iedc-go/vendor/github.com/robertkrimen/otto/parser/regexp.go

359 lines
7.4 KiB
Go
Raw Permalink Normal View History

2023-03-02 19:12:15 +00:00
package parser
import (
"bytes"
"fmt"
"strconv"
)
type _RegExp_parser struct {
str string
length int
chr rune // The current character
chrOffset int // The offset of current character
offset int // The offset after current character (may be greater than 1)
errors []error
invalid bool // The input is an invalid JavaScript RegExp
goRegexp *bytes.Buffer
}
// TransformRegExp transforms a JavaScript pattern into a Go "regexp" pattern.
//
// re2 (Go) cannot do backtracking, so the presence of a lookahead (?=) (?!) or
// backreference (\1, \2, ...) will cause an error.
//
// re2 (Go) has a different definition for \s: [\t\n\f\r ].
// The JavaScript definition, on the other hand, also includes \v, Unicode "Separator, Space", etc.
//
// If the pattern is invalid (not valid even in JavaScript), then this function
// returns the empty string and an error.
//
// If the pattern is valid, but incompatible (contains a lookahead or backreference),
// then this function returns the transformation (a non-empty string) AND an error.
func TransformRegExp(pattern string) (string, error) {
if pattern == "" {
return "", nil
}
// TODO If without \, if without (?=, (?!, then another shortcut
parser := _RegExp_parser{
str: pattern,
length: len(pattern),
goRegexp: bytes.NewBuffer(make([]byte, 0, 3*len(pattern)/2)),
}
parser.read() // Pull in the first character
parser.scan()
var err error
if len(parser.errors) > 0 {
err = parser.errors[0]
}
if parser.invalid {
return "", err
}
// Might not be re2 compatible, but is still a valid JavaScript RegExp
return parser.goRegexp.String(), err
}
func (self *_RegExp_parser) scan() {
for self.chr != -1 {
switch self.chr {
case '\\':
self.read()
self.scanEscape(false)
case '(':
self.pass()
self.scanGroup()
case '[':
self.pass()
self.scanBracket()
case ')':
self.error(-1, "Unmatched ')'")
self.invalid = true
self.pass()
default:
self.pass()
}
}
}
// (...)
func (self *_RegExp_parser) scanGroup() {
str := self.str[self.chrOffset:]
if len(str) > 1 { // A possibility of (?= or (?!
if str[0] == '?' {
if str[1] == '=' || str[1] == '!' {
self.error(-1, "re2: Invalid (%s) <lookahead>", self.str[self.chrOffset:self.chrOffset+2])
}
}
}
for self.chr != -1 && self.chr != ')' {
switch self.chr {
case '\\':
self.read()
self.scanEscape(false)
case '(':
self.pass()
self.scanGroup()
case '[':
self.pass()
self.scanBracket()
default:
self.pass()
continue
}
}
if self.chr != ')' {
self.error(-1, "Unterminated group")
self.invalid = true
return
}
self.pass()
}
// [...]
func (self *_RegExp_parser) scanBracket() {
for self.chr != -1 {
if self.chr == ']' {
break
} else if self.chr == '\\' {
self.read()
self.scanEscape(true)
continue
}
self.pass()
}
if self.chr != ']' {
self.error(-1, "Unterminated character class")
self.invalid = true
return
}
self.pass()
}
// \...
