mirror of
https://github.com/ceph/ceph-csi.git
synced 2024-11-23 23:00:19 +00:00
5a66991bb3
updating the kubernetes release to the latest in main go.mod Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
685 lines
16 KiB
Go
685 lines
16 KiB
Go
// Copyright (c) 2012-2022 The ANTLR Project. All rights reserved.
|
|
// Use of this file is governed by the BSD 3-clause license that
|
|
// can be found in the LICENSE.txt file in the project root.
|
|
|
|
package antlr
|
|
|
|
import (
|
|
"fmt"
|
|
"strconv"
|
|
)
|
|
|
|
const serializedVersion = 4
|
|
|
|
type loopEndStateIntPair struct {
|
|
item0 *LoopEndState
|
|
item1 int
|
|
}
|
|
|
|
type blockStartStateIntPair struct {
|
|
item0 BlockStartState
|
|
item1 int
|
|
}
|
|
|
|
type ATNDeserializer struct {
|
|
options *ATNDeserializationOptions
|
|
data []int32
|
|
pos int
|
|
}
|
|
|
|
func NewATNDeserializer(options *ATNDeserializationOptions) *ATNDeserializer {
|
|
if options == nil {
|
|
options = &defaultATNDeserializationOptions
|
|
}
|
|
|
|
return &ATNDeserializer{options: options}
|
|
}
|
|
|
|
//goland:noinspection GoUnusedFunction
|
|
func stringInSlice(a string, list []string) int {
|
|
for i, b := range list {
|
|
if b == a {
|
|
return i
|
|
}
|
|
}
|
|
|
|
return -1
|
|
}
|
|
|
|
func (a *ATNDeserializer) Deserialize(data []int32) *ATN {
|
|
a.data = data
|
|
a.pos = 0
|
|
a.checkVersion()
|
|
|
|
atn := a.readATN()
|
|
|
|
a.readStates(atn)
|
|
a.readRules(atn)
|
|
a.readModes(atn)
|
|
|
|
sets := a.readSets(atn, nil)
|
|
|
|
a.readEdges(atn, sets)
|
|
a.readDecisions(atn)
|
|
a.readLexerActions(atn)
|
|
a.markPrecedenceDecisions(atn)
|
|
a.verifyATN(atn)
|
|
|
|
if a.options.GenerateRuleBypassTransitions() && atn.grammarType == ATNTypeParser {
|
|
a.generateRuleBypassTransitions(atn)
|
|
// Re-verify after modification
|
|
a.verifyATN(atn)
|
|
}
|
|
|
|
return atn
|
|
|
|
}
|
|
|
|
func (a *ATNDeserializer) checkVersion() {
|
|
version := a.readInt()
|
|
|
|
if version != serializedVersion {
|
|
panic("Could not deserialize ATN with version " + strconv.Itoa(version) + " (expected " + strconv.Itoa(serializedVersion) + ").")
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) readATN() *ATN {
|
|
grammarType := a.readInt()
|
|
maxTokenType := a.readInt()
|
|
|
|
return NewATN(grammarType, maxTokenType)
|
|
}
|
|
|
|
func (a *ATNDeserializer) readStates(atn *ATN) {
|
|
nstates := a.readInt()
|
|
|
|
// Allocate worst case size.
|
|
loopBackStateNumbers := make([]loopEndStateIntPair, 0, nstates)
|
|
endStateNumbers := make([]blockStartStateIntPair, 0, nstates)
|
|
|
|
// Preallocate states slice.
