Lukas Zilka | e7962cc | 2018-03-28 18:09:48 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2017 The Android Open Source Project |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | #include "zlib-utils.h" |
| 18 | |
| 19 | #include <memory> |
| 20 | |
| 21 | #include "util/base/logging.h" |
Lukas Zilka | 434442d | 2018-04-25 11:38:51 +0200 | [diff] [blame] | 22 | #include "util/flatbuffers.h" |
Lukas Zilka | e7962cc | 2018-03-28 18:09:48 +0200 | [diff] [blame] | 23 | |
| 24 | namespace libtextclassifier2 { |
| 25 | |
| 26 | std::unique_ptr<ZlibDecompressor> ZlibDecompressor::Instance() { |
| 27 | std::unique_ptr<ZlibDecompressor> result(new ZlibDecompressor()); |
| 28 | if (!result->initialized_) { |
| 29 | result.reset(); |
| 30 | } |
| 31 | return result; |
| 32 | } |
| 33 | |
| 34 | ZlibDecompressor::ZlibDecompressor() { |
| 35 | memset(&stream_, 0, sizeof(stream_)); |
| 36 | stream_.zalloc = Z_NULL; |
| 37 | stream_.zfree = Z_NULL; |
| 38 | initialized_ = (inflateInit(&stream_) == Z_OK); |
| 39 | } |
| 40 | |
| 41 | ZlibDecompressor::~ZlibDecompressor() { |
| 42 | if (initialized_) { |
| 43 | inflateEnd(&stream_); |
| 44 | } |
| 45 | } |
| 46 | |
| 47 | bool ZlibDecompressor::Decompress(const CompressedBuffer* compressed_buffer, |
| 48 | std::string* out) { |
| 49 | out->resize(compressed_buffer->uncompressed_size()); |
| 50 | stream_.next_in = |
| 51 | reinterpret_cast<const Bytef*>(compressed_buffer->buffer()->Data()); |
| 52 | stream_.avail_in = compressed_buffer->buffer()->Length(); |
| 53 | stream_.next_out = reinterpret_cast<Bytef*>(const_cast<char*>(out->c_str())); |
| 54 | stream_.avail_out = compressed_buffer->uncompressed_size(); |
| 55 | return (inflate(&stream_, Z_SYNC_FLUSH) == Z_OK); |
| 56 | } |
| 57 | |
| 58 | std::unique_ptr<ZlibCompressor> ZlibCompressor::Instance() { |
| 59 | std::unique_ptr<ZlibCompressor> result(new ZlibCompressor()); |
| 60 | if (!result->initialized_) { |
| 61 | result.reset(); |
| 62 | } |
| 63 | return result; |
| 64 | } |
| 65 | |
| 66 | ZlibCompressor::ZlibCompressor(int level, int tmp_buffer_size) { |
| 67 | memset(&stream_, 0, sizeof(stream_)); |
| 68 | stream_.zalloc = Z_NULL; |
| 69 | stream_.zfree = Z_NULL; |
| 70 | buffer_size_ = tmp_buffer_size; |
| 71 | buffer_.reset(new Bytef[buffer_size_]); |
| 72 | initialized_ = (deflateInit(&stream_, level) == Z_OK); |
| 73 | } |
| 74 | |
| 75 | ZlibCompressor::~ZlibCompressor() { deflateEnd(&stream_); } |
| 76 | |
| 77 | void ZlibCompressor::Compress(const std::string& uncompressed_content, |
| 78 | CompressedBufferT* out) { |
| 79 | out->uncompressed_size = uncompressed_content.size(); |
| 80 | out->buffer.clear(); |
| 81 | stream_.next_in = |
| 82 | reinterpret_cast<const Bytef*>(uncompressed_content.c_str()); |
| 83 | stream_.avail_in = uncompressed_content.size(); |
| 84 | stream_.next_out = buffer_.get(); |
| 85 | stream_.avail_out = buffer_size_; |
| 86 | unsigned char* buffer_deflate_start_position = |
| 87 | reinterpret_cast<unsigned char*>(buffer_.get()); |
| 88 | int status; |
| 89 | do { |
| 90 | // Deflate chunk-wise. |
| 91 | // Z_SYNC_FLUSH causes all pending output to be flushed, but doesn't |
| 92 | // reset the compression state. |
| 93 | // As we do not know how big the compressed buffer will be, we compress |
| 94 | // chunk wise and append the flushed content to the output string buffer. |
| 95 | // As we store the uncompressed size, we do not have to do this during |
| 96 | // decompression. |
| 97 | status = deflate(&stream_, Z_SYNC_FLUSH); |
| 98 | unsigned char* buffer_deflate_end_position = |
| 99 | reinterpret_cast<unsigned char*>(stream_.next_out); |
| 100 | if (buffer_deflate_end_position != buffer_deflate_start_position) { |
