consul/vendor/github.com/hashicorp/hil/y.go

663 lines
15 KiB
Go
Raw Normal View History

//line lang.y:6
package hil
import __yyfmt__ "fmt"
//line lang.y:6
import (
"github.com/hashicorp/hil/ast"
)
//line lang.y:14
type parserSymType struct {
yys int
node ast.Node
nodeList []ast.Node
str string
token *parserToken
}
const PROGRAM_BRACKET_LEFT = 57346
const PROGRAM_BRACKET_RIGHT = 57347
const PROGRAM_STRING_START = 57348
const PROGRAM_STRING_END = 57349
const PAREN_LEFT = 57350
const PAREN_RIGHT = 57351
const COMMA = 57352
const SQUARE_BRACKET_LEFT = 57353
const SQUARE_BRACKET_RIGHT = 57354
const ARITH_OP = 57355
const IDENTIFIER = 57356
const INTEGER = 57357
const FLOAT = 57358
const STRING = 57359
var parserToknames = [...]string{
"$end",
"error",
"$unk",
"PROGRAM_BRACKET_LEFT",
"PROGRAM_BRACKET_RIGHT",
"PROGRAM_STRING_START",
"PROGRAM_STRING_END",
"PAREN_LEFT",
"PAREN_RIGHT",
"COMMA",
"SQUARE_BRACKET_LEFT",
"SQUARE_BRACKET_RIGHT",
"ARITH_OP",
"IDENTIFIER",
"INTEGER",
"FLOAT",
"STRING",
}
var parserStatenames = [...]string{}
const parserEofCode = 1
const parserErrCode = 2
const parserInitialStackSize = 16
//line lang.y:196
//line yacctab:1
var parserExca = [...]int{
-1, 1,
1, -1,
-2, 0,
}
const parserNprod = 21
const parserPrivate = 57344
var parserTokenNames []string
var parserStates []string
const parserLast = 37
var parserAct = [...]int{
9, 7, 29, 17, 23, 16, 17, 3, 17, 20,
8, 18, 21, 17, 6, 19, 27, 28, 22, 8,
1, 25, 26, 7, 11, 2, 24, 10, 4, 30,
5, 0, 14, 15, 12, 13, 6,
}
var parserPact = [...]int{
-3, -1000, -3, -1000, -1000, -1000, -1000, 19, -1000, 0,
19, -3, -1000, -1000, 19, 1, -1000, 19, -5, -1000,
19, 19, -1000, -1000, 7, -7, -10, -1000, 19, -1000,
-7,
}
var parserPgo = [...]int{
0, 0, 30, 28, 24, 7, 26, 20,
}
var parserR1 = [...]int{
0, 7, 7, 4, 4, 5, 5, 2, 1, 1,
1, 1, 1, 1, 1, 1, 1, 6, 6, 6,
3,
}
var parserR2 = [...]int{
0, 0, 1, 1, 2, 1, 1, 3, 3, 1,
1, 1, 2, 3, 1, 4, 4, 0, 3, 1,
1,
}
var parserChk = [...]int{
-1000, -7, -4, -5, -3, -2, 17, 4, -5, -1,
8, -4, 15, 16, 13, 14, 5, 13, -1, -1,
8, 11, -1, 9, -6, -1, -1, 9, 10, 12,
-1,
}
var parserDef = [...]int{
1, -2, 2, 3, 5, 6, 20, 0, 4, 0,
0, 9, 10, 11, 0, 14, 7, 0, 0, 12,
17, 0, 13, 8, 0, 19, 0, 15, 0, 16,
18,
}
var parserTok1 = [...]int{
1,
}
var parserTok2 = [...]int{
2, 3, 4, 5, 6, 7, 8, 9, 10, 11,
12, 13, 14, 15, 16, 17,
}
var parserTok3 = [...]int{
0,
}
var parserErrorMessages = [...]struct {
state int
token int
msg string
}{}
//line yaccpar:1
/* parser for yacc output */
var (
parserDebug = 0
parserErrorVerbose = false
)
type parserLexer interface {
Lex(lval *parserSymType) int
Error(s string)
}
type parserParser interface {
Parse(parserLexer) int
Lookahead() int
}
type parserParserImpl struct {
lval parserSymType
stack [parserInitialStackSize]parserSymType
char int
}
func (p *parserParserImpl) Lookahead() int {
return p.char
}
func parserNewParser() parserParser {
return &parserParserImpl{}
}
const parserFlag = -1000
func parserTokname(c int) string {
if c >= 1 && c-1 < len(parserToknames) {
if parserToknames[c-1] != "" {
return parserToknames[c-1]
}
}
return __yyfmt__.Sprintf("tok-%v", c)
}
func parserStatname(s int) string {
if s >= 0 && s < len(parserStatenames) {
if parserStatenames[s] != "" {
return parserStatenames[s]
}
}
return __yyfmt__.Sprintf("state-%v", s)
}
func parserErrorMessage(state, lookAhead int) string {
const TOKSTART = 4
if !parserErrorVerbose {
return "syntax error"
}
for _, e := range parserErrorMessages {
if e.state == state && e.token == lookAhead {
return "syntax error: " + e.msg
}
}
res := "syntax error: unexpected " + parserTokname(lookAhead)
// To match Bison, suggest at most four expected tokens.