|
|
atn.states = make([]ATNState, 0, nstates)
|
|
|
|
for i := 0; i < nstates; i++ {
|
|
stype := a.readInt()
|
|
|
|
// Ignore bad types of states
|
|
if stype == ATNStateInvalidType {
|
|
atn.addState(nil)
|
|
continue
|
|
}
|
|
|
|
ruleIndex := a.readInt()
|
|
|
|
s := a.stateFactory(stype, ruleIndex)
|
|
|
|
if stype == ATNStateLoopEnd {
|
|
loopBackStateNumber := a.readInt()
|
|
|
|
loopBackStateNumbers = append(loopBackStateNumbers, loopEndStateIntPair{s.(*LoopEndState), loopBackStateNumber})
|
|
} else if s2, ok := s.(BlockStartState); ok {
|
|
endStateNumber := a.readInt()
|
|
|
|
endStateNumbers = append(endStateNumbers, blockStartStateIntPair{s2, endStateNumber})
|
|
}
|
|
|
|
atn.addState(s)
|
|
}
|
|
|
|
// Delay the assignment of loop back and end states until we know all the state
|
|
// instances have been initialized
|
|
for _, pair := range loopBackStateNumbers {
|
|
pair.item0.loopBackState = atn.states[pair.item1]
|
|
}
|
|
|
|
for _, pair := range endStateNumbers {
|
|
pair.item0.setEndState(atn.states[pair.item1].(*BlockEndState))
|
|
}
|
|
|
|
numNonGreedyStates := a.readInt()
|
|
for j := 0; j < numNonGreedyStates; j++ {
|
|
stateNumber := a.readInt()
|
|
|
|
atn.states[stateNumber].(DecisionState).setNonGreedy(true)
|
|
}
|
|
|
|
numPrecedenceStates := a.readInt()
|
|
for j := 0; j < numPrecedenceStates; j++ {
|
|
stateNumber := a.readInt()
|
|
|
|
atn.states[stateNumber].(*RuleStartState).isPrecedenceRule = true
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) readRules(atn *ATN) {
|
|
nrules := a.readInt()
|
|
|
|
if atn.grammarType == ATNTypeLexer {
|
|
atn.ruleToTokenType = make([]int, nrules)
|
|
}
|
|
|
|
atn.ruleToStartState = make([]*RuleStartState, nrules)
|
|
|
|
for i := range atn.ruleToStartState {
|
|
s := a.readInt()
|
|
startState := atn.states[s].(*RuleStartState)
|
|
|
|
atn.ruleToStartState[i] = startState
|
|
|
|
if atn.grammarType == ATNTypeLexer {
|
|
tokenType := a.readInt()
|
|
|
|
atn.ruleToTokenType[i] = tokenType
|
|
}
|
|
}
|
|
|
|
atn.ruleToStopState = make([]*RuleStopState, nrules)
|
|
|
|
for _, state := range atn.states {
|
|
if s2, ok := state.(*RuleStopState); ok {
|
|
atn.ruleToStopState[s2.ruleIndex] = s2
|
|
atn.ruleToStartState[s2.ruleIndex].stopState = s2
|
|
}
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) readModes(atn *ATN) {
|
|
nmodes := a.readInt()
|
|
atn.modeToStartState = make([]*TokensStartState, nmodes)
|
|
|
|
for i := range atn.modeToStartState {
|
|
s := a.readInt()
|
|
|
|
atn.modeToStartState[i] = atn.states[s].(*TokensStartState)
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) readSets(_ *ATN, sets []*IntervalSet) []*IntervalSet {
|
|
m := a.readInt()
|
|
|
|
// Preallocate the needed capacity.
