181 lines
3.5 KiB
Go
181 lines
3.5 KiB
Go
package lexpr
|
|
|
|
import (
|
|
"context"
|
|
"strings"
|
|
"unicode/utf8"
|
|
)
|
|
|
|
// EOF const.
|
|
const EOF rune = -1
|
|
|
|
// lex holds current scanner state.
|
|
type lex struct {
|
|
input string // Input string.
|
|
start int // Start position of current lexem.
|
|
pos int // Pos at input string.
|
|
output chan lexem // Lexems channel.
|
|
width int // Width of last rune.
|
|
}
|
|
|
|
// newLex returns new scanner for input string.
|
|
func newLex() *lex {
|
|
return &lex{
|
|
input: "",
|
|
start: 0,
|
|
pos: 0,
|
|
output: nil,
|
|
width: 0,
|
|
}
|
|
}
|
|
|
|
// parse input to lexems.
|
|
func (l *lex) parse(ctx context.Context, input string) <-chan lexem {
|
|
l.input = input
|
|
l.output = make(chan lexem)
|
|
go func() {
|
|
defer close(l.output)
|
|
for {
|
|
if ctx.Err() != nil {
|
|
return
|
|
}
|
|
switch {
|
|
case l.acceptWhile(" \n\t", false):
|
|
l.ignore()
|
|
case l.accept("("):
|
|
l.emit(lp)
|
|
case l.accept(")"):
|
|
l.emit(rp)
|
|
case l.accept(","):
|
|
l.emit(sep)
|
|
case scanNumber(l):
|
|
l.emit(number)
|
|
case scanOps(l):
|
|
l.emit(op)
|
|
case scanWord(l):
|
|
l.emit(word)
|
|
case scanQuotedString(l, `"`):
|
|
l.emit(str)
|
|
case l.peek() == EOF:
|
|
return
|
|
default:
|
|
l.emit(tokError)
|
|
return
|
|
}
|
|
}
|
|
}()
|
|
return l.output
|
|
}
|
|
|
|
// emit current lexem to output.
|
|
func (l *lex) emit(typ lexType) {
|
|
l.output <- lexem{
|
|
Type: typ,
|
|
Value: l.input[l.start:l.pos],
|
|
Start: l.start,
|
|
End: l.pos,
|
|
}
|
|
l.start = l.pos
|
|
}
|
|
|
|
// next rune from input.
|
|
func (l *lex) next() (r rune) {
|
|
if int(l.pos) >= len(l.input) {
|
|
l.width = 0
|
|
return EOF
|
|
}
|
|
r, l.width = utf8.DecodeRuneInString(l.input[l.pos:])
|
|
l.pos += l.width
|
|
return r
|
|
}
|
|
|
|
// back move position to previos rune.
|
|
func (l *lex) back() {
|
|
l.pos -= l.width
|
|
}
|
|
|
|
// ignore previosly buffered text.
|
|
func (l *lex) ignore() {
|
|
l.start = l.pos
|
|
l.width = 0
|
|
}
|
|
|
|
// peek rune at current position without moving position.
|
|
func (l *lex) peek() (r rune) {
|
|
r = l.next()
|
|
l.back()
|
|
return r
|
|
}
|
|
|
|
// accept any rune from valid string. Returns true if next rune was in valid string.
|
|
func (l *lex) accept(valid string) bool {
|
|
if strings.ContainsRune(valid, l.next()) {
|
|
return true
|
|
}
|
|
l.back()
|
|
return false
|
|
}
|
|
|
|
// acceptString returns true if given string was at position.
|
|
func (l *lex) acceptString(s string, caseInsentive bool) bool {
|
|
input := l.input
|
|
if caseInsentive {
|
|
input = strings.ToLower(input)
|
|
s = strings.ToLower(s)
|
|
}
|
|
if strings.HasPrefix(input, s) {
|
|
l.width = 0
|
|
l.pos += len(s)
|
|
return true
|
|
}
|
|
return false
|
|
}
|
|
|
|
// acceptAnyOf substrings. Retuns true if any of substrings was found.
|
|
func (l *lex) acceptAnyOf(s []string, caseInsentive bool) bool {
|
|
for _, substring := range s {
|
|
if l.acceptString(substring, caseInsentive) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
// acceptWhile passing symbols from input while they at `valid` string.
|
|
func (l *lex) acceptWhile(valid string, ignoreEscaped bool) bool {
|
|
start := l.pos
|
|
for {
|
|
ch := l.next()
|
|
switch {
|
|
case ch == EOF:
|
|
return false
|
|
case ch == '\\' && ignoreEscaped:
|
|
l.next()
|
|
case !strings.ContainsRune(valid, ch):
|
|
l.back()
|
|
return l.pos > start
|
|
}
|
|
}
|
|
}
|
|
|
|
// acceptWhileNot passing symbols from input while they NOT in `invalid` string.
|
|
func (l *lex) acceptWhileNot(invalid string, ignoreEscaped bool) bool {
|
|
start := l.pos
|
|
for {
|
|
ch := l.next()
|
|
switch {
|
|
case ch == EOF:
|
|
return false
|
|
case ch == '\\' && ignoreEscaped:
|
|
l.next()
|
|
case strings.ContainsRune(invalid, ch):
|
|
l.back()
|
|
return l.pos > start
|
|
}
|
|
}
|
|
}
|
|
|
|
// atStart returns true if current lexem not empty
|
|
func (l *lex) atStart() bool {
|
|
return l.pos == l.start
|
|
}
|