mirror of
https://github.com/go-gitea/gitea
synced 2024-11-09 11:44:27 +00:00
274 lines
4.4 KiB
Go
274 lines
4.4 KiB
Go
|
package lexer
|
||
|
|
||
|
import (
|
||
|
"bytes"
|
||
|
"fmt"
|
||
|
"github.com/gobwas/glob/util/runes"
|
||
|
"unicode/utf8"
|
||
|
)
|
||
|
|
||
|
const (
|
||
|
char_any = '*'
|
||
|
char_comma = ','
|
||
|
char_single = '?'
|
||
|
char_escape = '\\'
|
||
|
char_range_open = '['
|
||
|
char_range_close = ']'
|
||
|
char_terms_open = '{'
|
||
|
char_terms_close = '}'
|
||
|
char_range_not = '!'
|
||
|
char_range_between = '-'
|
||
|
)
|
||
|
|
||
|
var specials = []byte{
|
||
|
char_any,
|
||
|
char_single,
|
||
|
char_escape,
|
||
|
char_range_open,
|
||
|
char_range_close,
|
||
|
char_terms_open,
|
||
|
char_terms_close,
|
||
|
}
|
||
|
|
||
|
func Special(c byte) bool {
|
||
|
return bytes.IndexByte(specials, c) != -1
|
||
|
}
|
||
|
|
||
|
type tokens []Token
|
||
|
|
||
|
func (i *tokens) shift() (ret Token) {
|
||
|
ret = (*i)[0]
|
||
|
copy(*i, (*i)[1:])
|
||
|
*i = (*i)[:len(*i)-1]
|
||
|
return
|
||
|
}
|
||
|
|
||
|
func (i *tokens) push(v Token) {
|
||
|
*i = append(*i, v)
|
||
|
}
|
||
|
|
||
|
func (i *tokens) empty() bool {
|
||
|
return len(*i) == 0
|
||
|
}
|
||
|
|
||
|
var eof rune = 0
|
||
|
|
||
|
type lexer struct {
|
||
|
data string
|
||
|
pos int
|
||
|
err error
|
||
|
|
||
|
tokens tokens
|
||
|
termsLevel int
|
||
|
|
||
|
lastRune rune
|
||
|
lastRuneSize int
|
||
|
hasRune bool
|
||
|
}
|
||
|
|
||
|
func NewLexer(source string) *lexer {
|
||
|
l := &lexer{
|
||
|
data: source,
|
||
|
tokens: tokens(make([]Token, 0, 4)),
|
||
|
}
|
||
|
return l
|
||
|
}
|
||
|
|
||
|
func (l *lexer) Next() Token {
|
||
|
if l.err != nil {
|
||
|
return Token{Error, l.err.Error()}
|
||
|
}
|
||
|
if !l.tokens.empty() {
|
||
|
return l.tokens.shift()
|
||
|
}
|
||
|
|
||
|
l.fetchItem()
|
||
|
return l.Next()
|
||
|
}
|
||
|
|
||
|
func (l *lexer) peek() (r rune, w int) {
|
||
|
if l.pos == len(l.data) {
|
||
|
return eof, 0
|
||
|
}
|
||
|
|
||
|
r, w = utf8.DecodeRuneInString(l.data[l.pos:])
|
||
|
if r == utf8.RuneError {
|
||
|
l.errorf("could not read rune")
|
||
|
r = eof
|
||
|
w = 0
|
||
|
}
|
||
|
|
||
|
return
|
||
|
}
|
||
|
|
||
|
func (l *lexer) read() rune {
|
||
|
if l.hasRune {
|
||
|
l.hasRune = false
|
||
|
l.seek(l.lastRuneSize)
|
||
|
return l.lastRune
|
||
|
}
|
||
|
|
||
|
r, s := l.peek()
|
||
|
l.seek(s)
|
||
|
|
||
|
l.lastRune = r
|
||
|
l.lastRuneSize = s
|
||
|
|
||
|
return r
|
||
|
}
|
||
|
|
||
|
func (l *lexer) seek(w int) {
|
||
|
l.pos += w
|
||
|
}
|
||
|
|
||
|
func (l *lexer) unread() {
|
||
|
if l.hasRune {
|
||
|
l.errorf("could not unread rune")
|
||
|
return
|
||
|
}
|
||
|
l.seek(-l.lastRuneSize)
|
||
|
l.hasRune = true
|
||
|
}
|
||
|
|
||
|
func (l *lexer) errorf(f string, v ...interface{}) {
|
||
|
l.err = fmt.Errorf(f, v...)
