diff options
author | Matt A. Tobin <mattatobin@localhost.localdomain> | 2018-02-02 04:16:08 -0500 |
---|---|---|
committer | Matt A. Tobin <mattatobin@localhost.localdomain> | 2018-02-02 04:16:08 -0500 |
commit | 5f8de423f190bbb79a62f804151bc24824fa32d8 (patch) | |
tree | 10027f336435511475e392454359edea8e25895d /gfx/ots/src/cff.cc | |
parent | 49ee0794b5d912db1f95dce6eb52d781dc210db5 (diff) | |
download | UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.gz UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.lz UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.xz UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.zip |
Add m-esr52 at 52.6.0
Diffstat (limited to 'gfx/ots/src/cff.cc')
-rw-r--r-- | gfx/ots/src/cff.cc | 1052 |
1 files changed, 1052 insertions, 0 deletions
diff --git a/gfx/ots/src/cff.cc b/gfx/ots/src/cff.cc new file mode 100644 index 000000000..23b6dadac --- /dev/null +++ b/gfx/ots/src/cff.cc @@ -0,0 +1,1052 @@ +// Copyright (c) 2012 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "cff.h" + +#include <cstring> +#include <utility> +#include <vector> + +#include "maxp.h" +#include "cff_type2_charstring.h" + +// CFF - PostScript font program (Compact Font Format) table +// http://www.microsoft.com/typography/otspec/cff.htm +// http://www.microsoft.com/typography/otspec/cffspec.htm + +#define TABLE_NAME "CFF" + +namespace { + +enum DICT_OPERAND_TYPE { + DICT_OPERAND_INTEGER, + DICT_OPERAND_REAL, + DICT_OPERATOR, +}; + +enum DICT_DATA_TYPE { + DICT_DATA_TOPLEVEL, + DICT_DATA_FDARRAY, +}; + +enum FONT_FORMAT { + FORMAT_UNKNOWN, + FORMAT_CID_KEYED, + FORMAT_OTHER, // Including synthetic fonts +}; + +// see Appendix. A +const size_t kNStdString = 390; + +bool ReadOffset(ots::Buffer *table, uint8_t off_size, uint32_t *offset) { + if (off_size > 4) { + return OTS_FAILURE(); + } + + uint32_t tmp32 = 0; + for (unsigned i = 0; i < off_size; ++i) { + uint8_t tmp8 = 0; + if (!table->ReadU8(&tmp8)) { + return OTS_FAILURE(); + } + tmp32 <<= 8; + tmp32 += tmp8; + } + *offset = tmp32; + return true; +} + +bool ParseIndex(ots::Buffer *table, ots::CFFIndex *index) { + index->off_size = 0; + index->offsets.clear(); + + if (!table->ReadU16(&(index->count))) { + return OTS_FAILURE(); + } + if (index->count == 0) { + // An empty INDEX. + index->offset_to_next = table->offset(); + return true; + } + + if (!table->ReadU8(&(index->off_size))) { + return OTS_FAILURE(); + } + if ((index->off_size == 0) || + (index->off_size > 4)) { + return OTS_FAILURE(); + } + + const size_t array_size = (index->count + 1) * index->off_size; + // less than ((64k + 1) * 4), thus does not overflow. + const size_t object_data_offset = table->offset() + array_size; + // does not overflow too, since offset() <= 1GB. + + if (object_data_offset >= table->length()) { + return OTS_FAILURE(); + } + + for (unsigned i = 0; i <= index->count; ++i) { // '<=' is not a typo. + uint32_t rel_offset = 0; + if (!ReadOffset(table, index->off_size, &rel_offset)) { + return OTS_FAILURE(); + } + if (rel_offset < 1) { + return OTS_FAILURE(); + } + if (i == 0 && rel_offset != 1) { + return OTS_FAILURE(); + } + + if (rel_offset > table->length()) { + return OTS_FAILURE(); + } + + // does not underflow. + if (object_data_offset > table->length() - (rel_offset - 1)) { + return OTS_FAILURE(); + } + + index->offsets.push_back( + object_data_offset + (rel_offset - 1)); // less than length(), 1GB. + } + + for (unsigned i = 1; i < index->offsets.size(); ++i) { + // We allow consecutive identical offsets here for zero-length strings. + // See http://crbug.com/69341 for more details. + if (index->offsets[i] < index->offsets[i - 1]) { + return OTS_FAILURE(); + } + } + + index->offset_to_next = index->offsets.back(); + return