You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
397 lines
14 KiB
Go
397 lines
14 KiB
Go
package main
|
|
|
|
import (
|
|
"fmt"
|
|
"sort"
|
|
)
|
|
|
|
// a Match stores a slice of all the capturing groups in a match.
|
|
type Match []Group
|
|
|
|
// a Group represents a group. It contains the start index and end index of the match
|
|
type Group struct {
|
|
startIdx int
|
|
endIdx int
|
|
}
|
|
|
|
func newMatch(size int) Match {
|
|
toRet := make([]Group, size)
|
|
for i := range toRet {
|
|
toRet[i].startIdx = -1
|
|
toRet[i].endIdx = -1
|
|
}
|
|
return toRet
|
|
}
|
|
|
|
// Returns the number of valid groups in the match
|
|
func (m Match) numValidGroups() int {
|
|
numValid := 0
|
|
for _, g := range m {
|
|
if g.startIdx >= 0 && g.endIdx >= 0 {
|
|
numValid++
|
|
}
|
|
}
|
|
return numValid
|
|
}
|
|
|
|
// Returns a string containing the indices of all (valid) groups in the match
|
|
func (m Match) toString() string {
|
|
var toRet string
|
|
for i, g := range m {
|
|
if g.isValid() {
|
|
toRet += fmt.Sprintf("Group %d\n", i)
|
|
toRet += g.toString()
|
|
toRet += "\n"
|
|
}
|
|
}
|
|
return toRet
|
|
}
|
|
|
|
// Converts the Group into a string representation:
|
|
func (idx Group) toString() string {
|
|
return fmt.Sprintf("%d\t%d", idx.startIdx, idx.endIdx)
|
|
}
|
|
|
|
// Returns whether a group contains valid indices
|
|
func (g Group) isValid() bool {
|
|
return g.startIdx >= 0 && g.endIdx >= 0
|
|
}
|
|
|
|
// takeZeroState takes the 0-state (if such a transition exists) for all states in the
|
|
// given slice. It returns the resulting states. If any of the resulting states is a 0-state,
|
|
// the second ret val is true.
|
|
// The third ret val is a list of all the group numbers of all the opening parentheses we crossed,
|
|
// and the fourth is a list of all the closing parentheses we passed
|
|
func takeZeroState(states []*State) (rtv []*State, isZero bool, openParenGroups []int, closeParenGroups []int) {
|
|
for _, state := range states {
|
|
if len(state.transitions[EPSILON]) > 0 {
|
|
for _, s := range state.transitions[EPSILON] {
|
|
if s.groupBegin {
|
|
openParenGroups = append(openParenGroups, s.groupNum)
|
|
}
|
|
if s.groupEnd {
|
|
closeParenGroups = append(closeParenGroups, s.groupNum)
|
|
}
|
|
}
|
|
rtv = append(rtv, state.transitions[EPSILON]...)
|
|
}
|
|
}
|
|
for _, state := range rtv {
|
|
if len(state.transitions[EPSILON]) > 0 {
|
|
return rtv, true, openParenGroups, closeParenGroups
|
|
}
|
|
}
|
|
return rtv, false, openParenGroups, closeParenGroups
|
|
}
|
|
|
|
// zeroMatchPossible returns true if a zero-length match is possible
|
|
// from any of the given states, given the string and our position in it.
|
|
// It uses the same algorithm to find zero-states as the one inside the loop,
|
|
// so I should probably put it in a function.
|
|
// It also returns all the capturing groups that both begin and end at the current index.
|
|
// This is because, by definition, zero-states don't move forward in the string.
|
|
func zeroMatchPossible(str []rune, idx int, states ...*State) (bool, []int, []int) {
|
|
allOpenParenGroups := make([]int, 0)
|
|
allCloseParenGroups := make([]int, 0)
|
|
zeroStates, isZero, openParenGroups, closeParenGroups := takeZeroState(states)
|
|
allOpenParenGroups = append(allOpenParenGroups, openParenGroups...)