|
|
if cap(sets)-len(sets) < m {
|
|
isets := make([]*IntervalSet, len(sets), len(sets)+m)
|
|
copy(isets, sets)
|
|
sets = isets
|
|
}
|
|
|
|
for i := 0; i < m; i++ {
|
|
iset := NewIntervalSet()
|
|
|
|
sets = append(sets, iset)
|
|
|
|
n := a.readInt()
|
|
containsEOF := a.readInt()
|
|
|
|
if containsEOF != 0 {
|
|
iset.addOne(-1)
|
|
}
|
|
|
|
for j := 0; j < n; j++ {
|
|
i1 := a.readInt()
|
|
i2 := a.readInt()
|
|
|
|
iset.addRange(i1, i2)
|
|
}
|
|
}
|
|
|
|
return sets
|
|
}
|
|
|
|
func (a *ATNDeserializer) readEdges(atn *ATN, sets []*IntervalSet) {
|
|
nedges := a.readInt()
|
|
|
|
for i := 0; i < nedges; i++ {
|
|
var (
|
|
src = a.readInt()
|
|
trg = a.readInt()
|
|
ttype = a.readInt()
|
|
arg1 = a.readInt()
|
|
arg2 = a.readInt()
|
|
arg3 = a.readInt()
|
|
trans = a.edgeFactory(atn, ttype, src, trg, arg1, arg2, arg3, sets)
|
|
srcState = atn.states[src]
|
|
)
|
|
|
|
srcState.AddTransition(trans, -1)
|
|
}
|
|
|
|
// Edges for rule stop states can be derived, so they are not serialized
|
|
for _, state := range atn.states {
|
|
for _, t := range state.GetTransitions() {
|
|
var rt, ok = t.(*RuleTransition)
|
|
|
|
if !ok {
|
|
continue
|
|
}
|
|
|
|
outermostPrecedenceReturn := -1
|
|
|
|
if atn.ruleToStartState[rt.getTarget().GetRuleIndex()].isPrecedenceRule {
|
|
if rt.precedence == 0 {
|
|
outermostPrecedenceReturn = rt.getTarget().GetRuleIndex()
|
|
}
|
|
}
|
|
|
|
trans := NewEpsilonTransition(rt.followState, outermostPrecedenceReturn)
|
|
|
|
atn.ruleToStopState[rt.getTarget().GetRuleIndex()].AddTransition(trans, -1)
|
|
}
|
|
}
|
|
|
|
for _, state := range atn.states {
|
|
if s2, ok := state.(BlockStartState); ok {
|
|
// We need to know the end state to set its start state
|
|
if s2.getEndState() == nil {
|
|
panic("IllegalState")
|
|
}
|
|
|
|
// Block end states can only be associated to a single block start state
|
|
if s2.getEndState().startState != nil {
|
|
panic("IllegalState")
|
|
}
|
|
|
|
s2.getEndState().startState = state
|
|
}
|
|
|
|
if s2, ok := state.(*PlusLoopbackState); ok {
|
|
for _, t := range s2.GetTransitions() {
|
|
if t2, ok := t.getTarget().(*PlusBlockStartState); ok {
|
|
t2.loopBackState = state
|
|
}
|
|
}
|
|
} else if s2, ok := state.(*StarLoopbackState); ok {
|
|
for _, t := range s2.GetTransitions() {
|
|
if t2, ok := t.getTarget().(*StarLoopEntryState); ok {
|
|
t2.loopBackState = state
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) readDecisions(atn *ATN) {
|
|
ndecisions := a.readInt()
|
|
|
|
for i := 0; i < ndecisions; i++ {
|
|
s := a.readInt()
|
|
decState := atn.states[s].(DecisionState)
|
|
|
|
atn.DecisionToState = append(atn.DecisionToState, decState)
|
|
decState.setDecision(i)
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) readLexerActions(atn *ATN) {
|
|
if atn.grammarType == ATNTypeLexer {
|
|
count := a.readInt()
|
|
|
|
atn.lexerActions = make([]LexerAction, count)
|
|
|
|
for i := range atn.lexerActions {
|
|
actionType := a.readInt()
|
|
data1 := a.readInt()
|
|
data2 := a.readInt()
|
|
atn.lexerActions[i] = a.lexerActionFactory(actionType, data1, data2)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) generateRuleBypassTransitions(atn *ATN) {
|
|
count := len(atn.ruleToStartState)
|
|
|
|
for i := 0; i < count; i++ {
|
|
atn.ruleToTokenType[i] = atn.maxTokenType + i + 1
|
|
}
|
|
|
|
for i := 0; i < count; i++ {
|
|
a.generateRuleBypassTransition(atn, i)
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) generateRuleBypassTransition(atn *ATN, idx int) {
|
|
bypassStart := NewBasicBlockStartState()
|
|
|
|
bypassStart.ruleIndex = idx
|
|
atn.addState(bypassStart)
|
|
|
|
bypassStop := NewBlockEndState()
|
|
|
|
bypassStop.ruleIndex = idx
|
|
atn.addState(bypassStop)
|
|
|
|
bypassStart.endState = bypassStop
|
|
|
|
atn.defineDecisionState(&bypassStart.BaseDecisionState)
|
|
|
|
bypassStop.startState = bypassStart
|
|
|
|
var excludeTransition Transition
|
|
var endState ATNState
|
|
|
|
if atn.ruleToStartState[idx].isPrecedenceRule {
|
|
// Wrap from the beginning of the rule to the StarLoopEntryState
|
|
endState = nil
|
|
|
|
for i := 0; i < len(atn.states); i++ {
|
|
state := atn.states[i]
|
|
|
|
if a.stateIsEndStateFor(state, idx) != nil {
|
|
endState = state
|
|
excludeTransition = state.(*StarLoopEntryState).loopBackState.GetTransitions()[0]
|
|
|
|
break
|
|
}
|
|
}
|
|
|
|
if excludeTransition == nil {
|
|
panic("Couldn't identify final state of the precedence rule prefix section.")
