package fsnoder
import (
"bytes"
"fmt"
"io"
"srcd.works/go-git.v4/utils/merkletrie/noder"
)
// New function creates a full merkle trie from the string description of
// a filesystem tree. See examples of the string format in the package
// description.
func New(s string) (noder.Noder, error) {
return decodeDir([]byte(s), root)
}
const (
root = true
nonRoot = false
)
// Expected data: a fsnoder description, for example: A(foo bar qux ...).
// When isRoot is true, unnamed dirs are supported, for example: (foo
// bar qux ...)
func decodeDir(data []byte, isRoot bool) (*dir, error) {
data = bytes.TrimSpace(data)
if len(data) == 0 {
return nil, io.EOF
}
// get the name of the dir (a single letter) and remove it from the
// data. In case the there is no name and isRoot is true, just use
// "" as the name.
var name string
if data[0] == dirStartMark {
if isRoot {
name = ""
} else {
return nil, fmt.Errorf("inner unnamed dirs not allowed: %s", data)
}
} else {
name = string(data[0])
data = data[1:]
}
// check that data is enclosed in parents and it is big enough and
// remove them.
if len(data) < 2 {
return nil, fmt.Errorf("malformed data: too short")
}
if data[0] != dirStartMark {
return nil, fmt.Errorf("malformed data: first %q not found",
dirStartMark)
}
if data[len(data)-1] != dirEndMark {
return nil, fmt.Errorf("malformed data: last %q not found",
dirEndMark)
}
data = data[1 : len(data)-1] // remove initial '(' and last ')'
children, err := decodeChildren(data)
if err != nil {
return nil, err
}
return newDir(name, children)
}
func isNumber(b byte) bool {
return '0' <= b && b <= '9'
}
func isLetter(b byte) bool {
return ('a' <= b && b <= 'z') || ('A' <= b && b <= 'Z')
}
func decodeChildren(data []byte) ([]noder.Noder, error) {
data = bytes.TrimSpace(data)
if len(data) == 0 {
return nil, nil
}
chunks := split(data)
ret := make([]noder.Noder, len(chunks))
var err error
for i, c := range chunks {
ret[i], err = decodeChild(c)
if err != nil {
return nil, fmt.Errorf("malformed element %d (%s): %s", i, c, err)
}
}
return ret, nil
}
// returns the description of the elements of a dir. It is just looking
// for spaces if they are not part of inner dirs.
func split(data []byte) [][]byte {
chunks := [][]byte{}
start := 0
dirDepth := 0
for i, b := range data {
switch b {
case dirStartMark:
dirDepth++
case dirEndMark:
dirDepth--
case dirElementSep:
if dirDepth == 0 {
chunks = append(chunks, data[start:i+1])
start = i + 1
}
}
}
chunks = append(chunks, data[start:])
return chunks
}
// A child can be a file or a dir.
func decodeChild(data []byte) (noder.Noder, error) {
clean := bytes.TrimSpace(data)
if len(data) < 3 {
return nil, fmt.Errorf("element too short: %s", clean)
}
switch clean[1] {
case fileStartMark:
return decodeFile(clean)
case dirStartMark:
return decodeDir(clean, nonRoot)
default:
if clean[0] == dirStartMark {
return nil, fmt.Errorf("non-root unnamed dir are not allowed: %s",
clean)
}
return nil, fmt.Errorf("malformed dir element: %s", clean)
}
}
func decodeFile(data []byte) (noder.Noder, error) {
if len(data) == 3 {
return decodeEmptyFile(data)
}
if len(data) != 4 {
return nil, fmt.Errorf("length is not 4")
}
if !isLetter(data[0]) {
return nil, fmt.Errorf("name must be a letter")
}
if data[1] != '<' {
return nil, fmt.Errorf("wrong file start character")
}
if !isNumber(data[2]) {
return nil, fmt.Errorf("contents must be a number")
}
if data[3] != '>' {
return nil, fmt.Errorf("wrong file end character")
}
name := string(data[0])
contents := string(data[2])
return newFile(name, contents)
}
func decodeEmptyFile(data []byte) (noder.Noder, error) {
if len(data) != 3 {
return nil, fmt.Errorf("length is not 3: %s", data)
}
if !isLetter(data[0]) {
return nil, fmt.Errorf("name must be a letter: %s", data)
}
if data[1] != '<' {
return nil, fmt.Errorf("wrong file start character: %s", data)
}
if data[2] != '>' {
return nil, fmt.Errorf("wrong file end character: %s", data)
}
name := string(data[0])
return newFile(name, "")
}
// HashEqual returns if a and b have the same hash.
func HashEqual(a, b noder.Hasher) bool {
return bytes.Equal(a.Hash(), b.Hash())
}