expected := make([]int, 0, 4)
// Look for shiftable tokens.
base := parserPact[state]
for tok := TOKSTART; tok-1 < len(parserToknames); tok++ {
if n := base + tok; n >= 0 && n < parserLast && parserChk[parserAct[n]] == tok {
if len(expected) == cap(expected) {
return res
}
expected = append(expected, tok)
}
}
if parserDef[state] == -2 {
i := 0
for parserExca[i] != -1 || parserExca[i+1] != state {
i += 2
}
// Look for tokens that we accept or reduce.
for i += 2; parserExca[i] >= 0; i += 2 {
tok := parserExca[i]
if tok < TOKSTART || parserExca[i+1] == 0 {
continue
}
if len(expected) == cap(expected) {
return res
}
expected = append(expected, tok)
}
// If the default action is to accept or reduce, give up.
if parserExca[i+1] != 0 {
return res
}
}
for i, tok := range expected {
if i == 0 {
res += ", expecting "
} else {
res += " or "
}
res += parserTokname(tok)
}
return res
}
func parserlex1(lex parserLexer, lval *parserSymType) (char, token int) {
token = 0
char = lex.Lex(lval)
if char <= 0 {
token = parserTok1[0]
goto out
}
if char < len(parserTok1) {
token = parserTok1[char]
goto out
}
if char >= parserPrivate {
if char < parserPrivate+len(parserTok2) {
token = parserTok2[char-parserPrivate]
goto out
}
}
for i := 0; i < len(parserTok3); i += 2 {
token = parserTok3[i+0]
if token == char {
token = parserTok3[i+1]
goto out
}
}
out:
if token == 0 {
token = parserTok2[1] /* unknown char */
}
if parserDebug >= 3 {
__yyfmt__.Printf("lex %s(%d)\n", parserTokname(token), uint(char))
}
return char, token
}
func parserParse(parserlex parserLexer) int {
return parserNewParser().Parse(parserlex)
}
func (parserrcvr *parserParserImpl) Parse(parserlex parserLexer) int {
var parsern int
var parserVAL parserSymType
var parserDollar []parserSymType
_ = parserDollar // silence set and not used
parserS := parserrcvr.stack[:]
Nerrs := 0 /* number of errors */
Errflag := 0 /* error recovery flag */
parserstate := 0
parserrcvr.char = -1
parsertoken := -1 // parserrcvr.char translated into internal numbering
defer func() {
// Make sure we report no lookahead when not parsing.
