encode.go raw

   1  // Copyright 2011 The Snappy-Go Authors. All rights reserved.
   2  // Use of this source code is governed by a BSD-style
   3  // license that can be found in the LICENSE file.
   4  
   5  package snapref
   6  
   7  import (
   8  	"encoding/binary"
   9  	"errors"
  10  	"io"
  11  )
  12  
  13  // Encode returns the encoded form of src. The returned slice may be a sub-
  14  // slice of dst if dst was large enough to hold the entire encoded block.
  15  // Otherwise, a newly allocated slice will be returned.
  16  //
  17  // The dst and src must not overlap. It is valid to pass a nil dst.
  18  //
  19  // Encode handles the Snappy block format, not the Snappy stream format.
  20  func Encode(dst, src []byte) []byte {
  21  	if n := MaxEncodedLen(len(src)); n < 0 {
  22  		panic(ErrTooLarge)
  23  	} else if cap(dst) < n {
  24  		dst = make([]byte, n)
  25  	} else {
  26  		dst = dst[:n]
  27  	}
  28  
  29  	// The block starts with the varint-encoded length of the decompressed bytes.
  30  	d := binary.PutUvarint(dst, uint64(len(src)))
  31  
  32  	for len(src) > 0 {
  33  		p := src
  34  		src = nil
  35  		if len(p) > maxBlockSize {
  36  			p, src = p[:maxBlockSize], p[maxBlockSize:]
  37  		}
  38  		if len(p) < minNonLiteralBlockSize {
  39  			d += emitLiteral(dst[d:], p)
  40  		} else {
  41  			d += encodeBlock(dst[d:], p)
  42  		}
  43  	}
  44  	return dst[:d]
  45  }
  46  
  47  // inputMargin is the minimum number of extra input bytes to keep, inside
  48  // encodeBlock's inner loop. On some architectures, this margin lets us
  49  // implement a fast path for emitLiteral, where the copy of short (<= 16 byte)
  50  // literals can be implemented as a single load to and store from a 16-byte
  51  // register. That literal's actual length can be as short as 1 byte, so this
  52  // can copy up to 15 bytes too much, but that's OK as subsequent iterations of
  53  // the encoding loop will fix up the copy overrun, and this inputMargin ensures
  54  // that we don't overrun the dst and src buffers.
  55  const inputMargin = 16 - 1
  56  
  57  // minNonLiteralBlockSize is the minimum size of the input to encodeBlock that
  58  // could be encoded with a copy tag. This is the minimum with respect to the
  59  // algorithm used by encodeBlock, not a minimum enforced by the file format.
  60  //
  61  // The encoded output must start with at least a 1 byte literal, as there are
  62  // no previous bytes to copy. A minimal (1 byte) copy after that, generated
  63  // from an emitCopy call in encodeBlock's main loop, would require at least
  64  // another inputMargin bytes, for the reason above: we want any emitLiteral
  65  // calls inside encodeBlock's main loop to use the fast path if possible, which
  66  // requires being able to overrun by inputMargin bytes. Thus,
  67  // minNonLiteralBlockSize equals 1 + 1 + inputMargin.
  68  //
  69  // The C++ code doesn't use this exact threshold, but it could, as discussed at
  70  // https://groups.google.com/d/topic/snappy-compression/oGbhsdIJSJ8/discussion
  71  // The difference between Go (2+inputMargin) and C++ (inputMargin) is purely an
  72  // optimization. It should not affect the encoded form. This is tested by
  73  // TestSameEncodingAsCppShortCopies.
  74  const minNonLiteralBlockSize = 1 + 1 + inputMargin
  75  
  76  // MaxEncodedLen returns the maximum length of a snappy block, given its
  77  // uncompressed length.
  78  //
  79  // It will return a negative value if srcLen is too large to encode.
  80  func MaxEncodedLen(srcLen int) int {
  81  	n := uint64(srcLen)
  82  	if n > 0xffffffff {
  83  		return -1
  84  	}
  85  	// Compressed data can be defined as:
  86  	//    compressed := item* literal*
  87  	//    item       := literal* copy
  88  	//
  89  	// The trailing literal sequence has a space blowup of at most 62/60
  90  	// since a literal of length 60 needs one tag byte + one extra byte
  91  	// for length information.
