common/compress: internalize encoders, add length wrappers

3 files changed, 125 insertions, 70 deletions

diff --git a/common/bitutil/compress.go b/common/bitutil/compress.go
index a806c0e8b..c057cee4a 100644
--- a/common/bitutil/compress.go
+++ b/common/bitutil/compress.go
@@ -19,21 +19,21 @@ package bitutil
 import "errors"
 
 var (
-	// ErrMissingData is returned from decompression if the byte referenced by
+	// errMissingData is returned from decompression if the byte referenced by
 	// the bitset header overflows the input data.
-	ErrMissingData = errors.New("missing bytes on input")
+	errMissingData = errors.New("missing bytes on input")
 
-	// ErrUnreferencedData is returned from decompression if not all bytes were used
+	// 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")
+	errUnreferencedData = errors.New("extra bytes on input")
 
-	// ErrExceededTarget is returned from decompression if the bitset header has
+	// 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")
+	errExceededTarget = errors.New("target data size exceeded")
 
-	// ErrZeroContent is returned from decompression if a data byte referenced in
+	// 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")
+	errZeroContent = errors.New("zero byte in input content")
 )
 
 // The compression algorithm implemented by CompressBytes and DecompressBytes is
@@ -55,8 +55,20 @@ var (
 //         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.
+// representation algorithm. If the result is bigger than the original input, no
+// compression is done.
 func CompressBytes(data []byte) []byte {
+	if out := bitsetEncodeBytes(data); len(out) < len(data) {
+		return out
+	}
+	cpy := make([]byte, len(data))
+	copy(cpy, data)
+	return cpy
+}
+
+// bitsetEncodeBytes compresses the input byte slice according to the sparse
+// bitset representation algorithm.
+func bitsetEncodeBytes(data []byte) []byte {
 	// Empty slices get compressed to nil
 	if len(data) == 0 {
 		return nil
@@ -81,27 +93,41 @@ func CompressBytes(data []byte) []byte {
 	if len(nonZeroBytes) == 0 {
 		return nil
 	}
-	return append(CompressBytes(nonZeroBitset), nonZeroBytes...)
+	return append(bitsetEncodeBytes(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 as the input slice may be longer.
+// DecompressBytes decompresses data with a known target size. If the input data
+// matches the size of the target, it means no compression was done in the first
+// place.
 func DecompressBytes(data []byte, target int) ([]byte, error) {
-	out, size, err := decompressBytes(data, target)
+	if len(data) > target {
+		return nil, errExceededTarget
+	}
+	if len(data) == target {
+		cpy := make([]byte, len(data))
+		copy(cpy, data)
+		return cpy, nil
+	}
+	return bitsetDecodeBytes(data, target)
+}
+
+// bitsetDecodeBytes decompresses data with a known target size.
+func bitsetDecodeBytes(data []byte, target int) ([]byte, error) {
+	out, size, err := bitsetDecodePartialBytes(data, target)
 	if err != nil {
 		return nil, err
 	}
 	if size != len(data) {
-		return nil, ErrUnreferencedData
+		return nil, errUnreferencedData
 	}
 	return out, nil
 }
 
-// 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) {
+// bitsetDecodePartialBytes decompresses data with a known target size, but does
+// not enforce consuming all the input bytes. 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 bitsetDecodePartialBytes(data []byte, target int) ([]byte, int, error) {
 	// Sanity check 0 targets to avoid infinite recursion
 	if target == 0 {
 		return nil, 0, nil
@@ -119,7 +145,7 @@ func decompressBytes(data []byte, target int) ([]byte, int, error) {
 		return decomp, 0, nil
 	}
 	// Decompress the bitset of set bytes and distribute the non zero bytes
-	nonZeroBitset, ptr, err := decompressBytes(data, (target+7)/8)
+	nonZeroBitset, ptr, err := bitsetDecodePartialBytes(data, (target+7)/8)
 	if err != nil {
 		return nil, ptr, err
 	}
@@ -127,14 +153,14 @@ func decompressBytes(data []byte, target int) ([]byte, int, error) {
 		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
+				return nil, 0, errMissingData
 			}
 			if i >= len(decomp) {
-				return nil, 0, ErrExceededTarget
+				return nil, 0, errExceededTarget
 			}
 			// Make sure the data is valid and push into the slot
 			if data[ptr] == 0 {
-				return nil, 0, ErrZeroContent
+				return nil, 0, errZeroContent
 			}
 			decomp[i] = data[ptr]
 			ptr++
diff --git a/common/bitutil/compress_fuzz.go b/common/bitutil/compress_fuzz.go
index 2b7fe2977..1b87f50ed 100644
--- a/common/bitutil/compress_fuzz.go
+++ b/common/bitutil/compress_fuzz.go
@@ -20,36 +20,36 @@ package bitutil
 
 import "bytes"
 
