDICT_FALSE(src_offset != Round4(length))) { + return FONT_COMPRESSION_FAILURE(); + } + + return true; +} + +// Write everything before the actual table data +bool WriteHeaders(const uint8_t* data, size_t length, RebuildMetadata* metadata, + WOFF2Header* hdr, WOFF2Out* out) { + std::vector<uint8_t> output(ComputeOffsetToFirstTable(*hdr), 0); + + // Re-order tables in output (OTSpec) order + std::vector<Table> sorted_tables(hdr->tables); + if (hdr->header_version) { + // collection; we have to sort the table offset vector in each font + for (auto& ttc_font : hdr->ttc_fonts) { + std::map<uint32_t, uint16_t> sorted_index_by_tag; + for (auto table_index : ttc_font.table_indices) { + sorted_index_by_tag[hdr->tables[table_index].tag] = table_index; + } + uint16_t index = 0; + for (auto& i : sorted_index_by_tag) { + ttc_font.table_indices[index++] = i.second; + } + } + } else { + // non-collection; we can just sort the tables + std::sort(sorted_tables.begin(), sorted_tables.end()); + } + + // Start building the font + uint8_t* result = &output[0]; + size_t offset = 0; + if (hdr->header_version) { + // TTC header + offset = StoreU32(result, offset, hdr->flavor); // TAG TTCTag + offset = StoreU32(result, offset, hdr->header_version); // FIXED Version + offset = StoreU32(result, offset, hdr->ttc_fonts.size()); // ULONG numFonts + // Space for ULONG OffsetTable[numFonts] (zeroed initially) + size_t offset_table = offset; // keep start of offset table for later + for (size_t i = 0; i < hdr->ttc_fonts.size(); i++) { + offset = StoreU32(result, offset, 0); // will fill real values in later + } + // space for DSIG fields for header v2 + if (hdr->header_version == 0x00020000) { + offset = StoreU32(result, offset, 0); // ULONG ulDsigTag + offset = StoreU32(result, offset, 0); // ULONG ulDsigLength + offset = StoreU32(result, offset, 0); // ULONG ulDsigOffset + } + + // write Offset Tables and store the location of each in TTC Header + metadata->font_infos.resize(hdr->ttc_fonts.size()); + for (size_t i = 0; i < hdr->ttc_fonts.size(); i++) { + TtcFont& ttc_font = hdr->ttc_fonts[i]; + + // write Offset Table location into TTC Header + offset_table = StoreU32(result, offset_table, offset); + + // write the actual offset table so our header doesn't lie + ttc_font.dst_offset = offset; + offset = StoreOffsetTable(result, offset, ttc_font.flavor, + ttc_font.table_indices.size()); + + for (const auto table_index : ttc_font.table_indices) { + uint32_t tag = hdr->tables[table_index].tag; + metadata->font_infos[i].table_entry_by_tag[tag] = offset; + offset = StoreTableEntry(result, offset, tag); + } + + ttc_font.header_checksum = ComputeULongSum(&output[ttc_font.dst_offset], + offset - ttc_font.dst_offset); + } + } else { + metadata->font_infos.resize(1); + offset = StoreOffsetTable(result, offset, hdr->flavor, hdr->num_tables); + for (uint16_t i = 0; i < hdr->num_tables; ++i) { + metadata->font_infos[0].table_entry_by_tag[sorted_tables[i].tag] = offset; + offset = StoreTableEntry(result, offset, sorted_tables[i].tag); + } + } + + if (PREDICT_FALSE(!out->Write(&output[0], output.size()))) { + return FONT_COMPRESSION_FAILURE(); + } + metadata->header_checksum = ComputeULongSum(&output[0], output.size()); + return true; +} + +} // namespace + +size_t ComputeWOFF2FinalSize(const uint8_t* data, size_t length) { + 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) { + WOFF2MemoryOut out(result, result_length); + return ConvertWOFF2ToTTF(data, length, &out); +} + +bool ConvertWOFF2ToTTF(const uint8_t* data, size_t length, + WOFF2Out* out) { + RebuildMetadata metadata; + WOFF2Header hdr; + if (!ReadWOFF2Header(data, length, &hdr)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (!WriteHeaders(data, length, &metadata, &hdr, out)) { + return FONT_COMPRESSION_FAILURE(); + } + + const float compression_ratio = (float) hdr.uncompressed_size / length; + if (compression_ratio > kMaxPlausibleCompressionRatio) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Implausible compression ratio %.01f\n", compression_ratio); +#endif + return FONT_COMPRESSION_FAILURE(); + } + + const uint8_t* src_buf = data + hdr.compressed_offset; + std::vector<uint8_t> uncompressed_buf(hdr.uncompressed_size); + if (PREDICT_FALSE(hdr.uncompressed_size < 1)) { + return FONT_COMPRESSION_FAILURE(); + } + if (PREDICT_FALSE(!Woff2Uncompress(&uncompressed_buf[0], + hdr.uncompressed_size, src_buf, + hdr.compressed_length))) { + return FONT_COMPRESSION_FAILURE(); + } + + for (size_t i = 0; i < metadata.font_infos.size(); i++) { + if (PREDICT_FALSE(!ReconstructFont(&uncompressed_buf[0], + hdr.uncompressed_size, + &metadata, &hdr, i, out))) { + return FONT_COMPRESSION_FAILURE(); + } + } + + return true; +} + +} // namespace woff2 diff --git a/modules/woff2/src/woff2_decompress.cc b/modules/woff2/src/woff2_decompress.cc new file mode 100644 index 0000000000..47394781ad --- /dev/null +++ b/modules/woff2/src/woff2_decompress.cc @@ -0,0 +1,41 @@ +/* Copyright 2013 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* A very simple commandline tool for decompressing woff2 format files to true + type font files. */ + +#include <string> + +#include "./file.h" +#include <woff2/decode.h> + + +int main(int argc, char **argv) { + if (argc != 2) { + fprintf(stderr, "One argument, the input filename, must be provided.\n"); + return 1; + } + + std::string filename(argv[1]); + std::string outfilename = filename.substr(0, filename.find_last_of(".")) + ".ttf"; + + // Note: update woff2_dec_fuzzer_new_entry.cc if this pattern changes. + std::string input = woff2::GetFileContent(filename); + const uint8_t* raw_input = reinterpret_cast<const uint8_t*>(input.data()); + std::string output( + std::min(woff2::ComputeWOFF2FinalSize(raw_input, input.size()), + woff2::kDefaultMaxSize), + 0); + woff2::WOFF2StringOut out(&output); + + const bool ok = woff2::ConvertWOFF2ToTTF(raw_input, input.size(), &out); + + if (ok) { + woff2::SetFileContents(outfilename, output.begin(), + output.begin() + out.Size()); + } + return ok ? 0 : 1; +} diff --git a/modules/woff2/src/woff2_enc.cc b/modules/woff2/src/woff2_enc.cc new file mode 100644 index 0000000000..f3f46e5759 --- /dev/null +++ b/modules/woff2/src/woff2_enc.cc @@ -0,0 +1,458 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Library for converting TTF format font files to their WOFF2 versions. */ + +#include <woff2/encode.h> + +#include <stdlib.h> +#include <complex> +#include <cstring> +#include <limits> +#include <string> +#include <vector> + +#include <brotli/encode.h> +#include "./buffer.h" +#include "./font.h" +#include "./normalize.h" +#include "./round.h" +#include "./store_bytes.h" +#include "./table_tags.h" +#include "./transform.h" +#include "./variable_length.h" +#include "./woff2_common.h" + +namespace woff2 { + + +namespace { + +const size_t kWoff2HeaderSize = 48; +const size_t kWoff2EntrySize = 20; + +bool Compress(const uint8_t* data, const size_t len, uint8_t* result, + uint32_t* result_len, BrotliEncoderMode mode, int quality) { + size_t compressed_len = *result_len; + if (BrotliEncoderCompress(quality, BROTLI_DEFAULT_WINDOW, mode, len, data, + &compressed_len, result) == 0) { + return false; + } + *result_len = compressed_len; + return true; +} + +bool Woff2Compress(const uint8_t* data, const size_t len, + uint8_t* result, uint32_t* result_len, + int quality) { + return Compress(data, len, result, result_len, + BROTLI_MODE_FONT, quality); +} + +bool TextCompress(const uint8_t* data, const size_t len, + uint8_t* result, uint32_t* result_len, + int quality) { + return Compress(data, len, result, result_len, + BROTLI_MODE_TEXT, quality); +} + +int KnownTableIndex(uint32_t tag) { + for (int i = 0; i < 63; ++i) { + if (tag == kKnownTags[i]) return i; + } + return 