mirror of
https://github.com/superseriousbusiness/gotosocial.git
synced 2024-11-23 20:26:39 +00:00
543 lines
15 KiB
Go
543 lines
15 KiB
Go
// Package parse contains a collection of parsers for various formats in its subpackages.
|
|
package parse
|
|
|
|
import (
|
|
"bytes"
|
|
"encoding/base64"
|
|
"errors"
|
|
"strconv"
|
|
)
|
|
|
|
var (
|
|
dataSchemeBytes = []byte("data:")
|
|
base64Bytes = []byte("base64")
|
|
textMimeBytes = []byte("text/plain")
|
|
)
|
|
|
|
// ErrBadDataURI is returned by DataURI when the byte slice does not start with 'data:' or is too short.
|
|
var ErrBadDataURI = errors.New("not a data URI")
|
|
|
|
// Number returns the number of bytes that parse as a number of the regex format (+|-)?([0-9]+(\.[0-9]+)?|\.[0-9]+)((e|E)(+|-)?[0-9]+)?.
|
|
func Number(b []byte) int {
|
|
if len(b) == 0 {
|
|
return 0
|
|
}
|
|
i := 0
|
|
if b[i] == '+' || b[i] == '-' {
|
|
i++
|
|
if i >= len(b) {
|
|
return 0
|
|
}
|
|
}
|
|
firstDigit := (b[i] >= '0' && b[i] <= '9')
|
|
if firstDigit {
|
|
i++
|
|
for i < len(b) && b[i] >= '0' && b[i] <= '9' {
|
|
i++
|
|
}
|
|
}
|
|
if i < len(b) && b[i] == '.' {
|
|
i++
|
|
if i < len(b) && b[i] >= '0' && b[i] <= '9' {
|
|
i++
|
|
for i < len(b) && b[i] >= '0' && b[i] <= '9' {
|
|
i++
|
|
}
|
|
} else if firstDigit {
|
|
// . could belong to the next token
|
|
i--
|
|
return i
|
|
} else {
|
|
return 0
|
|
}
|
|
} else if !firstDigit {
|
|
return 0
|
|
}
|
|
iOld := i
|
|
if i < len(b) && (b[i] == 'e' || b[i] == 'E') {
|
|
i++
|
|
if i < len(b) && (b[i] == '+' || b[i] == '-') {
|
|
i++
|
|
}
|
|
if i >= len(b) || b[i] < '0' || b[i] > '9' {
|
|
// e could belong to next token
|
|
return iOld
|
|
}
|
|
for i < len(b) && b[i] >= '0' && b[i] <= '9' {
|
|
i++
|
|
}
|
|
}
|
|
return i
|
|
}
|
|
|
|
// Dimension parses a byte-slice and returns the length of the number and its unit.
|
|
func Dimension(b []byte) (int, int) {
|
|
num := Number(b)
|
|
if num == 0 || num == len(b) {
|
|
return num, 0
|
|
} else if b[num] == '%' {
|
|
return num, 1
|
|
} else if b[num] >= 'a' && b[num] <= 'z' || b[num] >= 'A' && b[num] <= 'Z' {
|
|
i := num + 1
|
|
for i < len(b) && (b[i] >= 'a' && b[i] <= 'z' || b[i] >= 'A' && b[i] <= 'Z') {
|
|
i++
|
|
}
|
|
return num, i - num
|
|
}
|
|
return num, 0
|
|
}
|
|
|
|
// Mediatype parses a given mediatype and splits the mimetype from the parameters.
|
|
// It works similar to mime.ParseMediaType but is faster.