-// Fuzz implements a go-fuzz fuzzer method to test various compression method
+// Fuzz implements a go-fuzz fuzzer method to test various encoding method
 // invocations.
 func Fuzz(data []byte) int {
 	if len(data) == 0 {
 		return -1
 	}
 	if data[0]%2 == 0 {
-		return fuzzCompress(data[1:])
+		return fuzzEncode(data[1:])
 	}
-	return fuzzDecompress(data[1:])
+	return fuzzDecode(data[1:])
 }
 
-// fuzzCompress implements a go-fuzz fuzzer method to test the bit compression and
-// decompression algorithm.
-func fuzzCompress(data []byte) int {
-	proc, _ := DecompressBytes(CompressBytes(data), len(data))
+// fuzzEncode implements a go-fuzz fuzzer method to test the bitset encoding and
+// decoding algorithm.
+func fuzzEncode(data []byte) int {
+	proc, _ := bitsetDecodeBytes(bitsetEncodeBytes(data), len(data))
 	if !bytes.Equal(data, proc) {
 		panic("content mismatch")
 	}
 	return 0
 }
 
-// fuzzDecompress implements a go-fuzz fuzzer method to test the bit decompression
-// and recompression algorithm.
-func fuzzDecompress(data []byte) int {
-	blob, err := DecompressBytes(data, 1024)
+// fuzzDecode implements a go-fuzz fuzzer method to test the bit decoding and
+// reencoding algorithm.
+func fuzzDecode(data []byte) int {
+	blob, err := bitsetDecodeBytes(data, 1024)
 	if err != nil {
 		return 0
 	}
-	if comp := CompressBytes(blob); !bytes.Equal(comp, data) {
+	if comp := bitsetEncodeBytes(blob); !bytes.Equal(comp, data) {
 		panic("content mismatch")
 	}
 	return 0
diff --git a/common/bitutil/compress_test.go b/common/bitutil/compress_test.go
index ef38bc7b3..805ab0369 100644
--- a/common/bitutil/compress_test.go
+++ b/common/bitutil/compress_test.go
@@ -24,8 +24,8 @@ import (
 	"github.com/ethereum/go-ethereum/common/hexutil"
 )
 
-// Tests that data compression and decompression works correctly.
-func TestCompressCycle(t *testing.T) {
+// Tests that data bitset encoding and decoding works and is bijective.
+func TestEncodingCycle(t *testing.T) {
 	tests := []string{
 		// Tests generated by go-fuzz to maximize code coverage
 		"0x000000000000000000",
@@ -50,7 +50,7 @@ func TestCompressCycle(t *testing.T) {
 	for i, tt := range tests {
 		data := hexutil.MustDecode(tt)
 
-		proc, err := DecompressBytes(CompressBytes(data), len(data))
+		proc, err := bitsetDecodeBytes(bitsetEncodeBytes(data), len(data))
 		if err != nil {
 			t.Errorf("test %d: failed to decompress compressed data: %v", i, err)
 			continue
@@ -61,8 +61,8 @@ func TestCompressCycle(t *testing.T) {
 	}
 }
 
-// Tests that data decompression works
-func TestDecompress(t *testing.T) {
+// Tests that data bitset decoding and rencoding works and is bijective.
+func TestDecodingCycle(t *testing.T) {
 	tests := []struct {
 		size  int
 		input string
@@ -71,22 +71,22 @@ func TestDecompress(t *testing.T) {
 		{size: 0, input: "0x"},
 
 		// Crashers generated by go-fuzz
-		{size: 0, input: "0x0020", fail: ErrUnreferencedData},
-		{size: 0, input: "0x30", fail: ErrUnreferencedData},
-		{size: 1, input: "0x00", fail: ErrUnreferencedData},
-		{size: 2, input: "0x07", fail: ErrMissingData},
-		{size: 1024, input: "0x8000", fail: ErrZeroContent},
+		{size: 0, input: "0x0020", fail: errUnreferencedData},
+		{size: 0, input: "0x30", fail: errUnreferencedData},
+		{size: 1, input: "0x00", fail: errUnreferencedData},
+		{size: 2, input: "0x07", fail: errMissingData},
+		{size: 1024, input: "0x8000", fail: errZeroContent},
 