63; +} + +void StoreTableEntry(const Table& table, size_t* offset, uint8_t* dst) { + uint8_t flag_byte = (table.flags & 0xC0) | KnownTableIndex(table.tag); + dst[(*offset)++] = flag_byte; + // The index here is treated as a set of flag bytes because + // bits 6 and 7 of the byte are reserved for future use as flags. + // 0x3f or 63 means an arbitrary table tag. + if ((flag_byte & 0x3f) == 0x3f) { + StoreU32(table.tag, offset, dst); + } + StoreBase128(table.src_length, offset, dst); + if ((table.flags & kWoff2FlagsTransform) != 0) { + StoreBase128(table.transform_length, offset, dst); + } +} + +size_t TableEntrySize(const Table& table) { + uint8_t flag_byte = KnownTableIndex(table.tag); + size_t size = ((flag_byte & 0x3f) != 0x3f) ? 1 : 5; + size += Base128Size(table.src_length); + if ((table.flags & kWoff2FlagsTransform) != 0) { + size += Base128Size(table.transform_length); + } + return size; +} + +size_t ComputeWoff2Length(const FontCollection& font_collection, + const std::vector<Table>& tables, + std::map<std::pair<uint32_t, uint32_t>, uint16_t> + index_by_tag_offset, + size_t compressed_data_length, + size_t extended_metadata_length) { + size_t size = kWoff2HeaderSize; + + for (const auto& table : tables) { + size += TableEntrySize(table); + } + + // for collections only, collection tables + if (font_collection.flavor == kTtcFontFlavor) { + size += 4; // UInt32 Version of TTC Header + size += Size255UShort(font_collection.fonts.size()); // 255UInt16 numFonts + + size += 4 * font_collection.fonts.size(); // UInt32 flavor for each + + for (const auto& font : font_collection.fonts) { + size += Size255UShort(font.tables.size()); // 255UInt16 numTables + for (const auto& entry : font.tables) { + const Font::Table& table = entry.second; + // no collection entry for xform table + if (table.tag & 0x80808080) continue; + + std::pair<uint32_t, uint32_t> tag_offset(table.tag, table.offset); + uint16_t table_index = index_by_tag_offset[tag_offset]; + size += Size255UShort(table_index); // 255UInt16 index entry + } + } + } + + // compressed data + size += compressed_data_length; + size = Round4(size); + + size += extended_metadata_length; + return size; +} + +size_t ComputeUncompressedLength(const Font& font) { + // sfnt header + offset table + size_t size = 12 + 16 * font.num_tables; + for (const auto& entry : font.tables) { + const Font::Table& table = entry.second; + if (table.tag & 0x80808080) continue; // xform tables don't stay + if (table.IsReused()) continue; // don't have to pay twice + size += Round4(table.length); + } + return size; +} + +size_t ComputeUncompressedLength(const FontCollection& font_collection) { + if (font_collection.flavor != kTtcFontFlavor) { + return ComputeUncompressedLength(font_collection.fonts[0]); + } + size_t size = CollectionHeaderSize(font_collection.header_version, + font_collection.fonts.size()); + for (const auto& font : font_collection.fonts) { + size += ComputeUncompressedLength(font); + } + return size; +} + +size_t ComputeTotalTransformLength(const Font& font) { + size_t total = 0; + for (const auto& i : font.tables) { + const Font::Table& table = i.second; + if (table.IsReused()) { + continue; + } + if (table.tag & 0x80808080 || !font.FindTable(table.tag ^ 0x80808080)) { + // Count transformed tables and non-transformed tables that do not have + // transformed versions. + total += table.length; + } + } + return total; +} + +} // namespace + +size_t MaxWOFF2CompressedSize(const uint8_t* data, size_t length) { + return MaxWOFF2CompressedSize(data, length, ""); +} + +size_t MaxWOFF2CompressedSize(const uint8_t* data, size_t length, + const std::string& extended_metadata) { + // Except for the header size, which is 32 bytes larger in woff2 format, + // all other parts should be smaller (table header in short format, + // transformations and compression). Just to be sure, we will give some + // headroom anyway. + return length + 1024 + extended_metadata.length(); +} + +uint32_t CompressedBufferSize(uint32_t original_size) { + return 1.2 * original_size + 10240; +} + +bool TransformFontCollection(FontCollection* font_collection) { + for (auto& font : font_collection->fonts) { + if (!TransformGlyfAndLocaTables(&font)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "glyf/loca transformation failed.