true; +} + +bool ParseNameData( + ots::Buffer *table, const ots::CFFIndex &index, std::string* out_name) { + uint8_t name[256] = {0}; + if (index.offsets.size() == 0) { // just in case. + return OTS_FAILURE(); + } + for (unsigned i = 1; i < index.offsets.size(); ++i) { + const size_t length = index.offsets[i] - index.offsets[i - 1]; + // font names should be no longer than 127 characters. + if (length > 127) { + return OTS_FAILURE(); + } + + table->set_offset(index.offsets[i - 1]); + if (!table->Read(name, length)) { + return OTS_FAILURE(); + } + + for (size_t j = 0; j < length; ++j) { + // setting the first byte to NUL is allowed. + if (j == 0 && name[j] == 0) continue; + // non-ASCII characters are not recommended (except the first character). + if (name[j] < 33 || name[j] > 126) { + return OTS_FAILURE(); + } + // [, ], ... are not allowed. + if (std::strchr("[](){}<>/% ", name[j])) { + return OTS_FAILURE(); + } + } + } + + *out_name = reinterpret_cast<char *>(name); + return true; +} + +bool CheckOffset(const std::pair<uint32_t, DICT_OPERAND_TYPE>& operand, + size_t table_length) { + if (operand.second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + if (operand.first >= table_length) { + return OTS_FAILURE(); + } + return true; +} + +bool CheckSid(const std::pair<uint32_t, DICT_OPERAND_TYPE>& operand, + size_t sid_max) { + if (operand.second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + if (operand.first > sid_max) { + return OTS_FAILURE(); + } + return true; +} + +bool ParseDictDataBcd( + ots::Buffer *table, + std::vector<std::pair<uint32_t, DICT_OPERAND_TYPE> > *operands) { + bool read_decimal_point = false; + bool read_e = false; + + uint8_t nibble = 0; + size_t count = 0; + while (true) { + if (!table->ReadU8(&nibble)) { + return OTS_FAILURE(); + } + if ((nibble & 0xf0) == 0xf0) { + if ((nibble & 0xf) == 0xf) { + // TODO(yusukes): would be better to store actual double value, + // rather than the dummy integer. + operands->push_back(std::make_pair(static_cast<uint32_t>(0), + DICT_OPERAND_REAL)); + return true; + } + return OTS_FAILURE(); + } + if ((nibble & 0x0f) == 0x0f) { + operands->push_back(std::make_pair(static_cast<uint32_t>(0), + DICT_OPERAND_REAL)); + return true; + } + + // check number format + uint8_t nibbles[2]; + nibbles[0] = (nibble & 0xf0) >> 8; + nibbles[1] = (nibble & 0x0f); + for (unsigned i = 0; i < 2; ++i) { + if (nibbles[i] == 0xd) { // reserved number + return OTS_FAILURE(); + } + if ((nibbles[i] == 0xe) && // minus + ((count > 0) || (i > 0))) { + return OTS_FAILURE(); // minus sign should be the first character. + } + if (nibbles[i] == 0xa) { // decimal point + if (!read_decimal_point) { + read_decimal_point = true; + } else { + return OTS_FAILURE(); // two or more points. + } + } + if ((nibbles[i] == 0xb) || // E+ + (nibbles[i] == 0xc)) { // E- + if (!read_e) { + read_e = true; + } else { + return OTS_FAILURE(); // two or more E's. + } + } + } + ++count; + } +} + +bool ParseDictDataEscapedOperator( + ots::Buffer *table, + std::vector<std::pair<uint32_t, DICT_OPERAND_TYPE> > *operands) { + uint8_t op = 0; + if (!table->ReadU8(&op)) { + return OTS_FAILURE(); + } + + if ((op <= 14) || + (op >= 17 && op <= 23) || + (op >= 30 && op <= 38)) { + operands->push_back(std::make_pair((12U << 8) + op, DICT_OPERATOR)); + return true; + } + + // reserved area. + return OTS_FAILURE(); +} + +bool ParseDictDataNumber( + ots::Buffer *table, uint8_t b0, + std::vector<std::pair<uint32_t, DICT_OPERAND_TYPE> > *operands) { + uint8_t b1 = 0; + uint8_t b2 = 0; + uint8_t b3 = 0; + uint8_t b4 = 0; + + switch (b0) { + case 28: // shortint + if (!table->ReadU8(&b1) || + !table->ReadU8(&b2)) { + return OTS_FAILURE(); + } + operands->push_back(std::make_pair( + static_cast<uint32_t>((b1 << 8) + b2), DICT_OPERAND_INTEGER)); + return true; + + case 29: // longint + if (!table->ReadU8(&b1) || + !table->ReadU8(&b2) || + !table->ReadU8(&b3) || + !table->ReadU8(&b4)) { + return OTS_FAILURE(); + } + operands->push_back(std::make_pair( + static_cast<uint32_t>((b1 << 24) + (b2 << 16) + (b3 << 8) + b4), + DICT_OPERAND_INTEGER)); + return true; + + case 30: // binary coded decimal + return ParseDictDataBcd(table, operands); + + default: + break; + } + + uint32_t result; + if (b0 >=32 && b0 <=246) { + result = b0 - 139; + } else if (b0 >=247 && b0 <= 250) { + if (!table->ReadU8(&b1)) { + return OTS_FAILURE(); + } + result = (b0 - 247) * 256 + b1 + 108; + } else if (b0 >= 251 && b0 <= 254) { + if (!table->ReadU8(&b1)) { + return OTS_FAILURE(); + } + result = -(b0 - 251) * 256 + b1 - 108; + } else { + return OTS_FAILURE(); + } + + operands->push_back(std::make_pair(result, DICT_OPERAND_INTEGER)); + return true; +} + +bool ParseDictDataReadNext( + ots::Buffer *table, + std::vector<std::pair<uint32_t, DICT_OPERAND_TYPE> > *operands) { + uint8_t op = 0; + if (!table->ReadU8(&op)) { + return OTS_FAILURE(); + } + if (op <= 21) { + if (op == 12) { + return ParseDictDataEscapedOperator(table, operands); + } + operands->push_back(std::make_pair( + static_cast<uint32_t>(op), DICT_OPERATOR)); + return true; + } else if (op <= 27 || op == 31 || op == 255) { + // reserved area. + return OTS_FAILURE(); + } + + return ParseDictDataNumber(table, op, operands); +} + +bool ParsePrivateDictData( + const uint8_t *data, + size_t table_length, size_t offset, size_t dict_length, + DICT_DATA_TYPE type, ots::OpenTypeCFF *out_cff) { + ots::Buffer table(data + offset, dict_length); + std::vector<std::pair<uint32_t, DICT_OPERAND_TYPE> > operands; + + // Since a Private DICT for FDArray might not have a Local Subr (e.g. Hiragino + // Kaku Gothic Std W8), we create an empty Local Subr here to match the size + // of FDArray the size of |local_subrs_per_font|. + if (type == DICT_DATA_FDARRAY) { + out_cff->local_subrs_per_font.push_back(new ots::CFFIndex); + } + + while (table.offset() < dict_length) { + if (!ParseDictDataReadNext(&table, &operands)) { + return OTS_FAILURE(); + } + if (operands.empty()) { + return OTS_FAILURE(); + } + if (operands.size() > 48) { + // An operator may be preceded by up to a maximum of 48 operands. + return OTS_FAILURE(); + } + if (operands.back().second != DICT_OPERATOR) { + continue; + } + + // got operator + const uint32_t op = operands.back().first; + operands.pop_back(); + + switch (op) { + // hints + case 6: // BlueValues + case 7: // OtherBlues + case 8: // FamilyBlues + case 9: // FamilyOtherBlues + if (operands.empty() || (operands.size() % 2) != 0) { + return OTS_FAILURE(); + } + break; + + // array + case (12U << 8) + 12: // StemSnapH (delta) + case (12U << 8) + 13: // StemSnapV (delta) + if (operands.empty()) { + return OTS_FAILURE(); + } + break; + + // number + case 10: // StdHW + case 11: // StdVW + case 20: // defaultWidthX + case 21: // nominalWidthX + case (12U << 8) + 9: // BlueScale + case (12U << 8) + 10: // BlueShift + case (12U << 8) + 11: // BlueFuzz + case (12U << 8) + 17: // LanguageGroup + case (12U << 8) + 18: // ExpansionFactor + case (12U << 8) + 19: // initialRandomSeed + if (operands.size() != 1) { + return OTS_FAILURE(); + } + break; + + // Local Subrs INDEX, offset(self) + case 19: { + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (operands.back().second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + if (operands.back().first >= 1024 * 1024 * 1024) { + return OTS_FAILURE(); + } + if (operands.back().first + offset >= table_length) { + return OTS_FAILURE(); + } + // parse "16. Local Subrs INDEX" + ots::Buffer