|
|
}
|
|
} else {
|
|
endState = atn.ruleToStopState[idx]
|
|
}
|
|
|
|
// All non-excluded transitions that currently target end state need to target
|
|
// blockEnd instead
|
|
for i := 0; i < len(atn.states); i++ {
|
|
state := atn.states[i]
|
|
|
|
for j := 0; j < len(state.GetTransitions()); j++ {
|
|
transition := state.GetTransitions()[j]
|
|
|
|
if transition == excludeTransition {
|
|
continue
|
|
}
|
|
|
|
if transition.getTarget() == endState {
|
|
transition.setTarget(bypassStop)
|
|
}
|
|
}
|
|
}
|
|
|
|
// All transitions leaving the rule start state need to leave blockStart instead
|
|
ruleToStartState := atn.ruleToStartState[idx]
|
|
count := len(ruleToStartState.GetTransitions())
|
|
|
|
for count > 0 {
|
|
bypassStart.AddTransition(ruleToStartState.GetTransitions()[count-1], -1)
|
|
ruleToStartState.SetTransitions([]Transition{ruleToStartState.GetTransitions()[len(ruleToStartState.GetTransitions())-1]})
|
|
}
|
|
|
|
// Link the new states
|
|
atn.ruleToStartState[idx].AddTransition(NewEpsilonTransition(bypassStart, -1), -1)
|
|
bypassStop.AddTransition(NewEpsilonTransition(endState, -1), -1)
|
|
|
|
MatchState := NewBasicState()
|
|
|
|
atn.addState(MatchState)
|
|
MatchState.AddTransition(NewAtomTransition(bypassStop, atn.ruleToTokenType[idx]), -1)
|
|
bypassStart.AddTransition(NewEpsilonTransition(MatchState, -1), -1)
|
|
}
|
|
|
|
func (a *ATNDeserializer) stateIsEndStateFor(state ATNState, idx int) ATNState {
|
|
if state.GetRuleIndex() != idx {
|
|
return nil
|
|
}
|
|
|
|
if _, ok := state.(*StarLoopEntryState); !ok {
|
|
return nil
|
|
}
|
|
|
|
maybeLoopEndState := state.GetTransitions()[len(state.GetTransitions())-1].getTarget()
|
|
|
|
if _, ok := maybeLoopEndState.(*LoopEndState); !ok {
|
|
return nil
|
|
}
|
|
|
|
var _, ok = maybeLoopEndState.GetTransitions()[0].getTarget().(*RuleStopState)
|
|
|
|
if maybeLoopEndState.(*LoopEndState).epsilonOnlyTransitions && ok {
|
|
return state
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// markPrecedenceDecisions analyzes the StarLoopEntryState states in the
|
|
// specified ATN to set the StarLoopEntryState.precedenceRuleDecision field to
|
|
// the correct value.
|
|
func (a *ATNDeserializer) markPrecedenceDecisions(atn *ATN) {
|
|
for _, state := range atn.states {
|
|
if _, ok := state.(*StarLoopEntryState); !ok {
|
|
continue
|
|
}
|
|
|
|
// We analyze the [ATN] to determine if an ATN decision state is the
|
|
// decision for the closure block that determines whether a
|
|
// precedence rule should continue or complete.