\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + } + + return true; +} + +bool ConvertTTFToWOFF2(const uint8_t *data, size_t length, + uint8_t *result, size_t *result_length) { + WOFF2Params params; + return ConvertTTFToWOFF2(data, length, result, result_length, + params); +} + +bool ConvertTTFToWOFF2(const uint8_t *data, size_t length, + uint8_t *result, size_t *result_length, + const WOFF2Params& params) { + FontCollection font_collection; + if (!ReadFontCollection(data, length, &font_collection)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Parsing of the input font failed.\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + + if (!NormalizeFontCollection(&font_collection)) { + return FONT_COMPRESSION_FAILURE(); + } + + if (params.allow_transforms && !TransformFontCollection(&font_collection)) { + return FONT_COMPRESSION_FAILURE(); + } else { + // glyf/loca use 11 to flag "not transformed" + for (auto& font : font_collection.fonts) { + Font::Table* glyf_table = font.FindTable(kGlyfTableTag); + Font::Table* loca_table = font.FindTable(kLocaTableTag); + if (glyf_table) { + glyf_table->flag_byte |= 0xc0; + } + if (loca_table) { + loca_table->flag_byte |= 0xc0; + } + } + } + + // Although the compressed size of each table in the final woff2 file won't + // be larger than its transform_length, we have to allocate a large enough + // buffer for the compressor, since the compressor can potentially increase + // the size. If the compressor overflows this, it should return false and + // then this function will also return false. + + size_t total_transform_length = 0; + for (const auto& font : font_collection.fonts) { + total_transform_length += ComputeTotalTransformLength(font); + } + size_t compression_buffer_size = CompressedBufferSize(total_transform_length); + std::vector<uint8_t> compression_buf(compression_buffer_size); + uint32_t total_compressed_length = compression_buffer_size; + + // Collect all transformed data into one place in output order. + std::vector<uint8_t> transform_buf(total_transform_length); + size_t transform_offset = 0; + for (const auto& font : font_collection.fonts) { + for (const auto tag : font.OutputOrderedTags()) { + const Font::Table& original = font.tables.at(tag); + if (original.IsReused()) continue; + if (tag & 0x80808080) continue; + const Font::Table* table_to_store = font.FindTable(tag ^ 0x80808080); + if (table_to_store == NULL) table_to_store = &original; + + StoreBytes(table_to_store->data, table_to_store->length, + &transform_offset, &transform_buf[0]); + } + } + + // Compress all transformed data in one stream. + if (!Woff2Compress(transform_buf.data(), total_transform_length, + &compression_buf[0], + &total_compressed_length, + params.brotli_quality)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Compression of combined table failed.\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Compressed %zu to %u.\n", total_transform_length, + total_compressed_length); +#endif + + // Compress the extended metadata + // TODO(user): how does this apply to collections + uint32_t compressed_metadata_buf_length = + CompressedBufferSize(params.extended_metadata.length()); + std::vector<uint8_t> compressed_metadata_buf(compressed_metadata_buf_length); + + if (params.extended_metadata.length() > 0) { + if (!TextCompress((const uint8_t*)params.extended_metadata.data(), + params.extended_metadata.length(), + compressed_metadata_buf.data(), + &compressed_metadata_buf_length, + params.brotli_quality)) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Compression of extended metadata failed.