  92  	//
  93  	// Item blowup is trickier to measure. Suppose the "copy" op copies
  94  	// 4 bytes of data. Because of a special check in the encoding code,
  95  	// we produce a 4-byte copy only if the offset is < 65536. Therefore
  96  	// the copy op takes 3 bytes to encode, and this type of item leads
  97  	// to at most the 62/60 blowup for representing literals.
  98  	//
  99  	// Suppose the "copy" op copies 5 bytes of data. If the offset is big
 100  	// enough, it will take 5 bytes to encode the copy op. Therefore the
 101  	// worst case here is a one-byte literal followed by a five-byte copy.
 102  	// That is, 6 bytes of input turn into 7 bytes of "compressed" data.
 103  	//
 104  	// This last factor dominates the blowup, so the final estimate is:
 105  	n = 32 + n + n/6
 106  	if n > 0xffffffff {
 107  		return -1
 108  	}
 109  	return int(n)
 110  }
 111  
 112  var errClosed = errors.New("snappy: Writer is closed")
 113  
 114  // NewWriter returns a new Writer that compresses to w.
 115  //
 116  // The Writer returned does not buffer writes. There is no need to Flush or
 117  // Close such a Writer.
 118  //
 119  // Deprecated: the Writer returned is not suitable for many small writes, only
 120  // for few large writes. Use NewBufferedWriter instead, which is efficient
 121  // regardless of the frequency and shape of the writes, and remember to Close
 122  // that Writer when done.
 123  func NewWriter(w io.Writer) *Writer {
 124  	return &Writer{
 125  		w:    w,
 126  		obuf: make([]byte, obufLen),
 127  	}
 128  }
 129  
 130  // NewBufferedWriter returns a new Writer that compresses to w, using the
 131  // framing format described at
 132  // https://github.com/google/snappy/blob/master/framing_format.txt
 133  //
 134  // The Writer returned buffers writes. Users must call Close to guarantee all
 135  // data has been forwarded to the underlying io.Writer. They may also call
 136  // Flush zero or more times before calling Close.
 137  func NewBufferedWriter(w io.Writer) *Writer {
 138  	return &Writer{
 139  		w:    w,
 140  		ibuf: make([]byte, 0, maxBlockSize),
 141  		obuf: make([]byte, obufLen),
 142  	}
 143  }
 144  
 145  // Writer is an io.Writer that can write Snappy-compressed bytes.
 146  //
 147  // Writer handles the Snappy stream format, not the Snappy block format.
 148  type Writer struct {
 149  	w   io.Writer
 150  	err error
 151  
 152  	// ibuf is a buffer for the incoming (uncompressed) bytes.
 153  	//
 154  	// Its use is optional. For backwards compatibility, Writers created by the
 155  	// NewWriter function have ibuf == nil, do not buffer incoming bytes, and
 156  	// therefore do not need to be Flush'ed or Close'd.
 157  	ibuf []byte
 158  
 159  	// obuf is a buffer for the outgoing (compressed) bytes.
 160  	obuf []byte
 161  
 162  	// wroteStreamHeader is whether we have written the stream header.
 163  	wroteStreamHeader bool
 164  }
 165  
 166  // Reset discards the writer's state and switches the Snappy writer to write to
 167  // w. This permits reusing a Writer rather than allocating a new one.
 168  func (w *Writer) Reset(writer io.Writer) {
 169  	w.w = writer
 170  	w.err = nil
 171  	if w.ibuf != nil {
 172  		w.ibuf = w.ibuf[:0]
 173  	}
 174  	w.wroteStreamHeader = false
 175  }
 176  
 177  // Write satisfies the io.Writer interface.
 178  func (w *Writer) Write(p []byte) (nRet int, errRet error) {
 179  	if w.ibuf == nil {
 180  		// Do not buffer incoming bytes. This does not perform or compress well
 181  		// if the caller of Writer.Write writes many small slices. This
 182  		// behavior is therefore deprecated, but still supported for backwards
 183  		// compatibility with code that doesn't explicitly Flush or Close.
 184  		return w.write(p)
 185  	}
 186  
 187  	// The remainder of this method is based on bufio.Writer.Write from the
 188  	// standard library.
