2 * Copyright 2017 Facebook, Inc.
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
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
25 #include <folly/Range.h>
26 #include <folly/io/IOBuf.h>
29 * Compression / decompression over IOBufs
32 namespace folly { namespace io {
34 enum class CodecType {
36 * This codec type is not defined; getCodec() will throw an exception
37 * if used. Useful if deriving your own classes from Codec without
38 * going through the getCodec() interface.
49 * Use LZ4 compression.
50 * Levels supported: 1 = fast, 2 = best; default = 1
55 * Use Snappy compression.
61 * Use zlib compression.
62 * Levels supported: 0 = no compression, 1 = fast, ..., 9 = best; default = 6
67 * Use LZ4 compression, prefixed with size (as Varint).
72 * Use LZMA2 compression.
73 * Levels supported: 0 = no compression, 1 = fast, ..., 9 = best; default = 6
76 LZMA2_VARINT_SIZE = 7,
79 * Use ZSTD compression.
84 * Use gzip compression. This is the same compression algorithm as ZLIB but
85 * gzip-compressed files tend to be easier to work with from the command line.
86 * Levels supported: 0 = no compression, 1 = fast, ..., 9 = best; default = 6
91 * Use LZ4 frame compression.
92 * Levels supported: 0 = fast, 16 = best; default = 0
97 * Use bzip2 compression.
98 * Levels supported: 1 = fast, 9 = best; default = 9
102 NUM_CODEC_TYPES = 12,
110 * Return the maximum length of data that may be compressed with this codec.
111 * NO_COMPRESSION and ZLIB support arbitrary lengths;
112 * LZ4 supports up to 1.9GiB; SNAPPY supports up to 4GiB.
113 * May return UNLIMITED_UNCOMPRESSED_LENGTH if unlimited.
115 uint64_t maxUncompressedLength() const;
118 * Return the codec's type.
120 CodecType type() const { return type_; }
123 * Does this codec need the exact uncompressed length on decompression?
125 bool needsUncompressedLength() const;
128 * Compress data, returning an IOBuf (which may share storage with data).
129 * Throws std::invalid_argument if data is larger than
130 * maxUncompressedLength().
132 * Regardless of the behavior of the underlying compressor, compressing
133 * an empty IOBuf chain will return an empty IOBuf chain.
135 std::unique_ptr<IOBuf> compress(const folly::IOBuf* data);
138 * Compresses data. May involve additional copies compared to the overload
139 * that takes and returns IOBufs. Has the same error semantics as the IOBuf
142 std::string compress(StringPiece data);
145 * Uncompress data. Throws std::runtime_error on decompression error.
147 * Some codecs (LZ4) require the exact uncompressed length; this is indicated
148 * by needsUncompressedLength().
150 * For other codes (zlib), knowing the exact uncompressed length ahead of
151 * time might be faster.
153 * Regardless of the behavior of the underlying compressor, uncompressing
154 * an empty IOBuf chain will return an empty IOBuf chain.
156 static constexpr uint64_t UNKNOWN_UNCOMPRESSED_LENGTH = uint64_t(-1);
157 static constexpr uint64_t UNLIMITED_UNCOMPRESSED_LENGTH = uint64_t(-2);
159 std::unique_ptr<IOBuf> uncompress(
161 uint64_t uncompressedLength = UNKNOWN_UNCOMPRESSED_LENGTH);
164 * Uncompresses data. May involve additional copies compared to the overload
165 * that takes and returns IOBufs. Has the same error semantics as the IOBuf
168 std::string uncompress(
170 uint64_t uncompressedLength = UNKNOWN_UNCOMPRESSED_LENGTH);
173 explicit Codec(CodecType type);
177 * Returns a superset of the set of prefixes for which canUncompress() will
178 * return true. A superset is allowed for optimizations in canUncompress()
179 * based on other knowledge such as length. None of the prefixes may be empty.
180 * default: No prefixes.
182 virtual std::vector<std::string> validPrefixes() const;
185 * Returns true if the codec thinks it can uncompress the data.
186 * If a codec doesn't have magic bytes at the beginning, like LZ4 and Snappy,
187 * it can always return false.
188 * default: Returns false.
190 virtual bool canUncompress(
191 const folly::IOBuf* data,
192 uint64_t uncompressedLength = UNKNOWN_UNCOMPRESSED_LENGTH) const;
195 // default: no limits (save for special value UNKNOWN_UNCOMPRESSED_LENGTH)
196 virtual uint64_t doMaxUncompressedLength() const;
197 // default: doesn't need uncompressed length
198 virtual bool doNeedsUncompressedLength() const;
199 virtual std::unique_ptr<IOBuf> doCompress(const folly::IOBuf* data) = 0;
200 virtual std::unique_ptr<IOBuf> doUncompress(const folly::IOBuf* data,
201 uint64_t uncompressedLength) = 0;
202 // default: an implementation is provided by default to wrap the strings into
203 // IOBufs and delegate to the IOBuf methods. This incurs a copy of the output
204 // from IOBuf to string. Implementers, at their discretion, can override
205 // these methods to avoid the copy.
206 virtual std::string doCompressString(StringPiece data);
207 virtual std::string doUncompressString(
209 uint64_t uncompressedLength);
214 constexpr int COMPRESSION_LEVEL_FASTEST = -1;
215 constexpr int COMPRESSION_LEVEL_DEFAULT = -2;
216 constexpr int COMPRESSION_LEVEL_BEST = -3;
219 * Return a codec for the given type. Throws on error. The level
220 * is a non-negative codec-dependent integer indicating the level of
221 * compression desired, or one of the following constants:
223 * COMPRESSION_LEVEL_FASTEST is fastest (uses least CPU / memory,
225 * COMPRESSION_LEVEL_DEFAULT is the default (likely a tradeoff between
227 * COMPRESSION_LEVEL_BEST is the best compression (uses most CPU / memory,
230 * When decompressing, the compression level is ignored. All codecs will
231 * decompress all data compressed with the a codec of the same type, regardless
232 * of compression level.
234 std::unique_ptr<Codec> getCodec(CodecType type,
235 int level = COMPRESSION_LEVEL_DEFAULT);
238 * Returns a codec that can uncompress any of the given codec types as well as
239 * {LZ4_FRAME, ZSTD, ZLIB, GZIP, LZMA2, BZIP2}. Appends each default codec to
240 * customCodecs in order, so long as a codec with the same type() isn't already
241 * present. When uncompress() is called, each codec's canUncompress() is called
242 * in the order that they are given. Appended default codecs are checked last.
243 * uncompress() is called on the first codec whose canUncompress() returns true.
244 * An exception is thrown if no codec canUncompress() the data.
245 * An exception is thrown if the chosen codec's uncompress() throws on the data.
246 * An exception is thrown if compress() is called on the returned codec.
248 * Requirements are checked in debug mode and are as follows:
249 * Let headers be the concatenation of every codec's validPrefixes().
250 * 1. Each codec must override validPrefixes() and canUncompress().
251 * 2. No codec's validPrefixes() may be empty.
252 * 3. No header in headers may be empty.
253 * 4. headers must not contain any duplicate elements.
254 * 5. No strict non-empty prefix of any header in headers may be in headers.
256 std::unique_ptr<Codec> getAutoUncompressionCodec(
257 std::vector<std::unique_ptr<Codec>> customCodecs = {});
260 * Check if a specified codec is supported.
262 bool hasCodec(CodecType type);