\n"); +#endif + return FONT_COMPRESSION_FAILURE(); + } + } else { + compressed_metadata_buf_length = 0; + } + + std::vector<Table> tables; + std::map<std::pair<uint32_t, uint32_t>, uint16_t> index_by_tag_offset; + + for (const auto& font : font_collection.fonts) { + + for (const auto tag : font.OutputOrderedTags()) { + const Font::Table& src_table = font.tables.at(tag); + if (src_table.IsReused()) { + continue; + } + + std::pair<uint32_t, uint32_t> tag_offset(src_table.tag, src_table.offset); + if (index_by_tag_offset.find(tag_offset) == index_by_tag_offset.end()) { + index_by_tag_offset[tag_offset] = tables.size(); + } else { + return false; + } + + Table table; + table.tag = src_table.tag; + table.flags = src_table.flag_byte; + table.src_length = src_table.length; + table.transform_length = src_table.length; + const uint8_t* transformed_data = src_table.data; + const Font::Table* transformed_table = + font.FindTable(src_table.tag ^ 0x80808080); + if (transformed_table != NULL) { + table.flags = transformed_table->flag_byte; + table.flags |= kWoff2FlagsTransform; + table.transform_length = transformed_table->length; + transformed_data = transformed_table->data; + + } + tables.push_back(table); + } + } + + size_t woff2_length = ComputeWoff2Length(font_collection, tables, + index_by_tag_offset, total_compressed_length, + compressed_metadata_buf_length); + if (woff2_length > *result_length) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Result allocation was too small (%zd vs %zd bytes).\n", + *result_length, woff2_length); +#endif + return FONT_COMPRESSION_FAILURE(); + } + *result_length = woff2_length; + + size_t offset = 0; + + // start of woff2 header (http://www.w3.org/TR/WOFF2/#woff20Header) + StoreU32(kWoff2Signature, &offset, result); + if (font_collection.flavor != kTtcFontFlavor) { + StoreU32(font_collection.fonts[0].flavor, &offset, result); + } else { + StoreU32(kTtcFontFlavor, &offset, result); + } + StoreU32(woff2_length, &offset, result); + Store16(tables.size(), &offset, result); + Store16(0, &offset, result); // reserved + // totalSfntSize + StoreU32(ComputeUncompressedLength(font_collection), &offset, result); + StoreU32(total_compressed_length, &offset, result); // totalCompressedSize + + // Let's just all be v1.0 + Store16(1, &offset, result); // majorVersion + Store16(0, &offset, result); // minorVersion + if (compressed_metadata_buf_length > 0) { + StoreU32(woff2_length - compressed_metadata_buf_length, + &offset, result); // metaOffset + StoreU32(compressed_metadata_buf_length, &offset, result); // metaLength + StoreU32(params.extended_metadata.length(), + &offset, result); // metaOrigLength + } else { + StoreU32(0, &offset, result); // metaOffset + StoreU32(0, &offset, result); // metaLength + StoreU32(0, &offset, result); // metaOrigLength + } + StoreU32(0, &offset, result); // privOffset + StoreU32(0, &offset, result); // privLength + // end of woff2 header + + // table directory (http://www.w3.org/TR/WOFF2/#table_dir_format) + for (const auto& table : tables) { + StoreTableEntry(table, &offset, result); + } + + // for collections only, collection table directory + if (font_collection.flavor == kTtcFontFlavor) { + StoreU32(font_collection.header_version, &offset, result); + Store255UShort(font_collection.fonts.size(), &offset, result); + for (const Font& font : font_collection.fonts) { + + uint16_t num_tables = 0; + for (const auto& entry : font.tables) { + const Font::Table& table = entry.second; + if (table.tag & 0x80808080) continue; // don't write xform tables + num_tables++; + } + Store255UShort(num_tables, &offset, result); + + StoreU32(font.flavor, &offset, result); + for (const auto& entry : font.tables) { + const Font::Table& table = entry.second; + if (table.tag & 0x80808080) continue; // don't write xform tables + + // for reused tables, only the original has an updated offset + uint32_t table_offset = + table.IsReused() ? table.reuse_of->offset : table.offset; + uint32_t table_length = + table.IsReused() ? table.reuse_of->length : table.length; + std::pair<uint32_t, uint32_t> tag_offset(table.tag, table_offset); + if (index_by_tag_offset.find(tag_offset) == index_by_tag_offset.end()) { +#ifdef FONT_COMPRESSION_BIN +fprintf(stderr, "Missing table index for offset 0x%08x\n", + table_offset); +#endif + return FONT_COMPRESSION_FAILURE(); + } + uint16_t index = index_by_tag_offset[tag_offset]; + Store255UShort(index, &offset, result); + + } + + } + } + + // compressed data format (http://www.w3.org/TR/WOFF2/#table_format) + + StoreBytes(&compression_buf[0], total_compressed_length, &offset, result); + offset = Round4(offset); + + StoreBytes(compressed_metadata_buf.data(), compressed_metadata_buf_length, + &offset, result); + + if (*result_length != offset) { +#ifdef FONT_COMPRESSION_BIN + fprintf(stderr, "Mismatch between computed and actual length " + "(%zd vs %zd)\n", *result_length, offset); +#endif + return FONT_COMPRESSION_FAILURE(); + } + return true; +} + +} // namespace woff2 diff --git a/modules/woff2/src/woff2_info.cc b/modules/woff2/src/woff2_info.cc new file mode 100644 index 0000000000..b132304537 --- /dev/null +++ b/modules/woff2/src/woff2_info.cc @@ -0,0 +1,142 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* A commandline tool for dumping info about a woff2 file. */ + +#include <string> + +#include "file.h" +#include "./woff2_common.h" +#include "./buffer.h" +#include "./font.h" +#include "./table_tags.h" +#include "./variable_length.h" + +std::string PrintTag(int tag) { + if (tag & 0x80808080) { + return std::string("_xfm"); // print _xfm for xform tables (else garbage) + } + char printable[] = { + static_cast<char>((tag >> 24) & 0xFF), + static_cast<char>((tag >> 16) & 0xFF), + static_cast<char>((tag >> 8) & 0xFF), + static_cast<char>(tag & 0xFF) + }; + return std::string(printable, 4); +} + +int main(int argc, char **argv) { + if (argc != 2) { + fprintf(stderr, "One argument, the input filename, must be provided.\n"); + return 1; + } + + std::string filename(argv[1]); + std::string outfilename = filename.substr(0, filename.find_last_of(".")) + ".woff2"; + fprintf(stdout, "Processing %s => %s\n", + filename.c_str(), outfilename.c_str()); + std::string input = woff2::GetFileContent(filename); + + woff2::Buffer file(reinterpret_cast<const uint8_t*>(input.data()), + input.size()); + + printf("WOFF2Header\n"); + uint32_t signature, flavor, length, totalSfntSize, totalCompressedSize; + uint32_t metaOffset, metaLength, metaOrigLength, privOffset, privLength; + uint16_t num_tables, reserved, major, minor; + if (!file.ReadU32(&signature)) return 1; + if (!file.ReadU32(&flavor)) return 1; + if (!file.ReadU32(&length)) return 1; + if (!file.ReadU16(&num_tables)) return 1; + if (!file.ReadU16(&reserved)) return 1; + if (!file.ReadU32(&totalSfntSize)) return 1; + if (!file.ReadU32(&totalCompressedSize)) return 1; + if (!file.ReadU16(&major)) return 1; + if (!file.ReadU16(&minor)) return 1; + if (!file.ReadU32(&metaOffset)) return 1; + if (!file.ReadU32(&metaLength)) return 1; + if (!file.ReadU32(&metaOrigLength)) return 1; + if (!file.ReadU32(&privOffset)) return 1; + if (!file.ReadU32(&privLength)) return 1; + + if (signature != 0x774F4632) { + printf("Invalid signature: %08x\n", signature); + return 1; + } + printf("signature 0x%08x\n", signature); + printf("flavor 0x%08x\n", flavor); + printf("length %d\n", length); + printf("numTables %d\n", num_tables); + printf("reserved %d\n", reserved); + printf("totalSfntSize %d\n", totalSfntSize); + printf("totalCompressedSize %d\n", totalCompressedSize); + printf("majorVersion %d\n", major); + printf("minorVersion %d\n", minor); + printf("metaOffset %d\n", metaOffset); + printf("metaLength %d\n", metaLength); + printf("metaOrigLength %d\n", metaOrigLength); + printf("privOffset %d\n", privOffset); + printf("privLength %d\n", privLength); + + std::vector<uint32_t> table_tags; + printf("TableDirectory starts at +%zu\n", file.offset()); + printf("Entry offset flags tag origLength txLength\n"); + for (auto i = 0; i < num_tables; i++) { + size_t offset = file.offset(); + uint8_t flags; + uint32_t tag, origLength, transformLength; + if (!file.ReadU8(&flags)) return 1; + if ((flags & 0x3f) == 0x3f) { + if (!file.ReadU32(&tag)) return 1; + } else { + tag = woff2::kKnownTags[flags & 0x3f]; + } + table_tags.push_back(tag); + if (!ReadBase128(&file, &origLength)) return 1; + + printf("%5d %6zu 0x%02x %s %10d", i, offset, flags, + PrintTag(tag).c_str(), origLength); + + uint8_t xform_version = (flags >> 6) & 0x3; + if (tag == woff2::kGlyfTableTag || tag == woff2::kLocaTableTag) { + if (xform_version == 0) { + if (!ReadBase128(&file, &transformLength)) return 1; + printf(" %8d", transformLength); + } + } else if (xform_version > 0) { + if (!ReadBase128(&file, &transformLength)) return 1; + printf(" %8d", transformLength); + } + printf("\n"); + } + + // Collection header + if (flavor == woff2::kTtcFontFlavor) { + uint32_t version, numFonts; + if (!file.ReadU32(&version)) return 1; + if (!woff2::Read255UShort(&file, &numFonts)) return 1; + printf("CollectionHeader 0x%08x %d fonts\n", version, numFonts); + + for (auto i = 0; i < numFonts; i++) { + uint32_t numTables, flavor; + if (!woff2::Read255UShort(&file, &numTables)) return 1; + if (!file.ReadU32(&flavor)) return 1; + printf("CollectionFontEntry %d flavor 0x%08x %d tables\n", i, flavor, + numTables); + for (auto j = 0; j < numTables; j++) { + uint32_t table_idx; + if (!woff2::Read255UShort(&file, &table_idx)) return 1; + if (table_idx >= table_tags.size()) return 1; + printf(" %d %s (idx %d)\n", j, + PrintTag(table_tags[table_idx]).c_str(), table_idx); + } + } + } + + printf("TableDirectory ends at +%zu\n", file.offset()); + + return 0; +} diff --git a/modules/woff2/src/woff2_out.cc b/modules/woff2/src/woff2_out.cc new file mode 100644 index 0000000000..a22d3bf8be --- /dev/null +++ b/modules/woff2/src/woff2_out.cc @@ -0,0 +1,63 @@ +/* Copyright 2014 Google Inc. All Rights Reserved. + + Distributed under MIT license. + See file LICENSE for detail or copy at https://opensource.org/licenses/MIT +*/ + +/* Output buffer for WOFF2 decompression. */ + +#include <woff2/output.h> + +namespace woff2 { + +WOFF2StringOut::WOFF2StringOut(std::string *buf) + : buf_(buf), max_size_(kDefaultMaxSize), offset_(0) {} + +bool WOFF2StringOut::Write(const void *buf, size_t n) { + return Write(buf, offset_, n); +} + +bool WOFF2StringOut::Write(const void *buf, size_t offset, size_t n) { + if (offset > max_size_ || n > max_size_ - offset) { + return false; + } + if (offset == buf_->size()) { + buf_->append(static_cast<const char*>(buf), n); + } else { + if (offset + n > buf_->size()) { + buf_->append(offset + n - buf_->size(), 0); + } + buf_->replace(offset, n, static_cast<const char*>(buf), n); + } + offset_ = std::max(offset_, offset + n); + + return true; +} + +void WOFF2StringOut::SetMaxSize(size_t max_size) { + max_size_ = max_size; + if (offset_ > max_size_) { + offset_ = max_size_; + } +} + +WOFF2MemoryOut::WOFF2MemoryOut(uint8_t* buf, size_t buf_size) + : buf_(buf), + buf_size_(buf_size), + offset_(0) {} + +bool WOFF2MemoryOut::Write(const void *buf, size_t n) { + return Write(buf, offset_, n); +} + +bool WOFF2MemoryOut::Write(const void *buf, size_t offset, size_t n) { + if (offset > buf_size_ || n > buf_size_ - offset) { + return false; + } + std::memcpy(buf_ + offset, buf, n); + offset_ = std::max(offset_, offset + n); + + return true; +} + +} // namespace woff2 |