|
|
func Mediatype(b []byte) ([]byte, map[string]string) {
|
|
i := 0
|
|
for i < len(b) && b[i] == ' ' {
|
|
i++
|
|
}
|
|
b = b[i:]
|
|
n := len(b)
|
|
mimetype := b
|
|
var params map[string]string
|
|
for i := 3; i < n; i++ { // mimetype is at least three characters long
|
|
if b[i] == ';' || b[i] == ' ' {
|
|
mimetype = b[:i]
|
|
if b[i] == ' ' {
|
|
i++ // space
|
|
for i < n && b[i] == ' ' {
|
|
i++
|
|
}
|
|
if n <= i || b[i] != ';' {
|
|
break
|
|
}
|
|
}
|
|
params = map[string]string{}
|
|
s := string(b)
|
|
PARAM:
|
|
i++ // semicolon
|
|
for i < n && s[i] == ' ' {
|
|
i++
|
|
}
|
|
start := i
|
|
for i < n && s[i] != '=' && s[i] != ';' && s[i] != ' ' {
|
|
i++
|
|
}
|
|
key := s[start:i]
|
|
for i < n && s[i] == ' ' {
|
|
i++
|
|
}
|
|
if i < n && s[i] == '=' {
|
|
i++
|
|
for i < n && s[i] == ' ' {
|
|
i++
|
|
}
|
|
start = i
|
|
for i < n && s[i] != ';' && s[i] != ' ' {
|
|
i++
|
|
}
|
|
} else {
|
|
start = i
|
|
}
|
|
params[key] = s[start:i]
|
|
for i < n && s[i] == ' ' {
|
|
i++
|
|
}
|
|
if i < n && s[i] == ';' {
|
|
goto PARAM
|
|
}
|
|
break
|
|
}
|
|
}
|
|
return mimetype, params
|
|
}
|
|
|
|
// DataURI parses the given data URI and returns the mediatype, data and ok.
|
|
func DataURI(dataURI []byte) ([]byte, []byte, error) {
|
|
if len(dataURI) > 5 && bytes.Equal(dataURI[:5], dataSchemeBytes) {
|
|
dataURI = dataURI[5:]
|
|
inBase64 := false
|
|
var mediatype []byte
|
|
i := 0
|
|
for j := 0; j < len(dataURI); j++ {
|
|
c := dataURI[j]
|
|
if c == '=' || c == ';' || c == ',' {
|
|
if c != '=' && bytes.Equal(TrimWhitespace(dataURI[i:j]), base64Bytes) {
|
|
if len(mediatype) > 0 {
|
|
mediatype = mediatype[:len(mediatype)-1]
|
|
}
|
|
inBase64 = true
|
|
i = j
|
|
} else if c != ',' {
|
|
mediatype = append(append(mediatype, TrimWhitespace(dataURI[i:j])...), c)
|
|
i = j + 1
|
|
} else {
|
|
mediatype = append(mediatype, TrimWhitespace(dataURI[i:j])...)
|
|
}
|
|
if c == ',' {
|
|
if len(mediatype) == 0 || mediatype[0] == ';' {
|
|
mediatype = textMimeBytes
|
|
}
|
|
data := dataURI[j+1:]
|
|
if inBase64 {
|
|
decoded := make([]byte, base64.StdEncoding.DecodedLen(len(data)))
|
|
n, err := base64.StdEncoding.Decode(decoded, data)
|
|
if err != nil {
|
|
return nil, nil, err
|
|
}
|
|
data = decoded[:n]
|
|
} else {
|
|
data = DecodeURL(data)
|
|
}
|
|
return mediatype, data, nil
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return nil, nil, ErrBadDataURI
|
|
}
|
|
|
|
// QuoteEntity parses the given byte slice and returns the quote that got matched (' or ") and its entity length.