parserstate = -1
parserrcvr.char = -1
parsertoken = -1
}()
parserp := -1
goto parserstack
ret0:
return 0
ret1:
return 1
parserstack:
/* put a state and value onto the stack */
if parserDebug >= 4 {
__yyfmt__.Printf("char %v in %v\n", parserTokname(parsertoken), parserStatname(parserstate))
}
parserp++
if parserp >= len(parserS) {
nyys := make([]parserSymType, len(parserS)*2)
copy(nyys, parserS)
parserS = nyys
}
parserS[parserp] = parserVAL
parserS[parserp].yys = parserstate
parsernewstate:
parsern = parserPact[parserstate]
if parsern <= parserFlag {
goto parserdefault /* simple state */
}
if parserrcvr.char < 0 {
parserrcvr.char, parsertoken = parserlex1(parserlex, &parserrcvr.lval)
}
parsern += parsertoken
if parsern < 0 || parsern >= parserLast {
goto parserdefault
}
parsern = parserAct[parsern]
if parserChk[parsern] == parsertoken { /* valid shift */
parserrcvr.char = -1
parsertoken = -1
parserVAL = parserrcvr.lval
parserstate = parsern
if Errflag > 0 {
Errflag--
}
goto parserstack
}
parserdefault:
/* default state action */
parsern = parserDef[parserstate]
if parsern == -2 {
if parserrcvr.char < 0 {
parserrcvr.char, parsertoken = parserlex1(parserlex, &parserrcvr.lval)
}
/* look through exception table */
xi := 0
for {
if parserExca[xi+0] == -1 && parserExca[xi+1] == parserstate {
break
}
xi += 2
}
for xi += 2; ; xi += 2 {
parsern = parserExca[xi+0]
if parsern < 0 || parsern == parsertoken {
break
}
}
parsern = parserExca[xi+1]
if parsern < 0 {
goto ret0
}
}
if parsern == 0 {
/* error ... attempt to resume parsing */
switch Errflag {
case 0: /* brand new error */
parserlex.Error(parserErrorMessage(parserstate, parsertoken))
Nerrs++
if parserDebug >= 1 {
__yyfmt__.Printf("%s", parserStatname(parserstate))
__yyfmt__.Printf(" saw %s\n", parserTokname(parsertoken))
}
fallthrough
case 1, 2: /* incompletely recovered error ... try again */
Errflag = 3
/* find a state where "error" is a legal shift action */
for parserp >= 0 {
parsern = parserPact[parserS[parserp].yys] + parserErrCode
if parsern >= 0 && parsern < parserLast {
parserstate = parserAct[parsern] /* simulate a shift of "error" */
if parserChk[parserstate] == parserErrCode {
goto parserstack
}
}
/* the current p has no shift on "error", pop stack */
if parserDebug >= 2 {
__yyfmt__.Printf("error recovery pops state %d\n", parserS[parserp].yys)
}
parserp--
}
/* there is no state on the stack with an error shift ... abort */
goto ret1
case 3: /* no shift yet; clobber input char */
if parserDebug >= 2 {
__yyfmt__.Printf("error recovery discards %s\n", parserTokname(parsertoken))
}
if parsertoken == parserEofCode {
goto ret1
}
parserrcvr.char = -1
parsertoken = -1
goto parsernewstate /* try again in the same state */
}
}
/* reduction by production parsern */
if parserDebug >= 2 {
__yyfmt__.Printf("reduce %v in:\n\t%v\n", parsern, parserStatname(parserstate))
}
parsernt := parsern
parserpt := parserp
_ = parserpt // guard against "declared and not used"
parserp -= parserR2[parsern]
// parserp is now the index of $0. Perform the default action. Iff the
// reduced production is ε, $1 is possibly out of range.
if parserp+1 >= len(parserS) {
nyys := make([]parserSymType, len(parserS)*2)
copy(nyys, parserS)
parserS = nyys
}
parserVAL = parserS[parserp+1]
/* consult goto table to find next state */
parsern = parserR1[parsern]
parserg := parserPgo[parsern]
parserj := parserg + parserS[parserp].yys + 1
if parserj >= parserLast {
parserstate = parserAct[parserg]
} else {
parserstate = parserAct[parserj]
if parserChk[parserstate] != -parsern {
parserstate = parserAct[parserg]
}
}
// dummy call; replaced with literal code
switch parsernt {
case 1:
parserDollar = parserS[parserpt-0 : parserpt+1]
//line lang.y:36
{
parserResult = &ast.LiteralNode{
Value: "",
Typex: ast.TypeString,
Posx: ast.Pos{Column: 1, Line: 1},
}
}
case 2:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:44
{
parserResult = parserDollar[1].node
// We want to make sure that the top value is always an Output
// so that the return value is always a string, list of map from an
// interpolation.