| 101 | out->buffer.insert(out->buffer.end(), buffer_deflate_start_position, |
| 102 | buffer_deflate_end_position); |
| 103 | stream_.next_out = buffer_deflate_start_position; |
| 104 | stream_.avail_out = buffer_size_; |
| 105 | } else { |
| 106 | break; |
| 107 | } |
| 108 | } while (status == Z_OK); |
| 109 | } |
| 110 | |
| 111 | // Compress rule fields in the model. |
| 112 | bool CompressModel(ModelT* model) { |
| 113 | std::unique_ptr<ZlibCompressor> zlib_compressor = ZlibCompressor::Instance(); |
| 114 | if (!zlib_compressor) { |
| 115 | TC_LOG(ERROR) << "Cannot compress model."; |
| 116 | return false; |
| 117 | } |
| 118 | |
| 119 | // Compress regex rules. |
| 120 | if (model->regex_model != nullptr) { |
| 121 | for (int i = 0; i < model->regex_model->patterns.size(); i++) { |
| 122 | RegexModel_::PatternT* pattern = model->regex_model->patterns[i].get(); |
| 123 | pattern->compressed_pattern.reset(new CompressedBufferT); |
| 124 | zlib_compressor->Compress(pattern->pattern, |
| 125 | pattern->compressed_pattern.get()); |
| 126 | pattern->pattern.clear(); |
| 127 | } |
| 128 | } |
| 129 | |
| 130 | // Compress date-time rules. |
| 131 | if (model->datetime_model != nullptr) { |
| 132 | for (int i = 0; i < model->datetime_model->patterns.size(); i++) { |
| 133 | DatetimeModelPatternT* pattern = model->datetime_model->patterns[i].get(); |
| 134 | for (int j = 0; j < pattern->regexes.size(); j++) { |
| 135 | DatetimeModelPattern_::RegexT* regex = pattern->regexes[j].get(); |
| 136 | regex->compressed_pattern.reset(new CompressedBufferT); |
| 137 | zlib_compressor->Compress(regex->pattern, |
| 138 | regex->compressed_pattern.get()); |
| 139 | regex->pattern.clear(); |
| 140 | } |
| 141 | } |
| 142 | for (int i = 0; i < model->datetime_model->extractors.size(); i++) { |
| 143 | DatetimeModelExtractorT* extractor = |
| 144 | model->datetime_model->extractors[i].get(); |
| 145 | extractor->compressed_pattern.reset(new CompressedBufferT); |
| 146 | zlib_compressor->Compress(extractor->pattern, |
| 147 | extractor->compressed_pattern.get()); |
| 148 | extractor->pattern.clear(); |
| 149 | } |
| 150 | } |
| 151 | return true; |
| 152 | } |
| 153 | |
Lukas Zilka | 434442d | 2018-04-25 11:38:51 +0200 | [diff] [blame] | 154 | namespace { |
| 155 | |
| 156 | bool DecompressBuffer(const CompressedBufferT* compressed_pattern, |
| 157 | ZlibDecompressor* zlib_decompressor, |
| 158 | std::string* uncompressed_pattern) { |
| 159 | std::string packed_pattern = |
| 160 | PackFlatbuffer<CompressedBuffer>(compressed_pattern); |
| 161 | if (!zlib_decompressor->Decompress( |
| 162 | LoadAndVerifyFlatbuffer<CompressedBuffer>(packed_pattern), |
| 163 | uncompressed_pattern)) { |
| 164 | return false; |
| 165 | } |
| 166 | return true; |
| 167 | } |
| 168 | |
| 169 | } // namespace |
| 170 | |
| 171 | bool DecompressModel(ModelT* model) { |
| 172 | std::unique_ptr<ZlibDecompressor> zlib_decompressor = |
| 173 | ZlibDecompressor::Instance(); |
| 174 | if (!zlib_decompressor) { |
| 175 | TC_LOG(ERROR) << "Cannot initialize decompressor."; |
| 176 | return false; |
| 177 | } |
| 178 | |
| 179 | // Decompress regex rules. |
| 180 | if (model->regex_model != nullptr) { |
| 181 | for (int i = 0; i < model->regex_model->patterns.size(); i++) { |
| 182 | RegexModel_::PatternT* pattern = model->regex_model->patterns[i].get(); |
| 183 | if (!DecompressBuffer(pattern->compressed_pattern.get(), |
| 184 | zlib_decompressor.get(), &pattern->pattern)) { |
| 185 | TC_LOG(ERROR) << "Cannot decompress pattern: " << i; |
| 186 | return false; |
| 187 | } |
| 188 | pattern->compressed_pattern.reset(nullptr); |
| 189 | } |
| 190 | } |
| 191 | |
| 192 | // Decompress date-time rules. |
