common/bitutil: fix decompression corner cases; fuzz, test & bench

This commit is contained in:
Péter Szilágyi
2017-05-06 18:35:59 +03:00
parent fd5d51c9ae
commit cf19586cfb
3 changed files with 302 additions and 43 deletions

View File

@ -16,78 +16,129 @@
package bitutil
/*
The compression algorithm implemented by CompressBytes and DecompressBytes is
optimized for "sparse" input data which contains a lot of zero bytes. Decompression
requires knowledge of the decompressed data length. Compression works as follows:
import "errors"
if data only contains zeroes,
CompressBytes(data) == nil
otherwise if len(data) <= 1,
CompressBytes(data) == data
otherwise:
CompressBytes(data) == append(CompressBytes(nonZeroBits(data)), nonZeroBytes(data)...)
where
nonZeroBits(data) is a bit vector with len(data) bits (MSB first):
nonZeroBits(data)[i/8] && (1 << (7-i%8)) != 0 if data[i] != 0
len(nonZeroBits(data)) == (len(data)+7)/8
nonZeroBytes(data) contains the non-zero bytes of data in the same order
*/
var (
// ErrMissingData is returned from decompression if the byte referenced by
// the bitset header overflows the input data.
ErrMissingData = errors.New("missing bytes on input")
// CompressBytes compresses the input byte slice
// ErrUnreferencedData is returned from decompression if not all bytes were used
// up from the input data after decompressing it.
ErrUnreferencedData = errors.New("extra bytes on input")
// ErrExceededTarget is returned from decompression if the bitset header has
// more bits defined than the number of target buffer space available.
ErrExceededTarget = errors.New("target data size exceeded")
// ErrZeroContent is returned from decompression if a data byte referenced in
// the bitset header is actually a zero byte.
ErrZeroContent = errors.New("zero byte in input content")
)
// The compression algorithm implemented by CompressBytes and DecompressBytes is
// optimized for sparse input data which contains a lot of zero bytes. Decompression
// requires knowledge of the decompressed data length.
//
// Compression works as follows:
//
// if data only contains zeroes,
// CompressBytes(data) == nil
// otherwise if len(data) <= 1,
// CompressBytes(data) == data
// otherwise:
// CompressBytes(data) == append(CompressBytes(nonZeroBitset(data)), nonZeroBytes(data)...)
// where
// nonZeroBitset(data) is a bit vector with len(data) bits (MSB first):
// nonZeroBitset(data)[i/8] && (1 << (7-i%8)) != 0 if data[i] != 0
// len(nonZeroBitset(data)) == (len(data)+7)/8
// nonZeroBytes(data) contains the non-zero bytes of data in the same order
// CompressBytes compresses the input byte slice according to the sparse bitset
// representation algorithm.
func CompressBytes(data []byte) []byte {
// Empty slices get compressed to nil
if len(data) == 0 {
return nil
}
// One byte slices compress to nil or retain the single byte
if len(data) == 1 {
if data[0] == 0 {
return nil
} else {
return data
}
return data
}
bitsLen := (len(data) + 7) / 8
nonZeroBits := make([]byte, bitsLen)
// Calculate the bitset of set bytes, and gather the non-zero bytes
nonZeroBitset := make([]byte, (len(data)+7)/8)
nonZeroBytes := make([]byte, 0, len(data))
for i, b := range data {
if b != 0 {
nonZeroBytes = append(nonZeroBytes, b)
nonZeroBits[i/8] |= 1 << byte(7-i%8)
nonZeroBitset[i/8] |= 1 << byte(7-i%8)
}
}
if len(nonZeroBytes) == 0 {
return nil
}
return append(CompressBytes(nonZeroBits), nonZeroBytes...)
return append(CompressBytes(nonZeroBitset), nonZeroBytes...)
}
// DecompressBytes decompresses data with a known target size.
// In addition to the decompressed output, the function returns the length of
// compressed input data corresponding to the output. The input slice may be longer.
// If the input slice is too short, (nil, -1) is returned.
func DecompressBytes(data []byte, targetLen int) ([]byte, int) {
decomp := make([]byte, targetLen)
if len(data) == 0 {
return decomp, 0
// DecompressBytes decompresses data with a known target size. In addition to the
// decompressed output, the function returns the length of compressed input data
// corresponding to the output as the input slice may be longer.
func DecompressBytes(data []byte, target int) ([]byte, error) {
out, size, err := decompressBytes(data, target)
if err != nil {
return nil, err
}
if targetLen == 1 {
return data[0:1], 1
if size != len(data) {
return nil, ErrUnreferencedData
}
return out, nil
}
bitsLen := (targetLen + 7) / 8
nonZeroBits, ptr := DecompressBytes(data, bitsLen)
if ptr < 0 {
return nil, -1
// decompressBytes decompresses data with a known target size. In addition to the
// decompressed output, the function returns the length of compressed input data
// corresponding to the output as the input slice may be longer.
func decompressBytes(data []byte, target int) ([]byte, int, error) {
// Sanity check 0 targets to avoid infinite recursion
if target == 0 {
return nil, 0, nil
}
for i, _ := range decomp {
if nonZeroBits[i/8]&(1<<byte(7-i%8)) != 0 {
if ptr == len(data) {
return nil, -1
// Handle the zero and single byte corner cases
decomp := make([]byte, target)
if len(data) == 0 {
return decomp, 0, nil
}
if target == 1 {
decomp[0] = data[0] // copy to avoid referencing the input slice
if data[0] != 0 {
return decomp, 1, nil
}
return decomp, 0, nil
}
// Decompress the bitset of set bytes and distribute the non zero bytes
nonZeroBitset, ptr, err := decompressBytes(data, (target+7)/8)
if err != nil {
return nil, ptr, err
}
for i := 0; i < 8*len(nonZeroBitset); i++ {
if nonZeroBitset[i/8]&(1<<byte(7-i%8)) != 0 {
// Make sure we have enough data to push into the correct slot
if ptr >= len(data) {
return nil, 0, ErrMissingData
}
if i >= len(decomp) {
return nil, 0, ErrExceededTarget
}
// Make sure the data is valid and push into the slot
if data[ptr] == 0 {
return nil, 0, ErrZeroContent
}
decomp[i] = data[ptr]
ptr++
}
}
return decomp, ptr
return decomp, ptr, nil
}