 		// Tests generated by go-fuzz to maximize code coverage
-		{size: 29490, input: "0x343137343733323134333839373334323073333930783e3078333930783e70706336346c65303e", fail: ErrMissingData},
-		{size: 59395, input: "0x00", fail: ErrUnreferencedData},
-		{size: 52574, input: "0x70706336346c65c0de", fail: ErrExceededTarget},
-		{size: 42264, input: "0x07", fail: ErrMissingData},
-		{size: 52, input: "0xa5045bad48f4", fail: ErrExceededTarget},
-		{size: 52574, input: "0xc0de", fail: ErrMissingData},
+		{size: 29490, input: "0x343137343733323134333839373334323073333930783e3078333930783e70706336346c65303e", fail: errMissingData},
+		{size: 59395, input: "0x00", fail: errUnreferencedData},
+		{size: 52574, input: "0x70706336346c65c0de", fail: errExceededTarget},
+		{size: 42264, input: "0x07", fail: errMissingData},
+		{size: 52, input: "0xa5045bad48f4", fail: errExceededTarget},
+		{size: 52574, input: "0xc0de", fail: errMissingData},
 		{size: 52574, input: "0x"},
-		{size: 29490, input: "0x34313734373332313433383937333432307333393078073034333839373334323073333930783e3078333937333432307333393078073061333930783e70706336346c65303e", fail: ErrMissingData},
-		{size: 29491, input: "0x3973333930783e30783e", fail: ErrMissingData},
+		{size: 29490, input: "0x34313734373332313433383937333432307333393078073034333839373334323073333930783e3078333937333432307333393078073061333930783e70706336346c65303e", fail: errMissingData},
+		{size: 29491, input: "0x3973333930783e30783e", fail: errMissingData},
 
 		{size: 1024, input: "0x808080608080"},
 		{size: 1024, input: "0x808470705e3632383337363033313434303137393130306c6580ef46806380635a80"},
@@ -101,37 +101,66 @@ func TestDecompress(t *testing.T) {
 	for i, tt := range tests {
 		data := hexutil.MustDecode(tt.input)
 
-		orig, err := DecompressBytes(data, tt.size)
+		orig, err := bitsetDecodeBytes(data, tt.size)
 		if err != tt.fail {
 			t.Errorf("test %d: failure mismatch: have %v, want %v", i, err, tt.fail)
 		}
 		if err != nil {
 			continue
 		}
-		if comp := CompressBytes(orig); !bytes.Equal(comp, data) {
+		if comp := bitsetEncodeBytes(orig); !bytes.Equal(comp, data) {
 			t.Errorf("test %d: decompress/compress mismatch: have %x, want %x", i, comp, data)
 		}
 	}
 }
 
+// TestCompression tests that compression works by returning either the bitset
+// encoded input, or the actual input if the bitset version is longer.
+func TestCompression(t *testing.T) {
+	// Check the the compression returns the bitset encoding is shorter
+	in := hexutil.MustDecode("0x4912385c0e7b64000000")
+	out := hexutil.MustDecode("0x80fe4912385c0e7b64")
+
+	if data := CompressBytes(in); bytes.Compare(data, out) != 0 {
+		t.Errorf("encoding mismatch for sparse data: have %x, want %x", data, out)
+	}
+	if data, err := DecompressBytes(out, len(in)); err != nil || bytes.Compare(data, in) != 0 {
+		t.Errorf("decoding mismatch for sparse data: have %x, want %x, error %v", data, in, err)
+	}
+	// Check the the compression returns the input if the bitset encoding is longer
+	in = hexutil.MustDecode("0xdf7070533534333636313639343638373532313536346c1bc33339343837313070706336343035336336346c65fefb3930393233383838ac2f65fefb")
+	out = hexutil.MustDecode("0xdf7070533534333636313639343638373532313536346c1bc33339343837313070706336343035336336346c65fefb3930393233383838ac2f65fefb")
+
+	if data := CompressBytes(in); bytes.Compare(data, out) != 0 {
+		t.Errorf("encoding mismatch for dense data: have %x, want %x", data, out)
+	}
+	if data, err := DecompressBytes(out, len(in)); err != nil || bytes.Compare(data, in) != 0 {
+		t.Errorf("decoding mismatch for dense data: have %x, want %x, error %v", data, in, err)
+	}
+	// Check that decompressing a longer input than the target fails
+	if _, err := DecompressBytes([]byte{0xc0, 0x01, 0x01}, 2); err != errExceededTarget {
+		t.Errorf("decoding error mismatch for long data: have %v, want %v", err, errExceededTarget)
+	}
+}
+
 // Crude benchmark for compressing random slices of bytes.
-func BenchmarkCompress1KBVerySparse(b *testing.B) { benchmarkCompress(b, 1024, 0.0001) }
-func BenchmarkCompress2KBVerySparse(b *testing.B) { benchmarkCompress(b, 2048, 0.0001) }
-func BenchmarkCompress4KBVerySparse(b *testing.B) { benchmarkCompress(b, 4096, 0.0001) }
+func BenchmarkEncoding1KBVerySparse(b *testing.B) { benchmarkEncoding(b, 1024, 0.0001) }
+func BenchmarkEncoding2KBVerySparse(b *testing.B) { benchmarkEncoding(b, 2048, 0.0001) }
+func BenchmarkEncoding4KBVerySparse(b *testing.B) { benchmarkEncoding(b, 4096, 0.0001) }
 
