gotty/vendor/github.com/yudai/hcl/json/lex.go
2016-04-13 16:17:57 +09:00

257 lines
4.1 KiB
Go

package json
import (
"bytes"
"fmt"
"strconv"
"unicode"
"unicode/utf8"
)
//go:generate go tool yacc -p "json" parse.y
// This marks the end of the lexer
const lexEOF = 0
// The parser uses the type <prefix>Lex as a lexer. It must provide
// the methods Lex(*<prefix>SymType) int and Error(string).
type jsonLex struct {
Input string
pos int
width int
col, line int
err error
}
// The parser calls this method to get each new token.
func (x *jsonLex) Lex(yylval *jsonSymType) int {
for {
c := x.next()
if c == lexEOF {
return lexEOF
}
// Ignore all whitespace except a newline which we handle
// specially later.
if unicode.IsSpace(c) {
continue
}
// If it is a number, lex the number
if c >= '0' && c <= '9' {
x.backup()
return x.lexNumber(yylval)
}
switch c {
case 'e':
fallthrough
case 'E':
switch x.next() {
case '+':
return EPLUS
case '-':
return EMINUS
default:
x.backup()
return EPLUS
}
case '.':
return PERIOD
case '-':
return MINUS
case ':':
return COLON
case ',':
return COMMA
case '[':
return LEFTBRACKET
case ']':
return RIGHTBRACKET
case '{':
return LEFTBRACE
case '}':
return RIGHTBRACE
case '"':
return x.lexString(yylval)
default:
x.backup()
return x.lexId(yylval)
}
}
}
// lexId lexes an identifier
func (x *jsonLex) lexId(yylval *jsonSymType) int {
var b bytes.Buffer
first := true
for {
c := x.next()
if c == lexEOF {
break
}
if !unicode.IsDigit(c) && !unicode.IsLetter(c) && c != '_' && c != '-' {
x.backup()
if first {
x.createErr("Invalid identifier")
return lexEOF
}
break
}
first = false
if _, err := b.WriteRune(c); err != nil {
return lexEOF
}
}
switch v := b.String(); v {
case "true":
return TRUE
case "false":
return FALSE
case "null":
return NULL
default:
x.createErr(fmt.Sprintf("Invalid identifier: %s", v))
return lexEOF
}
}
// lexNumber lexes out a number
func (x *jsonLex) lexNumber(yylval *jsonSymType) int {
var b bytes.Buffer
gotPeriod := false
for {
c := x.next()
if c == lexEOF {
break
}
if c == '.' {
if gotPeriod {
x.backup()
break
}
gotPeriod = true
} else if c < '0' || c > '9' {
x.backup()
break
}
if _, err := b.WriteRune(c); err != nil {
x.createErr(fmt.Sprintf("Internal error: %s", err))
return lexEOF
}
}
if !gotPeriod {
v, err := strconv.ParseInt(b.String(), 0, 0)
if err != nil {
x.createErr(fmt.Sprintf("Expected number: %s", err))
return lexEOF
}
yylval.num = int(v)
return NUMBER
}
f, err := strconv.ParseFloat(b.String(), 64)
if err != nil {
x.createErr(fmt.Sprintf("Expected float: %s", err))
return lexEOF
}
yylval.f = float64(f)
return FLOAT
}
// lexString extracts a string from the input
func (x *jsonLex) lexString(yylval *jsonSymType) int {
var b bytes.Buffer
for {
c := x.next()
if c == lexEOF {
break
}
// String end
if c == '"' {
break
}
// If we're escaping a quote, then escape the quote
if c == '\\' {
n := x.next()
switch n {
case '"':
c = n
case 'n':
c = '\n'
case '\\':
c = n
default:
x.backup()
}
}
if _, err := b.WriteRune(c); err != nil {
return lexEOF
}
}
yylval.str = b.String()
return STRING
}
// Return the next rune for the lexer.
func (x *jsonLex) next() rune {
if int(x.pos) >= len(x.Input) {
x.width = 0
return lexEOF
}
r, w := utf8.DecodeRuneInString(x.Input[x.pos:])
x.width = w
x.pos += x.width
x.col += 1
if x.line == 0 {
x.line = 1
}
if r == '\n' {
x.line += 1
x.col = 0
}
return r
}
// peek returns but does not consume the next rune in the input
func (x *jsonLex) peek() rune {
r := x.next()
x.backup()
return r
}
// backup steps back one rune. Can only be called once per next.
func (x *jsonLex) backup() {
x.col -= 1
x.pos -= x.width
}
// createErr records the given error
func (x *jsonLex) createErr(msg string) {
x.err = fmt.Errorf("Line %d, column %d: %s", x.line, x.col, msg)
}
// The parser calls this method on a parse error.
func (x *jsonLex) Error(s string) {
x.createErr(s)
}