 189  
 190  	for len(p) > (cap(w.ibuf)-len(w.ibuf)) && w.err == nil {
 191  		var n int
 192  		if len(w.ibuf) == 0 {
 193  			// Large write, empty buffer.
 194  			// Write directly from p to avoid copy.
 195  			n, _ = w.write(p)
 196  		} else {
 197  			n = copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p)
 198  			w.ibuf = w.ibuf[:len(w.ibuf)+n]
 199  			w.Flush()
 200  		}
 201  		nRet += n
 202  		p = p[n:]
 203  	}
 204  	if w.err != nil {
 205  		return nRet, w.err
 206  	}
 207  	n := copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p)
 208  	w.ibuf = w.ibuf[:len(w.ibuf)+n]
 209  	nRet += n
 210  	return nRet, nil
 211  }
 212  
 213  func (w *Writer) write(p []byte) (nRet int, errRet error) {
 214  	if w.err != nil {
 215  		return 0, w.err
 216  	}
 217  	for len(p) > 0 {
 218  		obufStart := len(magicChunk)
 219  		if !w.wroteStreamHeader {
 220  			w.wroteStreamHeader = true
 221  			copy(w.obuf, magicChunk)
 222  			obufStart = 0
 223  		}
 224  
 225  		var uncompressed []byte
 226  		if len(p) > maxBlockSize {
 227  			uncompressed, p = p[:maxBlockSize], p[maxBlockSize:]
 228  		} else {
 229  			uncompressed, p = p, nil
 230  		}
 231  		checksum := crc(uncompressed)
 232  
 233  		// Compress the buffer, discarding the result if the improvement
 234  		// isn't at least 12.5%.
 235  		compressed := Encode(w.obuf[obufHeaderLen:], uncompressed)
 236  		chunkType := uint8(chunkTypeCompressedData)
 237  		chunkLen := 4 + len(compressed)
 238  		obufEnd := obufHeaderLen + len(compressed)
 239  		if len(compressed) >= len(uncompressed)-len(uncompressed)/8 {
 240  			chunkType = chunkTypeUncompressedData
 241  			chunkLen = 4 + len(uncompressed)
 242  			obufEnd = obufHeaderLen
 243  		}
 244  
 245  		// Fill in the per-chunk header that comes before the body.
 246  		w.obuf[len(magicChunk)+0] = chunkType
 247  		w.obuf[len(magicChunk)+1] = uint8(chunkLen >> 0)
 248  		w.obuf[len(magicChunk)+2] = uint8(chunkLen >> 8)
 249  		w.obuf[len(magicChunk)+3] = uint8(chunkLen >> 16)
 250  		w.obuf[len(magicChunk)+4] = uint8(checksum >> 0)
 251  		w.obuf[len(magicChunk)+5] = uint8(checksum >> 8)
 252  		w.obuf[len(magicChunk)+6] = uint8(checksum >> 16)
 253  		w.obuf[len(magicChunk)+7] = uint8(checksum >> 24)
 254  
 255  		if _, err := w.w.Write(w.obuf[obufStart:obufEnd]); err != nil {
 256  			w.err = err
 257  			return nRet, err
 258  		}
 259  		if chunkType == chunkTypeUncompressedData {
 260  			if _, err := w.w.Write(uncompressed); err != nil {
 261  				w.err = err
 262  				return nRet, err
 263  			}
 264  		}
 265  		nRet += len(uncompressed)
 266  	}
 267  	return nRet, nil
 268  }
 269  
 270  // Flush flushes the Writer to its underlying io.Writer.
 271  func (w *Writer) Flush() error {
 272  	if w.err != nil {
 273  		return w.err
 274  	}
 275  	if len(w.ibuf) == 0 {
 276  		return nil
 277  	}
 278  	w.write(w.ibuf)
 279  	w.ibuf = w.ibuf[:0]
 280  	return w.err
 281  }
 282  
 283  // Close calls Flush and then closes the Writer.
 284  func (w *Writer) Close() error {
 285  	w.Flush()
 286  	ret := w.err
 287  	if w.err == nil {
 288  		w.err = errClosed
 289  	}
 290  	return ret
 291  }
 292