|
|
// TODO: deprecated
|
|
func QuoteEntity(b []byte) (quote byte, n int) {
|
|
if len(b) < 5 || b[0] != '&' {
|
|
return 0, 0
|
|
}
|
|
if b[1] == '#' {
|
|
if b[2] == 'x' {
|
|
i := 3
|
|
for i < len(b) && b[i] == '0' {
|
|
i++
|
|
}
|
|
if i+2 < len(b) && b[i] == '2' && b[i+2] == ';' {
|
|
if b[i+1] == '2' {
|
|
return '"', i + 3 // "
|
|
} else if b[i+1] == '7' {
|
|
return '\'', i + 3 // '
|
|
}
|
|
}
|
|
} else {
|
|
i := 2
|
|
for i < len(b) && b[i] == '0' {
|
|
i++
|
|
}
|
|
if i+2 < len(b) && b[i] == '3' && b[i+2] == ';' {
|
|
if b[i+1] == '4' {
|
|
return '"', i + 3 // "
|
|
} else if b[i+1] == '9' {
|
|
return '\'', i + 3 // '
|
|
}
|
|
}
|
|
}
|
|
} else if len(b) >= 6 && b[5] == ';' {
|
|
if bytes.Equal(b[1:5], []byte{'q', 'u', 'o', 't'}) {
|
|
return '"', 6 // "
|
|
} else if bytes.Equal(b[1:5], []byte{'a', 'p', 'o', 's'}) {
|
|
return '\'', 6 // '
|
|
}
|
|
}
|
|
return 0, 0
|
|
}
|
|
|
|
// ReplaceMultipleWhitespace replaces character series of space, \n, \t, \f, \r into a single space or newline (when the serie contained a \n or \r).
|
|
func ReplaceMultipleWhitespace(b []byte) []byte {
|
|
j, k := 0, 0 // j is write position, k is start of next text section
|
|
for i := 0; i < len(b); i++ {
|
|
if IsWhitespace(b[i]) {
|
|
start := i
|
|
newline := IsNewline(b[i])
|
|
i++
|
|
for ; i < len(b) && IsWhitespace(b[i]); i++ {
|
|
if IsNewline(b[i]) {
|
|
newline = true
|
|
}
|
|
}
|
|
if newline {
|
|
b[start] = '\n'
|
|
} else {
|
|
b[start] = ' '
|
|
}
|
|
if 1 < i-start { // more than one whitespace
|
|
if j == 0 {
|
|
j = start + 1
|
|
} else {
|
|
j += copy(b[j:], b[k:start+1])
|
|
}
|
|
k = i
|
|
}
|
|
}
|
|
}
|
|
if j == 0 {
|
|
return b
|
|
} else if j == 1 { // only if starts with whitespace
|
|
b[k-1] = b[0]
|
|
return b[k-1:]
|
|
} else if k < len(b) {
|
|
j += copy(b[j:], b[k:])
|
|
}
|
|
return b[:j]
|
|
}
|
|
|
|
// replaceEntities will replace in b at index i, assuming that b[i] == '&' and that i+3<len(b). The returned int will be the last character of the entity, so that the next iteration can safely do i++ to continue and not miss any entitites.
|
|
func replaceEntities(b []byte, i int, entitiesMap map[string][]byte, revEntitiesMap map[byte][]byte) ([]byte, int) {
|
|
const MaxEntityLength = 31 // longest HTML entity: CounterClockwiseContourIntegral
|
|
var r []byte
|
|
j := i + 1
|
|
if b[j] == '#' {
|
|
j++
|
|
if b[j] == 'x' {
|
|
j++
|
|
c := 0
|
|
for ; j < len(b) && (b[j] >= '0' && b[j] <= '9' || b[j] >= 'a' && b[j] <= 'f' || b[j] >= 'A' && b[j] <= 'F'); j++ {
|
|
if b[j] <= '9' {
|
|
c = c<<4 + int(b[j]-'0')
|
|
} else if b[j] <= 'F' {
|
|
c = c<<4 + int(b[j]-'A') + 10
|
|
} else if b[j] <= 'f' {
|
|
c = c<<4 + int(b[j]-'a') + 10
|
|
}
|
|
}
|
|
if j <= i+3 || 10000 <= c {
|
|
return b, j - 1
|
|
}
|
|
if c < 128 {
|
|
r = []byte{byte(c)}
|
|
} else {
|
|
r = append(r, '&', '#')
|
|
r = strconv.AppendInt(r, int64(c), 10)
|
|
r = append(r, ';')
|
|
}
|
|
} else {
|
|
c := 0
|
|
for ; j < len(b) && c < 128 && b[j] >= '0' && b[j] <= '9'; j++ {
|