cff_table(data, table_length); + cff_table.set_offset(operands.back().first + offset); + ots::CFFIndex *local_subrs_index = NULL; + if (type == DICT_DATA_FDARRAY) { + if (out_cff->local_subrs_per_font.empty()) { + return OTS_FAILURE(); // not reached. + } + local_subrs_index = out_cff->local_subrs_per_font.back(); + } else { // type == DICT_DATA_TOPLEVEL + if (out_cff->local_subrs) { + return OTS_FAILURE(); // two or more local_subrs? + } + local_subrs_index = new ots::CFFIndex; + out_cff->local_subrs = local_subrs_index; + } + if (!ParseIndex(&cff_table, local_subrs_index)) { + return OTS_FAILURE(); + } + break; + } + + // boolean + case (12U << 8) + 14: // ForceBold + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (operands.back().second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + if (operands.back().first >= 2) { + return OTS_FAILURE(); + } + break; + + default: + return OTS_FAILURE(); + } + operands.clear(); + } + + return true; +} + +bool ParseDictData(const uint8_t *data, size_t table_length, + const ots::CFFIndex &index, uint16_t glyphs, + size_t sid_max, DICT_DATA_TYPE type, + ots::OpenTypeCFF *out_cff) { + for (unsigned i = 1; i < index.offsets.size(); ++i) { + if (type == DICT_DATA_TOPLEVEL) { + out_cff->char_strings_array.push_back(new ots::CFFIndex); + } + size_t dict_length = index.offsets[i] - index.offsets[i - 1]; + ots::Buffer table(data + index.offsets[i - 1], dict_length); + + std::vector<std::pair<uint32_t, DICT_OPERAND_TYPE> > operands; + + FONT_FORMAT font_format = FORMAT_UNKNOWN; + bool have_ros = false; + uint16_t charstring_glyphs = 0; + size_t charset_offset = 0; + + while (table.offset() < dict_length) { + if (!ParseDictDataReadNext(&table, &operands)) { + return OTS_FAILURE(); + } + if (operands.empty()) { + return OTS_FAILURE(); + } + if (operands.size() > 48) { + // An operator may be preceded by up to a maximum of 48 operands. + return OTS_FAILURE(); + } + if (operands.back().second != DICT_OPERATOR) continue; + + // got operator + const uint32_t op = operands.back().first; + operands.pop_back(); + + switch (op) { + // SID + case 0: // version + case 1: // Notice + case 2: // Copyright + case 3: // FullName + case 4: // FamilyName + case (12U << 8) + 0: // Copyright + case (12U << 8) + 21: // PostScript + case (12U << 8) + 22: // BaseFontName + case (12U << 8) + 38: // FontName + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (!CheckSid(operands.back(), sid_max)) { + return OTS_FAILURE(); + } + break; + + // array + case 5: // FontBBox + case 14: // XUID + case (12U << 8) + 7: // FontMatrix + case (12U << 8) + 23: // BaseFontBlend (delta) + if (operands.empty()) { + return OTS_FAILURE(); + } + break; + + // number + case 13: // UniqueID + case (12U << 8) + 2: // ItalicAngle + case (12U << 8) + 3: // UnderlinePosition + case (12U << 8) + 4: // UnderlineThickness + case (12U << 8) + 5: // PaintType + case (12U << 8) + 8: // StrokeWidth + case (12U << 8) + 20: // SyntheticBase + if (operands.size() != 1) { + return OTS_FAILURE(); + } + break; + case (12U << 8) + 31: // CIDFontVersion + case (12U << 8) + 32: // CIDFontRevision + case (12U << 8) + 33: // CIDFontType + case (12U << 8) + 34: // CIDCount + case (12U << 8) + 35: // UIDBase + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (font_format != FORMAT_CID_KEYED) { + return OTS_FAILURE(); + } + break; + case (12U << 8) + 6: // CharstringType + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if(operands.back().second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + if (operands.back().first != 2) { + // We only support the "Type 2 Charstring Format." + // TODO(yusukes): Support Type 1 format? Is that still in use? + return OTS_FAILURE(); + } + break; + + // boolean + case (12U << 8) + 1: // isFixedPitch + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (operands.back().second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + if (operands.back().first >= 2) { + return OTS_FAILURE(); + } + break; + + // offset(0) + case 15: // charset + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (operands.back().first <= 2) { + // predefined charset, ISOAdobe, Expert or ExpertSubset, is used. + break; + } + if (!CheckOffset(operands.back(), table_length)) { + return OTS_FAILURE(); + } + if (charset_offset) { + return OTS_FAILURE(); // multiple charset tables? + } + charset_offset = operands.back().first; + break; + + case 16: { // Encoding + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (operands.back().first <= 1) { + break; // predefined encoding, "Standard" or "Expert", is used. + } + if (!CheckOffset(operands.back(), table_length)) { + return OTS_FAILURE(); + } + + // parse sub dictionary INDEX. + ots::Buffer cff_table(data, table_length); + cff_table.set_offset(operands.back().first); + uint8_t format = 0; + if (!cff_table.ReadU8(&format)) { + return OTS_FAILURE(); + } + if (format & 0x80) { + // supplemental encoding is not supported at the moment. + return OTS_FAILURE(); + } + // TODO(yusukes): support & parse supplemental encoding tables. + break; + } + + case 17: { // CharStrings + if (type != DICT_DATA_TOPLEVEL) { + return OTS_FAILURE(); + } + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (!CheckOffset(operands.back(), table_length)) { + return OTS_FAILURE(); + } + // parse "14. CharStrings INDEX" + ots::Buffer cff_table(data, table_length); + cff_table.set_offset(operands.back().first); + ots::CFFIndex *charstring_index = out_cff->char_strings_array.back(); + if (!ParseIndex(&cff_table, charstring_index)) { + return OTS_FAILURE(); + } + if (charstring_index->count < 2) { + return OTS_FAILURE(); + } + if (charstring_glyphs) { + return OTS_FAILURE(); // multiple charstring tables? + } + charstring_glyphs = charstring_index->count; + if (charstring_glyphs != glyphs) { + return OTS_FAILURE(); // CFF and maxp have different number of glyphs? + } + break; + } + + case (12U << 8) + 36: { // FDArray + if (type != DICT_DATA_TOPLEVEL) { + return OTS_FAILURE(); + } + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (!CheckOffset(operands.back(), table_length)) { + return OTS_FAILURE(); + } + + // parse sub dictionary INDEX. + ots::Buffer cff_table(data, table_length); + cff_table.set_offset(operands.back().first); + ots::CFFIndex sub_dict_index; + if (!ParseIndex(&cff_table, &sub_dict_index)) { + return OTS_FAILURE(); + } + if (!ParseDictData(data, table_length, + sub_dict_index, + glyphs, sid_max, DICT_DATA_FDARRAY, + out_cff)) { + return OTS_FAILURE(); + } + if (out_cff->font_dict_length != 0) { + return OTS_FAILURE(); // two or more FDArray found. + } + out_cff->font_dict_length = sub_dict_index.count; + break; + } + + case (12U << 8) + 37: { // FDSelect + if (type != DICT_DATA_TOPLEVEL) { + return OTS_FAILURE(); + } + if (operands.size() != 1) { + return OTS_FAILURE(); + } + if (!CheckOffset(operands.back(), table_length)) { + return OTS_FAILURE(); + } + + // parse FDSelect data structure + ots::Buffer cff_table(data, table_length); + cff_table.set_offset(operands.back().first); + uint8_t format = 0; + if (!cff_table.ReadU8(&format)) { + return OTS_FAILURE(); + } + if (format == 0) { + for (uint16_t j = 0; j < glyphs; ++j) { + uint8_t fd_index = 0; + if (!cff_table.ReadU8(&fd_index)) { + return OTS_FAILURE(); + } + (out_cff->fd_select)[j] = fd_index; + } + } else if (format == 3) { + uint16_t n_ranges = 0; + if (!cff_table.ReadU16(&n_ranges)) { + return OTS_FAILURE(); + } + if (n_ranges == 0) { + return OTS_FAILURE(); + } + + uint16_t last_gid = 0; + uint8_t fd_index = 0; + for (unsigned j = 0; j < n_ranges; ++j) { + uint16_t first = 0; // GID + if (!cff_table.ReadU16(&first)) { + return OTS_FAILURE(); + } + + // Sanity checks. + if ((j == 0) && (first != 0)) { + return OTS_FAILURE(); + } + if ((j != 0) && (last_gid >= first)) { + return OTS_FAILURE(); // not increasing order. + } + + // Copy the mapping to |out_cff->fd_select|. + if (j != 0) { + for (uint16_t k = last_gid; k < first; ++k) { + if (!out_cff->fd_select.insert( + std::make_pair(k, fd_index)).second) { + return OTS_FAILURE(); + } + } + } + + if (!cff_table.ReadU8(&fd_index)) { + return OTS_FAILURE(); + } + last_gid = first; + // TODO(yusukes): check GID? + } + uint16_t sentinel = 0; + if (!cff_table.ReadU16(&sentinel)) { + return OTS_FAILURE(); + } + if (last_gid >= sentinel) { + return OTS_FAILURE(); + } + for (uint16_t k = last_gid; k < sentinel; ++k) { + if (!out_cff->fd_select.insert( + std::make_pair(k, fd_index)).second) { + return OTS_FAILURE(); + } + } + } else { + // unknown format + return OTS_FAILURE(); + } + break; + } + + // Private DICT (2 * number) + case 18: { + if (operands.size() != 2) { + return OTS_FAILURE(); + } + if (operands.back().second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + const uint32_t private_offset = operands.back().first; + operands.pop_back(); + if (operands.back().second != DICT_OPERAND_INTEGER) { + return OTS_FAILURE(); + } + const uint32_t private_length = operands.back().first; + if (private_offset > table_length) { + return OTS_FAILURE(); + } + if (private_length >= table_length) { + return OTS_FAILURE(); + } + if (private_length + private_offset > table_length) { + return OTS_FAILURE(); + } + // parse "15. Private DICT Data" + if (!ParsePrivateDictData(data, table_length, + private_offset, private_length, + type, out_cff)) { + return OTS_FAILURE(); + } + break; + } + + // ROS + case (12U << 8) + 30: + if (font_format != FORMAT_UNKNOWN) { + return OTS_FAILURE(); + } + font_format = FORMAT_CID_KEYED; + if (operands.size() != 3) { + return OTS_FAILURE(); + } + // check SIDs + operands.pop_back(); // ignore the first number. + if (!CheckSid(operands.back(), sid_max)) { + return OTS_FAILURE(); + } + operands.pop_back(); + if (!CheckSid(operands.back(), sid_max)) { + return OTS_FAILURE(); + } + if (have_ros) { + return OTS_FAILURE(); // multiple ROS tables? + } + have_ros = true; + break; + + default: + return OTS_FAILURE(); + } + operands.clear(); + + if (font_format == FORMAT_UNKNOWN) { + font_format = FORMAT_OTHER; + } + } + + // parse "13. Charsets" + if (charset_offset) { + ots::Buffer cff_table(data, table_length); + cff_table.set_offset(charset_offset); + uint8_t format = 0; + if (!cff_table.ReadU8(&format)) { + return OTS_FAILURE(); + } + switch (format) { + case 0: + for (uint16_t j = 1 /* .notdef is omitted */; j < glyphs; ++j) { + uint16_t sid = 0; + if (!cff_table.ReadU16(&sid)) { + return OTS_FAILURE(); + } + if (!have_ros && (sid > sid_max)) { + return OTS_FAILURE(); + } + // TODO(yusukes): check CIDs when have_ros is true. + } + break; + + case 1: + case 2: { + uint32_t total = 1; // .notdef is omitted. + while (total < glyphs) { + uint16_t sid = 0; + if (!cff_table.ReadU16(&sid)) { + return OTS_FAILURE(); + } + if (!have_ros && (sid > sid_max)) { + return OTS_FAILURE(); + } + // TODO(yusukes): check CIDs when have_ros is true. + + if (format == 1) { + uint8_t left = 0; + if (!cff_table.ReadU8(&left)) { + return OTS_FAILURE(); + } + total += (left + 1); + } else { + uint16_t left = 0; + if (!cff_table.ReadU16(&left)) { + return OTS_FAILURE(); + } + total += (left + 1); + } + } + break; + } + + default: + return OTS_FAILURE(); + } + } + } + return true; +} + +} // namespace + +namespace ots { + +bool ots_cff_parse(Font *font, const uint8_t *data, size_t length) { + Buffer table(data, length); + + font->cff = new OpenTypeCFF; + font->cff->data = data; + font->cff->length = length; + font->cff->font_dict_length = 0; + font->cff->local_subrs = NULL; + + // parse "6. Header" in the Adobe Compact Font Format Specification + uint8_t major = 0; + uint8_t minor = 0; + uint8_t hdr_size = 0; + uint8_t off_size = 0; + if (!table.ReadU8(&major)) { + return OTS_FAILURE(); + } + if (!table.ReadU8(&minor)) { + return OTS_FAILURE(); + } + if (!table.ReadU8(&hdr_size)) { + return OTS_FAILURE(); + } + if (!table.ReadU8(&off_size)) { + return OTS_FAILURE(); + } + if ((off_size == 0) || (off_size > 4)) { + return OTS_FAILURE(); + } + + if ((major != 1) || + (minor != 0) || + (hdr_size != 4)) { + return OTS_FAILURE(); + } + if (hdr_size >= length) { + return OTS_FAILURE(); + } + + // parse "7. Name INDEX" + table.set_offset(hdr_size); + CFFIndex name_index; + if (!ParseIndex(&table, &name_index)) { + return OTS_FAILURE(); + } + if (!ParseNameData(&table, name_index, &(font->cff->name))) { + return OTS_FAILURE(); + } + + // parse "8. Top DICT INDEX" + table.set_offset(name_index.offset_to_next); + CFFIndex top_dict_index; + if (!ParseIndex(&table, &top_dict_index)) { + return OTS_FAILURE(); + } + if (name_index.count != top_dict_index.count) { + return OTS_FAILURE(); + } + + // parse "10. String INDEX" + table.set_offset(top_dict_index.offset_to_next); + CFFIndex string_index; + if (!ParseIndex(&table, &string_index)) { + return OTS_FAILURE(); + } + if (string_index.count >= 65000 - kNStdString) { + return OTS_FAILURE(); + } + + const uint16_t num_glyphs = font->maxp->num_glyphs; + const size_t sid_max = string_index.count + kNStdString; + // string_index.count == 0 is allowed. + + // parse "9. Top DICT Data" + if (!ParseDictData(data, length, top_dict_index, + num_glyphs, sid_max, + DICT_DATA_TOPLEVEL, font->cff)) { + return OTS_FAILURE(); + } + + // parse "16. Global Subrs INDEX" + table.set_offset(string_index.offset_to_next); + CFFIndex global_subrs_index; + if (!ParseIndex(&table, &global_subrs_index)) { + return OTS_FAILURE(); + } + + // Check if all fd_index in FDSelect are valid. + std::map<uint16_t, uint8_t>::const_iterator iter; + std::map<uint16_t, uint8_t>::const_iterator end = font->cff->fd_select.end(); + for (iter = font->cff->fd_select.begin(); iter != end; ++iter) { + if (iter->second >= font->cff->font_dict_length) { + return OTS_FAILURE(); + } + } + + // Check if all charstrings (font hinting code for each glyph) are valid. + for (size_t i = 0; i < font->cff->char_strings_array.size(); ++i) { + if (!ValidateType2CharStringIndex(font, + *(font->cff->char_strings_array.at(i)), + global_subrs_index, + font->cff->fd_select, + font->cff->local_subrs_per_font, + font->cff->local_subrs, + &table)) { + return OTS_FAILURE_MSG("Failed validating charstring set %d", (int) i); + } + } + + return true; +} + +bool ots_cff_should_serialise(Font *font) { + return font->cff != NULL; +} + +bool ots_cff_serialise(OTSStream *out, Font *font) { + // TODO(yusukes): would be better to transcode the data, + // rather than simple memcpy. + if (!out->Write(font->cff->data, font->cff->length)) { + return OTS_FAILURE(); + } + return true; +} + +void ots_cff_reuse(Font *font, Font *other) { + font->cff = other->cff; + font->cff_reused = true; +} + +void ots_cff_free(Font *font) { + if (font->cff) { + for (size_t i = 0; i < font->cff->char_strings_array.size(); ++i) { + delete (font->cff->char_strings_array)[i]; + } + for (size_t i = 0; i < font->cff->local_subrs_per_font.size(); ++i) { + delete (font->cff->local_subrs_per_font)[i]; + } + delete font->cff->local_subrs; + delete font->cff; + } +} + +} // namespace ots + +#undef TABLE_NAME |