Split source files by patch type (text/binary)

The tests were already split up in this way, so it makes sense to have
the parsing functions split as well.

Author: bluekeyes

gitdiff/binary.go

@@ -0,0 +1,179 @@
+package gitdiff
+
+import (
+	"bytes"
+	"compress/zlib"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"strconv"
+	"strings"
+)
+
+func (p *parser) ParseBinaryFragments(f *File) (n int, err error) {
+	isBinary, hasData, err := p.ParseBinaryMarker()
+	if err != nil || !isBinary {
+		return 0, err
+	}
+
+	f.IsBinary = true
+	if !hasData {
+		return 0, nil
+	}
+
+	forward, err := p.ParseBinaryFragmentHeader()
+	if err != nil {
+		return 0, err
+	}
+	if forward == nil {
+		return 0, p.Errorf(0, "missing data for binary patch")
+	}
+	if err := p.ParseBinaryChunk(forward); err != nil {
+		return 0, err
+	}
+	f.BinaryFragment = forward
+
+	// valid for reverse to not exist, but it must be valid if present
+	reverse, err := p.ParseBinaryFragmentHeader()
+	if err != nil {
+		return 1, err
+	}
+	if reverse == nil {
+		return 1, nil
+	}
+	if err := p.ParseBinaryChunk(reverse); err != nil {
+		return 1, err
+	}
+	f.ReverseBinaryFragment = reverse
+
+	return 1, nil
+}
+
+func (p *parser) ParseBinaryMarker() (isBinary bool, hasData bool, err error) {
+	switch p.Line(0) {
+	case "GIT binary patch\n":
+		hasData = true
+	case "Binary files differ\n":
+	case "Files differ\n":
+	default:
+		return false, false, nil
+	}
+
+	if err = p.Next(); err != nil && err != io.EOF {
+		return false, false, err
+	}
+	return true, hasData, nil
+}
+
+func (p *parser) ParseBinaryFragmentHeader() (*BinaryFragment, error) {
+	parts := strings.SplitN(strings.TrimSuffix(p.Line(0), "\n"), " ", 2)
+	if len(parts) < 2 {
+		return nil, nil
+	}
+
+	frag := &BinaryFragment{}
+	switch parts[0] {
+	case "delta":
+		frag.Method = BinaryPatchDelta
+	case "literal":
+		frag.Method = BinaryPatchLiteral
+	default:
+		return nil, nil
+	}
+
+	var err error
+	if frag.Size, err = strconv.ParseInt(parts[1], 10, 64); err != nil {
+		nerr := err.(*strconv.NumError)
+		return nil, p.Errorf(0, "binary patch: invalid size: %v", nerr.Err)
+	}
+
+	if err := p.Next(); err != nil && err != io.EOF {
+		return nil, err
+	}
+	return frag, nil
+}
+
+func (p *parser) ParseBinaryChunk(frag *BinaryFragment) error {
+	// Binary fragments are encoded as a series of base85 encoded lines. Each
+	// line starts with a character in [A-Za-z] giving the number of bytes on
+	// the line, where A = 1 and z = 52, and ends with a newline character.
+	//
+	// The base85 encoding means each line is a multiple of 5 characters + 2
+	// additional characters for the length byte and the newline. The fragment
+	// ends with a blank line.
+	const (
+		shortestValidLine = "A00000\n"
+		maxBytesPerLine   = 52
+	)
+
+	var data bytes.Buffer
+	buf := make([]byte, maxBytesPerLine)
+	for {
+		line := p.Line(0)
+		if line == "\n" {
+			break
+		}
+		if len(line) < len(shortestValidLine) || (len(line)-2)%5 != 0 {
+			return p.Errorf(0, "binary patch: corrupt data line")
+		}
+
+		byteCount, seq := int(line[0]), line[1:len(line)-1]
+		switch {
+		case 'A' <= byteCount && byteCount <= 'Z':
+			byteCount = byteCount - 'A' + 1
+		case 'a' <= byteCount && byteCount <= 'z':
+			byteCount = byteCount - 'a' + 27
+		default:
+			return p.Errorf(0, "binary patch: invalid length byte")
+		}
+
+		// base85 encodes every 4 bytes into 5 characters, with up to 3 bytes of end padding
+		maxByteCount := len(seq) / 5 * 4
+		if byteCount > maxByteCount || byteCount < maxByteCount-3 {
+			return p.Errorf(0, "binary patch: incorrect byte count")
+		}
+
+		if err := base85Decode(buf[:byteCount], []byte(seq)); err != nil {
+			return p.Errorf(0, "binary patch: %v", err)
+		}
+		data.Write(buf[:byteCount])
+
+		if err := p.Next(); err != nil {
+			if err == io.EOF {
+				return p.Errorf(0, "binary patch: unexpected EOF")
+			}
+			return err
+		}
+	}
+
+	if err := inflateBinaryChunk(frag, &data); err != nil {
+		return p.Errorf(0, "binary patch: %v", err)
+	}
+
+	// consume the empty line that ended the fragment
+	if err := p.Next(); err != nil && err != io.EOF {
+		return err
+	}
+	return nil
+}
+
+func inflateBinaryChunk(frag *BinaryFragment, r io.Reader) error {
+	zr, err := zlib.NewReader(r)
+	if err != nil {
+		return err
+	}
+
+	data, err := ioutil.ReadAll(zr)
+	if err != nil {
+		return err
+	}
+	if err := zr.Close(); err != nil {
+		return err
+	}
+
+	if int64(len(data)) != frag.Size {
+		return fmt.Errorf("%d byte fragment inflated to %d", frag.Size, len(data))
+	}
+	frag.Data = data
+	return nil
+}

gitdiff/parser_binary_test.go renamed to gitdiff/binary_test.go

@@ -13,6 +13,53 @@ const (
 	devNull = "/dev/null"
 )
 
+// ParseNextFileHeader finds and parses the next file header in the stream. If
+// a header is found, it returns a file and all input before the header. It
+// returns nil if no headers are found before the end of the input.
+func (p *parser) ParseNextFileHeader() (*File, string, error) {
+	var preamble strings.Builder
+	var file *File
+	for {
+		// check for disconnected fragment headers (corrupt patch)
+		frag, err := p.ParseTextFragmentHeader()
+		if err != nil {
+			// not a valid header, nothing to worry about
+			goto NextLine
+		}
+		if frag != nil {
+			return nil, "", p.Errorf(-1, "patch fragment without file header: %s", frag.Header())
+		}
+
+		// check for a git-generated patch
+		file, err = p.ParseGitFileHeader()
+		if err != nil {
+			return nil, "", err
+		}
+		if file != nil {
+			return file, preamble.String(), nil
+		}
+
+		// check for a "traditional" patch
+		file, err = p.ParseTraditionalFileHeader()
+		if err != nil {
+			return nil, "", err
+		}
+		if file != nil {
+			return file, preamble.String(), nil
+		}
+
+	NextLine:
+		preamble.WriteString(p.Line(0))
+		if err := p.Next(); err != nil {
+			if err == io.EOF {
+				break
+			}
+			return nil, "", err
+		}
+	}
+	return nil, "", nil
+}
+
 func (p *parser) ParseGitFileHeader() (*File, error) {
 	const prefix = "diff --git "