|
|
if atn.ruleToStartState[state.GetRuleIndex()].isPrecedenceRule {
|
|
maybeLoopEndState := state.GetTransitions()[len(state.GetTransitions())-1].getTarget()
|
|
|
|
if s3, ok := maybeLoopEndState.(*LoopEndState); ok {
|
|
var _, ok2 = maybeLoopEndState.GetTransitions()[0].getTarget().(*RuleStopState)
|
|
|
|
if s3.epsilonOnlyTransitions && ok2 {
|
|
state.(*StarLoopEntryState).precedenceRuleDecision = true
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) verifyATN(atn *ATN) {
|
|
if !a.options.VerifyATN() {
|
|
return
|
|
}
|
|
|
|
// Verify assumptions
|
|
for _, state := range atn.states {
|
|
if state == nil {
|
|
continue
|
|
}
|
|
|
|
a.checkCondition(state.GetEpsilonOnlyTransitions() || len(state.GetTransitions()) <= 1, "")
|
|
|
|
switch s2 := state.(type) {
|
|
case *PlusBlockStartState:
|
|
a.checkCondition(s2.loopBackState != nil, "")
|
|
|
|
case *StarLoopEntryState:
|
|
a.checkCondition(s2.loopBackState != nil, "")
|
|
a.checkCondition(len(s2.GetTransitions()) == 2, "")
|
|
|
|
switch s2.transitions[0].getTarget().(type) {
|
|
case *StarBlockStartState:
|
|
_, ok := s2.transitions[1].getTarget().(*LoopEndState)
|
|
|
|
a.checkCondition(ok, "")
|
|
a.checkCondition(!s2.nonGreedy, "")
|
|
|
|
case *LoopEndState:
|
|
var _, ok = s2.transitions[1].getTarget().(*StarBlockStartState)
|
|
|
|
a.checkCondition(ok, "")
|
|
a.checkCondition(s2.nonGreedy, "")
|
|
|
|
default:
|
|
panic("IllegalState")
|
|
}
|
|
|
|
case *StarLoopbackState:
|
|
a.checkCondition(len(state.GetTransitions()) == 1, "")
|
|
|
|
var _, ok = state.GetTransitions()[0].getTarget().(*StarLoopEntryState)
|
|
|
|
a.checkCondition(ok, "")
|
|
|
|
case *LoopEndState:
|
|
a.checkCondition(s2.loopBackState != nil, "")
|
|
|
|
case *RuleStartState:
|
|
a.checkCondition(s2.stopState != nil, "")
|
|
|
|
case BlockStartState:
|
|
a.checkCondition(s2.getEndState() != nil, "")
|
|
|
|
case *BlockEndState:
|
|
a.checkCondition(s2.startState != nil, "")
|
|
|
|
case DecisionState:
|
|
a.checkCondition(len(s2.GetTransitions()) <= 1 || s2.getDecision() >= 0, "")
|
|
|
|
default:
|
|
var _, ok = s2.(*RuleStopState)
|
|
|
|
a.checkCondition(len(s2.GetTransitions()) <= 1 || ok, "")
|
|
}
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) checkCondition(condition bool, message string) {
|
|
if !condition {
|
|
if message == "" {
|
|
message = "IllegalState"
|
|
}
|
|
|
|
panic(message)
|
|
}
|
|
}
|
|
|
|
func (a *ATNDeserializer) readInt() int {
|
|
v := a.data[a.pos]
|
|
|
|
a.pos++
|
|
|
|
return int(v) // data is 32 bits but int is at least that big
|
|
}
|
|
|
|
func (a *ATNDeserializer) edgeFactory(atn *ATN, typeIndex, _, trg, arg1, arg2, arg3 int, sets []*IntervalSet) Transition {
|
|
target := atn.states[trg]
|
|
|
|
switch typeIndex {
|
|
case TransitionEPSILON:
|
|
return NewEpsilonTransition(target, -1)
|
|
|
|