|
c = c*10 + int(b[j]-'0')
|
|
}
|
|
if j <= i+2 || 128 <= c {
|
|
return b, j - 1
|
|
}
|
|
r = []byte{byte(c)}
|
|
}
|
|
} else {
|
|
for ; j < len(b) && j-i-1 <= MaxEntityLength && b[j] != ';'; j++ {
|
|
}
|
|
if j <= i+1 || len(b) <= j {
|
|
return b, j - 1
|
|
}
|
|
|
|
var ok bool
|
|
r, ok = entitiesMap[string(b[i+1:j])]
|
|
if !ok {
|
|
return b, j
|
|
}
|
|
}
|
|
|
|
// j is at semicolon
|
|
n := j + 1 - i
|
|
if j < len(b) && b[j] == ';' && 2 < n {
|
|
if len(r) == 1 {
|
|
if q, ok := revEntitiesMap[r[0]]; ok {
|
|
if len(q) == len(b[i:j+1]) && bytes.Equal(q, b[i:j+1]) {
|
|
return b, j
|
|
}
|
|
r = q
|
|
} else if r[0] == '&' {
|
|
// check if for example & is followed by something that could potentially be an entity
|
|
k := j + 1
|
|
if k < len(b) && (b[k] >= '0' && b[k] <= '9' || b[k] >= 'a' && b[k] <= 'z' || b[k] >= 'A' && b[k] <= 'Z' || b[k] == '#') {
|
|
return b, k
|
|
}
|
|
}
|
|
}
|
|
|
|
copy(b[i:], r)
|
|
copy(b[i+len(r):], b[j+1:])
|
|
b = b[:len(b)-n+len(r)]
|
|
return b, i + len(r) - 1
|
|
}
|
|
return b, i
|
|
}
|
|
|
|
// ReplaceEntities replaces all occurrences of entites (such as ") to their respective unencoded bytes.
|
|
func ReplaceEntities(b []byte, entitiesMap map[string][]byte, revEntitiesMap map[byte][]byte) []byte {
|
|
for i := 0; i < len(b); i++ {
|
|
if b[i] == '&' && i+3 < len(b) {
|
|
b, i = replaceEntities(b, i, entitiesMap, revEntitiesMap)
|
|
}
|
|
}
|
|
return b
|
|
}
|
|
|
|
// ReplaceMultipleWhitespaceAndEntities is a combination of ReplaceMultipleWhitespace and ReplaceEntities. It is faster than executing both sequentially.
|
|
func ReplaceMultipleWhitespaceAndEntities(b []byte, entitiesMap map[string][]byte, revEntitiesMap map[byte][]byte) []byte {
|
|
j, k := 0, 0 // j is write position, k is start of next text section
|
|
for i := 0; i < len(b); i++ {
|
|
if IsWhitespace(b[i]) {
|
|
start := i
|
|
newline := IsNewline(b[i])
|
|
i++
|
|
for ; i < len(b) && IsWhitespace(b[i]); i++ {
|
|
if IsNewline(b[i]) {
|
|
newline = true
|
|
}
|
|
}
|
|
if newline {
|
|
b[start] = '\n'
|
|
} else {
|
|
b[start] = ' '
|
|
}
|
|
if 1 < i-start { // more than one whitespace
|
|
if j == 0 {
|
|
j = start + 1
|
|
} else {
|
|
j += copy(b[j:], b[k:start+1])
|
|
}
|
|
k = i
|
|
}
|
|
}
|
|
if i+3 < len(b) && b[i] == '&' {
|
|
b, i = replaceEntities(b, i, entitiesMap, revEntitiesMap)
|
|
}
|
|
}
|
|
if j == 0 {
|
|
return b
|
|
} else if j == 1 { // only if starts with whitespace
|
|
b[k-1] = b[0]
|
|
return b[k-1:]
|
|
} else if k < len(b) {
|
|
j += copy(b[j:], b[k:])
|
|
}
|
|
return b[:j]
|
|
}
|
|
|
|
// URLEncodingTable is a charmap for which characters need escaping in the URL encoding scheme
|
|
var URLEncodingTable = [256]bool{
|
|
// ASCII
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, false, true, true, true, true, true, false, // space, ", #, $, %, &
|
|
false, false, false, true, true, false, false, true, // +, comma, /
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, true, true, true, true, true, true, // :, ;, <, =, >, ?