//
// The logic for checking for a LiteralNode is a little annoying
// because functionally the AST is the same, but we do that because
// it makes for an easy literal check later (to check if a string
// has any interpolations).
if _, ok := parserDollar[1].node.(*ast.Output); !ok {
if n, ok := parserDollar[1].node.(*ast.LiteralNode); !ok || n.Typex != ast.TypeString {
parserResult = &ast.Output{
Exprs: []ast.Node{parserDollar[1].node},
Posx: parserDollar[1].node.Pos(),
}
}
}
}
case 3:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:67
{
parserVAL.node = parserDollar[1].node
}
case 4:
parserDollar = parserS[parserpt-2 : parserpt+1]
//line lang.y:71
{
var result []ast.Node
if c, ok := parserDollar[1].node.(*ast.Output); ok {
result = append(c.Exprs, parserDollar[2].node)
} else {
result = []ast.Node{parserDollar[1].node, parserDollar[2].node}
}
parserVAL.node = &ast.Output{
Exprs: result,
Posx: result[0].Pos(),
}
}
case 5:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:87
{
parserVAL.node = parserDollar[1].node
}
case 6:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:91
{
parserVAL.node = parserDollar[1].node
}
case 7:
parserDollar = parserS[parserpt-3 : parserpt+1]
//line lang.y:97
{
parserVAL.node = parserDollar[2].node
}
case 8:
parserDollar = parserS[parserpt-3 : parserpt+1]
//line lang.y:103
{
parserVAL.node = parserDollar[2].node
}
case 9:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:107
{
parserVAL.node = parserDollar[1].node
}
case 10:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:111
{
parserVAL.node = &ast.LiteralNode{
Value: parserDollar[1].token.Value.(int),
Typex: ast.TypeInt,
Posx: parserDollar[1].token.Pos,
}
}
case 11:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:119
{
parserVAL.node = &ast.LiteralNode{
Value: parserDollar[1].token.Value.(float64),
Typex: ast.TypeFloat,
Posx: parserDollar[1].token.Pos,
}
}
case 12:
parserDollar = parserS[parserpt-2 : parserpt+1]
//line lang.y:127
{
// This is REALLY jank. We assume that a singular ARITH_OP
// means 0 ARITH_OP expr, which... is weird. We don't want to
// support *, /, etc., only -. We should fix this later with a pure
// Go scanner/parser.
if parserDollar[1].token.Value.(ast.ArithmeticOp) != ast.ArithmeticOpSub {
panic("Unary - is only allowed")
}
parserVAL.node = &ast.Arithmetic{
Op: parserDollar[1].token.Value.(ast.ArithmeticOp),
Exprs: []ast.Node{
&ast.LiteralNode{Value: 0, Typex: ast.TypeInt},
parserDollar[2].node,
},
Posx: parserDollar[2].node.Pos(),
}
}
case 13:
parserDollar = parserS[parserpt-3 : parserpt+1]
//line lang.y:146
{
parserVAL.node = &ast.Arithmetic{
Op: parserDollar[2].token.Value.(ast.ArithmeticOp),
Exprs: []ast.Node{parserDollar[1].node, parserDollar[3].node},
Posx: parserDollar[1].node.Pos(),
}
}
case 14:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:154
{
parserVAL.node = &ast.VariableAccess{Name: parserDollar[1].token.Value.(string), Posx: parserDollar[1].token.Pos}
}
case 15:
parserDollar = parserS[parserpt-4 : parserpt+1]
//line lang.y:158
{
parserVAL.node = &ast.Call{Func: parserDollar[1].token.Value.(string), Args: parserDollar[3].nodeList, Posx: parserDollar[1].token.Pos}
}
case 16:
parserDollar = parserS[parserpt-4 : parserpt+1]
//line lang.y:162
{
parserVAL.node = &ast.Index{
Target: &ast.VariableAccess{
Name: parserDollar[1].token.Value.(string),
Posx: parserDollar[1].token.Pos,
},
Key: parserDollar[3].node,
Posx: parserDollar[1].token.Pos,
}
}
case 17:
parserDollar = parserS[parserpt-0 : parserpt+1]
//line lang.y:174
{
parserVAL.nodeList = nil
}
case 18:
parserDollar = parserS[parserpt-3 : parserpt+1]
//line lang.y:178
{
parserVAL.nodeList = append(parserDollar[1].nodeList, parserDollar[3].node)
}
case 19:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:182
{
parserVAL.nodeList = append(parserVAL.nodeList, parserDollar[1].node)
}
case 20:
parserDollar = parserS[parserpt-1 : parserpt+1]
//line lang.y:188
{
parserVAL.node = &ast.LiteralNode{
Value: parserDollar[1].token.Value.(string),
Typex: ast.TypeString,
Posx: parserDollar[1].token.Pos,
}
}
}
goto parserstack /* stack new state and value */
}