case TransitionRANGE:
|
|
if arg3 != 0 {
|
|
return NewRangeTransition(target, TokenEOF, arg2)
|
|
}
|
|
|
|
return NewRangeTransition(target, arg1, arg2)
|
|
|
|
case TransitionRULE:
|
|
return NewRuleTransition(atn.states[arg1], arg2, arg3, target)
|
|
|
|
case TransitionPREDICATE:
|
|
return NewPredicateTransition(target, arg1, arg2, arg3 != 0)
|
|
|
|
case TransitionPRECEDENCE:
|
|
return NewPrecedencePredicateTransition(target, arg1)
|
|
|
|
case TransitionATOM:
|
|
if arg3 != 0 {
|
|
return NewAtomTransition(target, TokenEOF)
|
|
}
|
|
|
|
return NewAtomTransition(target, arg1)
|
|
|
|
case TransitionACTION:
|
|
return NewActionTransition(target, arg1, arg2, arg3 != 0)
|
|
|
|
case TransitionSET:
|
|
return NewSetTransition(target, sets[arg1])
|
|
|
|
case TransitionNOTSET:
|
|
return NewNotSetTransition(target, sets[arg1])
|
|
|
|
case TransitionWILDCARD:
|
|
return NewWildcardTransition(target)
|
|
}
|
|
|
|
panic("The specified transition type is not valid.")
|
|
}
|
|
|
|
func (a *ATNDeserializer) stateFactory(typeIndex, ruleIndex int) ATNState {
|
|
var s ATNState
|
|
|
|
switch typeIndex {
|
|
case ATNStateInvalidType:
|
|
return nil
|
|
|
|
case ATNStateBasic:
|
|
s = NewBasicState()
|
|
|
|
case ATNStateRuleStart:
|
|
s = NewRuleStartState()
|
|
|
|
case ATNStateBlockStart:
|
|
s = NewBasicBlockStartState()
|
|
|
|
case ATNStatePlusBlockStart:
|
|
s = NewPlusBlockStartState()
|
|
|
|
case ATNStateStarBlockStart:
|
|
s = NewStarBlockStartState()
|
|
|
|
case ATNStateTokenStart:
|
|
s = NewTokensStartState()
|
|
|
|
case ATNStateRuleStop:
|
|
s = NewRuleStopState()
|
|
|
|
case ATNStateBlockEnd:
|
|
s = NewBlockEndState()
|
|
|
|
case ATNStateStarLoopBack:
|
|
s = NewStarLoopbackState()
|
|
|
|
case ATNStateStarLoopEntry:
|
|
s = NewStarLoopEntryState()
|
|
|
|
case ATNStatePlusLoopBack:
|
|
s = NewPlusLoopbackState()
|
|
|
|
case ATNStateLoopEnd:
|
|
s = NewLoopEndState()
|
|
|
|
default:
|
|
panic(fmt.Sprintf("state type %d is invalid", typeIndex))
|
|
}
|
|
|
|
s.SetRuleIndex(ruleIndex)
|
|
|
|
return s
|
|
}
|
|
|
|
func (a *ATNDeserializer) lexerActionFactory(typeIndex, data1, data2 int) LexerAction {
|
|
switch typeIndex {
|
|
case LexerActionTypeChannel:
|
|
return NewLexerChannelAction(data1)
|
|
|
|
case LexerActionTypeCustom:
|
|
return NewLexerCustomAction(data1, data2)
|
|
|
|
case LexerActionTypeMode:
|
|
return NewLexerModeAction(data1)
|
|
|
|
case LexerActionTypeMore:
|
|
return LexerMoreActionINSTANCE
|
|
|
|
case LexerActionTypePopMode:
|
|
return LexerPopModeActionINSTANCE
|
|
|
|
case LexerActionTypePushMode:
|
|
return NewLexerPushModeAction(data1)
|
|
|
|
case LexerActionTypeSkip:
|
|
return LexerSkipActionINSTANCE
|
|
|
|
case LexerActionTypeType:
|
|
return NewLexerTypeAction(data1)
|
|
|
|
default:
|
|
panic(fmt.Sprintf("lexer action %d is invalid", typeIndex))
|
|
}
|
|
}
|