|
|
|
|
true, false, false, false, false, false, false, false, // @
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, true, true, true, true, false, // [, \, ], ^
|
|
|
|
true, false, false, false, false, false, false, false, // `
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, true, true, true, false, true, // {, |, }, DEL
|
|
|
|
// non-ASCII
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
}
|
|
|
|
// DataURIEncodingTable is a charmap for which characters need escaping in the Data URI encoding scheme
|
|
// Escape only non-printable characters, unicode and %, #, &.
|
|
// IE11 additionally requires encoding of \, [, ], ", <, >, `, {, }, |, ^ which is not required by Chrome, Firefox, Opera, Edge, Safari, Yandex
|
|
// To pass the HTML validator, restricted URL characters must be escaped: non-printable characters, space, <, >, #, %, "
|
|
var DataURIEncodingTable = [256]bool{
|
|
// ASCII
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, false, true, true, false, true, true, false, // space, ", #, %, &
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, false, true, false, true, false, // <, >
|
|
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, true, true, true, true, false, // [, \, ], ^
|
|
|
|
true, false, false, false, false, false, false, false, // `
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, false, false, false, false, false,
|
|
false, false, false, true, true, true, false, true, // {, |, }, DEL
|
|
|
|
// non-ASCII
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
true, true, true, true, true, true, true, true,
|
|
}
|
|
|
|
// EncodeURL encodes bytes using the URL encoding scheme
|
|
func EncodeURL(b []byte, table [256]bool) []byte {
|
|
for i := 0; i < len(b); i++ {
|
|
c := b[i]
|
|
if table[c] {
|
|
b = append(b, 0, 0)
|
|
copy(b[i+3:], b[i+1:])
|
|
b[i+0] = '%'
|
|
b[i+1] = "0123456789ABCDEF"[c>>4]
|
|
b[i+2] = "0123456789ABCDEF"[c&15]
|
|
}
|
|
}
|
|
return b
|
|
}
|
|
|
|
// DecodeURL decodes an URL encoded using the URL encoding scheme
|
|
func DecodeURL(b []byte) []byte {
|
|
for i := 0; i < len(b); i++ {
|
|
if b[i] == '%' && i+2 < len(b) {
|
|
j := i + 1
|
|
c := 0
|
|
for ; j < i+3 && (b[j] >= '0' && b[j] <= '9' || b[j] >= 'a' && b[j] <= 'f' || b[j] >= 'A' && b[j] <= 'F'); j++ {
|
|
if b[j] <= '9' {
|
|
c = c<<4 + int(b[j]-'0')
|
|
} else if b[j] <= 'F' {
|
|
c = c<<4 + int(b[j]-'A') + 10
|
|
} else if b[j] <= 'f' {
|
|
c = c<<4 + int(b[j]-'a') + 10
|
|
}
|
|
}
|
|
if j == i+3 && c < 128 {
|
|
b[i] = byte(c)
|
|
b = append(b[:i+1], b[i+3:]...)
|
|
}
|
|
} else if b[i] == '+' {
|
|
b[i] = ' '
|
|
}
|
|
}
|
|
return b
|
|
}
|