|
||
|
}
|
||
|
|
||
|
func (l *lexer) inTerms() bool {
|
||
|
return l.termsLevel > 0
|
||
|
}
|
||
|
|
||
|
func (l *lexer) termsEnter() {
|
||
|
l.termsLevel++
|
||
|
}
|
||
|
|
||
|
func (l *lexer) termsLeave() {
|
||
|
l.termsLevel--
|
||
|
}
|
||
|
|
||
|
var inTextBreakers = []rune{char_single, char_any, char_range_open, char_terms_open}
|
||
|
var inTermsBreakers = append(inTextBreakers, char_terms_close, char_comma)
|
||
|
|
||
|
func (l *lexer) fetchItem() {
|
||
|
r := l.read()
|
||
|
switch {
|
||
|
case r == eof:
|
||
|
l.tokens.push(Token{EOF, ""})
|
||
|
|
||
|
case r == char_terms_open:
|
||
|
l.termsEnter()
|
||
|
l.tokens.push(Token{TermsOpen, string(r)})
|
||
|
|
||
|
case r == char_comma && l.inTerms():
|
||
|
l.tokens.push(Token{Separator, string(r)})
|
||
|
|
||
|
case r == char_terms_close && l.inTerms():
|
||
|
l.tokens.push(Token{TermsClose, string(r)})
|
||
|
l.termsLeave()
|
||
|
|
||
|
case r == char_range_open:
|
||
|
l.tokens.push(Token{RangeOpen, string(r)})
|
||
|
l.fetchRange()
|
||
|
|
||
|
case r == char_single:
|
||
|
l.tokens.push(Token{Single, string(r)})
|
||
|
|
||
|
case r == char_any:
|
||
|
if l.read() == char_any {
|
||
|
l.tokens.push(Token{Super, string(r) + string(r)})
|
||
|
} else {
|
||
|
l.unread()
|
||
|
l.tokens.push(Token{Any, string(r)})
|
||
|
}
|
||
|
|
||
|
default:
|
||
|
l.unread()
|
||
|
|
||
|
var breakers []rune
|
||
|
if l.inTerms() {
|
||
|
breakers = inTermsBreakers
|
||
|
} else {
|
||
|
breakers = inTextBreakers
|
||
|
}
|
||
|
l.fetchText(breakers)
|
||
|
}
|
||
|
}
|
||
|
|
||
|
func (l *lexer) fetchRange() {
|
||
|
var wantHi bool
|
||
|
var wantClose bool
|
||
|
var seenNot bool
|
||
|
for {
|
||
|
r := l.read()
|
||
|
if r == eof {
|
||
|
l.errorf("unexpected end of input")
|
||
|
return
|
||
|
}
|
||
|
|
||
|
if wantClose {
|
||
|
if r != char_range_close {
|
||
|
l.errorf("expected close range character")
|
||
|
} else {
|
||
|
l.tokens.push(Token{RangeClose, string(r)})
|
||
|
}
|
||
|
return
|
||
|
}
|
||
|
|
||
|
if wantHi {
|
||
|
l.tokens.push(Token{RangeHi, string(r)})
|
||
|
wantClose = true
|
||
|
continue
|
||
|
}
|
||
|
|
||
|
if !seenNot && r == char_range_not {
|
||
|
l.tokens.push(Token{Not, string(r)})
|
||
|
seenNot = true
|
||
|
continue
|
||
|
}
|
||
|
|
||
|
if n, w := l.peek(); n == char_range_between {
|
||
|
l.seek(w)
|
||
|
l.tokens.push(Token{RangeLo, string(r)})
|
||
|
l.tokens.push(Token{RangeBetween, string(n)})
|
||
|
wantHi = true
|
||
|
continue
|
||
|
}
|
||
|
|
||
|
l.unread() // unread first peek and fetch as text
|
||
|
l.fetchText([]rune{char_range_close})
|
||
|
wantClose = true
|
||
|
}
|
||
|
}
|
||
|
|
||
|
func (l *lexer) fetchText(breakers []rune) {
|
||
|
var data []rune
|
||
|
var escaped bool
|
||
|
|
||
|
reading:
|
||
|
for {
|
||
|
r := l.read()
|
||
|
if r == eof {
|
||
|
break
|
||
|
}
|
||
|
|
||
|
if !escaped {
|
||
|
if r == char_escape {
|
||
|
escaped = true
|
||
|
continue
|
||
|
}
|
||
|
|
||
|
if runes.IndexRune(breakers, r) != -1 {
|
||
|
l.unread()
|
||
|
break reading
|
||
|
}
|
||
|
}
|
||
|
|
||
|
escaped = false
|
||
|
data = append(data, r)
|
||
|
}
|
||
|
|
||
|
if len(data) > 0 {
|
||
|
l.tokens.push(Token{Text, string(data)})
|
||
|
}
|
||
|
}
|