func (self *_RegExp_parser) scanEscape(inClass bool) {
offset := self.chrOffset
var length, base uint32
switch self.chr {
case '0', '1', '2', '3', '4', '5', '6', '7':
var value int64
size := 0
for {
digit := int64(digitValue(self.chr))
if digit >= 8 {
// Not a valid digit
break
}
value = value*8 + digit
self.read()
size += 1
}
if size == 1 { // The number of characters read
_, err := self.goRegexp.Write([]byte{'\\', byte(value) + '0'})
if err != nil {
self.errors = append(self.errors, err)
}
if value != 0 {
// An invalid backreference
self.error(-1, "re2: Invalid \\%d <backreference>", value)
}
return
}
tmp := []byte{'\\', 'x', '0', 0}
if value >= 16 {
tmp = tmp[0:2]
} else {
tmp = tmp[0:3]
}
tmp = strconv.AppendInt(tmp, value, 16)
_, err := self.goRegexp.Write(tmp)
if err != nil {
self.errors = append(self.errors, err)
}
return
case '8', '9':
size := 0
for {
digit := digitValue(self.chr)
if digit >= 10 {
// Not a valid digit
break
}
self.read()
size += 1
}
err := self.goRegexp.WriteByte('\\')
if err != nil {
self.errors = append(self.errors, err)
}
_, err = self.goRegexp.WriteString(self.str[offset:self.chrOffset])
if err != nil {
self.errors = append(self.errors, err)
}
self.error(-1, "re2: Invalid \\%s <backreference>", self.str[offset:self.chrOffset])
return
case 'x':
self.read()
length, base = 2, 16
case 'u':
self.read()
length, base = 4, 16
case 'b':
if inClass {
_, err := self.goRegexp.Write([]byte{'\\', 'x', '0', '8'})
if err != nil {
self.errors = append(self.errors, err)
}
self.read()
return
}
fallthrough
case 'B':
fallthrough
case 'd', 'D', 's', 'S', 'w', 'W':
// This is slightly broken, because ECMAScript
// includes \v in \s, \S, while re2 does not
fallthrough
case '\\':
fallthrough
case 'f', 'n', 'r', 't', 'v':
err := self.goRegexp.WriteByte('\\')
if err != nil {
self.errors = append(self.errors, err)
}
self.pass()
return
case 'c':
self.read()
var value int64
if 'a' <= self.chr && self.chr <= 'z' {
value = int64(self.chr) - 'a' + 1
} else if 'A' <= self.chr && self.chr <= 'Z' {
value = int64(self.chr) - 'A' + 1
} else {
err := self.goRegexp.WriteByte('c')
if err != nil {
self.errors = append(self.errors, err)
}
return
}
tmp := []byte{'\\', 'x', '0', 0}
if value >= 16 {
tmp = tmp[0:2]
} else {
tmp = tmp[0:3]
}
tmp = strconv.AppendInt(tmp, value, 16)
_, err := self.goRegexp.Write(tmp)
if err != nil {
self.errors = append(self.errors, err)
}
self.read()
return
default:
// $ is an identifier character, so we have to have
// a special case for it here
if self.chr == '$' || !isIdentifierPart(self.chr) {
// A non-identifier character needs escaping
err := self.goRegexp.WriteByte('\\')
if err != nil {
self.errors = append(self.errors, err)
}
} else {
// Unescape the character for re2
}
self.pass()
return
}
// Otherwise, we're a \u.... or \x...
valueOffset := self.chrOffset
var value uint32
{
length := length
for ; length > 0; length-- {
digit := uint32(digitValue(self.chr))
if digit >= base {
// Not a valid digit
goto skip
}
value = value*base + digit
self.read()
}
}
if length == 4 {
_, err := self.goRegexp.Write([]byte{
'\\',
'x',
'{',
self.str[valueOffset+0],
self.str[valueOffset+1],
self.str[valueOffset+2],
self.str[valueOffset+3],
'}',
})
if err != nil {
self.errors = append(self.errors, err)
}
} else if length == 2 {
_, err := self.goRegexp.Write([]byte{
'\\',
'x',
self.str[valueOffset+0],
self.str[valueOffset+1],
})
if err != nil {
self.errors = append(self.errors, err)
}
} else {
// Should never, ever get here...
self.error(-1, "re2: Illegal branch in scanEscape")
goto skip
}
return
skip:
_, err := self.goRegexp.WriteString(self.str[offset:self.chrOffset])
if err != nil {
self.errors = append(self.errors, err)
}
}
func (self *_RegExp_parser) pass() {
if self.chr != -1 {
_, err := self.goRegexp.WriteRune(self.chr)
if err != nil {
self.errors = append(self.errors, err)
}
}
self.read()
}
// TODO Better error reporting, use the offset, etc.
func (self *_RegExp_parser) error(offset int, msg string, msgValues ...interface{}) error {
err := fmt.Errorf(msg, msgValues...)
self.errors = append(self.errors, err)
return err
}