-func BenchmarkCompress1KBSparse(b *testing.B) { benchmarkCompress(b, 1024, 0.001) }
-func BenchmarkCompress2KBSparse(b *testing.B) { benchmarkCompress(b, 2048, 0.001) }
-func BenchmarkCompress4KBSparse(b *testing.B) { benchmarkCompress(b, 4096, 0.001) }
+func BenchmarkEncoding1KBSparse(b *testing.B) { benchmarkEncoding(b, 1024, 0.001) }
+func BenchmarkEncoding2KBSparse(b *testing.B) { benchmarkEncoding(b, 2048, 0.001) }
+func BenchmarkEncoding4KBSparse(b *testing.B) { benchmarkEncoding(b, 4096, 0.001) }
 
-func BenchmarkCompress1KBDense(b *testing.B) { benchmarkCompress(b, 1024, 0.1) }
-func BenchmarkCompress2KBDense(b *testing.B) { benchmarkCompress(b, 2048, 0.1) }
-func BenchmarkCompress4KBDense(b *testing.B) { benchmarkCompress(b, 4096, 0.1) }
+func BenchmarkEncoding1KBDense(b *testing.B) { benchmarkEncoding(b, 1024, 0.1) }
+func BenchmarkEncoding2KBDense(b *testing.B) { benchmarkEncoding(b, 2048, 0.1) }
+func BenchmarkEncoding4KBDense(b *testing.B) { benchmarkEncoding(b, 4096, 0.1) }
 
-func BenchmarkCompress1KBSaturated(b *testing.B) { benchmarkCompress(b, 1024, 0.5) }
-func BenchmarkCompress2KBSaturated(b *testing.B) { benchmarkCompress(b, 2048, 0.5) }
-func BenchmarkCompress4KBSaturated(b *testing.B) { benchmarkCompress(b, 4096, 0.5) }
+func BenchmarkEncoding1KBSaturated(b *testing.B) { benchmarkEncoding(b, 1024, 0.5) }
+func BenchmarkEncoding2KBSaturated(b *testing.B) { benchmarkEncoding(b, 2048, 0.5) }
+func BenchmarkEncoding4KBSaturated(b *testing.B) { benchmarkEncoding(b, 4096, 0.5) }
 
-func benchmarkCompress(b *testing.B, bytes int, fill float64) {
+func benchmarkEncoding(b *testing.B, bytes int, fill float64) {
 	// Generate a random slice of bytes to compress
 	random := rand.NewSource(0) // reproducible and comparable
 
@@ -143,10 +172,10 @@ func benchmarkCompress(b *testing.B, bytes int, fill float64) {
 		bit := uint(random.Int63() % 8)
 		data[idx] |= 1 << bit
 	}
-	// Reset the benchmark and measure compression/decompression
+	// Reset the benchmark and measure encoding/decoding
 	b.ResetTimer()
 	b.ReportAllocs()
 	for i := 0; i < b.N; i++ {
-		DecompressBytes(CompressBytes(data), len(data))
+		bitsetDecodeBytes(bitsetEncodeBytes(data), len(data))
 	}
 }