| 193 | if (model->datetime_model != nullptr) { |
| 194 | for (int i = 0; i < model->datetime_model->patterns.size(); i++) { |
| 195 | DatetimeModelPatternT* pattern = model->datetime_model->patterns[i].get(); |
| 196 | for (int j = 0; j < pattern->regexes.size(); j++) { |
| 197 | DatetimeModelPattern_::RegexT* regex = pattern->regexes[j].get(); |
| 198 | if (!DecompressBuffer(regex->compressed_pattern.get(), |
| 199 | zlib_decompressor.get(), ®ex->pattern)) { |
| 200 | TC_LOG(ERROR) << "Cannot decompress pattern: " << i << " " << j; |
| 201 | return false; |
| 202 | } |
| 203 | regex->compressed_pattern.reset(nullptr); |
| 204 | } |
| 205 | } |
| 206 | for (int i = 0; i < model->datetime_model->extractors.size(); i++) { |
| 207 | DatetimeModelExtractorT* extractor = |
| 208 | model->datetime_model->extractors[i].get(); |
| 209 | if (!DecompressBuffer(extractor->compressed_pattern.get(), |
| 210 | zlib_decompressor.get(), &extractor->pattern)) { |
| 211 | TC_LOG(ERROR) << "Cannot decompress pattern: " << i; |
| 212 | return false; |
| 213 | } |
| 214 | extractor->compressed_pattern.reset(nullptr); |
| 215 | } |
| 216 | } |
| 217 | return true; |
| 218 | } |
| 219 | |
Lukas Zilka | e7962cc | 2018-03-28 18:09:48 +0200 | [diff] [blame] | 220 | std::string CompressSerializedModel(const std::string& model) { |
| 221 | std::unique_ptr<ModelT> unpacked_model = UnPackModel(model.c_str()); |
| 222 | TC_CHECK(unpacked_model != nullptr); |
| 223 | TC_CHECK(CompressModel(unpacked_model.get())); |
| 224 | flatbuffers::FlatBufferBuilder builder; |
| 225 | FinishModelBuffer(builder, Model::Pack(builder, unpacked_model.get())); |
| 226 | return std::string(reinterpret_cast<const char*>(builder.GetBufferPointer()), |
| 227 | builder.GetSize()); |
| 228 | } |
| 229 | |
| 230 | std::unique_ptr<UniLib::RegexPattern> UncompressMakeRegexPattern( |
| 231 | const UniLib& unilib, const flatbuffers::String* uncompressed_pattern, |
Lukas Zilka | 434442d | 2018-04-25 11:38:51 +0200 | [diff] [blame] | 232 | const CompressedBuffer* compressed_pattern, ZlibDecompressor* decompressor, |
| 233 | std::string* result_pattern_text) { |
Lukas Zilka | e7962cc | 2018-03-28 18:09:48 +0200 | [diff] [blame] | 234 | UnicodeText unicode_regex_pattern; |
| 235 | std::string decompressed_pattern; |
| 236 | if (compressed_pattern != nullptr && |
| 237 | compressed_pattern->buffer() != nullptr) { |
| 238 | if (decompressor == nullptr || |
| 239 | !decompressor->Decompress(compressed_pattern, &decompressed_pattern)) { |
| 240 | TC_LOG(ERROR) << "Cannot decompress pattern."; |
| 241 | return nullptr; |
| 242 | } |
| 243 | unicode_regex_pattern = |
| 244 | UTF8ToUnicodeText(decompressed_pattern.data(), |
| 245 | decompressed_pattern.size(), /*do_copy=*/false); |
| 246 | } else { |
| 247 | if (uncompressed_pattern == nullptr) { |
| 248 | TC_LOG(ERROR) << "Cannot load uncompressed pattern."; |
| 249 | return nullptr; |
| 250 | } |
| 251 | unicode_regex_pattern = |
| 252 | UTF8ToUnicodeText(uncompressed_pattern->c_str(), |
| 253 | uncompressed_pattern->Length(), /*do_copy=*/false); |
| 254 | } |
| 255 | |
Lukas Zilka | 434442d | 2018-04-25 11:38:51 +0200 | [diff] [blame] | 256 | if (result_pattern_text != nullptr) { |
| 257 | *result_pattern_text = unicode_regex_pattern.ToUTF8String(); |
| 258 | } |
| 259 | |
Lukas Zilka | e7962cc | 2018-03-28 18:09:48 +0200 | [diff] [blame] | 260 | std::unique_ptr<UniLib::RegexPattern> regex_pattern = |
| 261 | unilib.CreateRegexPattern(unicode_regex_pattern); |
| 262 | if (!regex_pattern) { |
| 263 | TC_LOG(ERROR) << "Could not create pattern: " |
| 264 | << unicode_regex_pattern.ToUTF8String(); |
| 265 | } |
| 266 | return regex_pattern; |
| 267 | } |
| 268 | |
| 269 | } // namespace libtextclassifier2 |