|
|
allCloseParenGroups = append(allCloseParenGroups, closeParenGroups...)
|
|
tempstates := make([]*State, 0, len(zeroStates)+len(states))
|
|
tempstates = append(tempstates, states...)
|
|
tempstates = append(tempstates, zeroStates...)
|
|
num_appended := 0 // number of unique states addded to tempstates
|
|
for isZero == true {
|
|
zeroStates, isZero, openParenGroups, closeParenGroups = takeZeroState(tempstates)
|
|
allOpenParenGroups = append(allOpenParenGroups, openParenGroups...)
|
|
allCloseParenGroups = append(allCloseParenGroups, closeParenGroups...)
|
|
tempstates, num_appended = unique_append(tempstates, zeroStates...)
|
|
if num_appended == 0 { // break if we haven't appended any more unique values
|
|
break
|
|
}
|
|
}
|
|
for _, state := range tempstates {
|
|
if state.isEmpty && (state.assert == NONE || state.checkAssertion(str, idx)) && state.isLast {
|
|
return true, allOpenParenGroups, allCloseParenGroups
|
|
}
|
|
}
|
|
return false, allOpenParenGroups, allCloseParenGroups
|
|
}
|
|
|
|
// Prunes the slice by removing overlapping indices.
|
|
func pruneIndices(indices []Match) []Match {
|
|
// First, sort the slice by the start indices
|
|
sort.Slice(indices, func(i, j int) bool {
|
|
return indices[i][0].startIdx < indices[j][0].startIdx
|
|
})
|
|
toRet := make([]Match, 0, len(indices))
|
|
current := indices[0]
|
|
for _, idx := range indices[1:] {
|
|
// idx doesn't overlap with current (starts after current ends), so add current to result
|
|
// and update the current.
|
|
if idx[0].startIdx >= current[0].endIdx {
|
|
toRet = append(toRet, current)
|
|
current = idx
|
|
} else if idx[0].endIdx > current[0].endIdx {
|
|
// idx overlaps, but it is longer, so update current
|
|
current = idx
|
|
}
|
|
}
|
|
// Add last state
|
|
toRet = append(toRet, current)
|
|
return toRet
|
|
}
|
|
|
|
// findAllMatches tries to find all matches of the regex represented by given start-state, with
|
|
// the given string
|
|
func findAllMatches(start *State, str []rune, numGroups int) []Match {
|
|
idx := 0
|
|
var matchFound bool
|
|
var matchIdx Match
|
|
indices := make([]Match, 0)
|
|
for idx <= len(str) {
|
|
matchFound, matchIdx, idx = findAllMatchesHelper(start, str, idx, numGroups)
|
|
if matchFound {
|
|
indices = append(indices, matchIdx)
|
|
}
|
|
}
|
|
if len(indices) > 0 {
|
|
return pruneIndices(indices)
|
|
}
|
|
return indices
|
|
}
|
|
|
|
// Helper for findAllMatches. Returns whether it found a match, the
|
|
// first Match it finds, and how far it got into the string ie. where
|
|
// the next search should start from.
|
|
//
|
|
// Might return duplicates or overlapping indices, so care must be taken to prune the resulting array.
|
|
func findAllMatchesHelper(start *State, str []rune, offset int, numGroups int) (bool, Match, int) {
|
|
// Base case - exit if offset exceeds string's length
|
|
if offset > len(str) {
|
|
// The second value here shouldn't be used, because we should exit when the third return value is > than len(str)
|
|
return false, []Group{}, offset
|
|
}
|
|
// 'Base case' - if we are at the end of the string, check if we can add a zero-length match
|
|
if offset == len(str) {
|
|
// Get all zero-state matches. If we can get to a zero-state without matching anything, we
|
|
// can add a zero-length match. This is all true only if the start state itself matches nothing.
|
|
if start.isEmpty {
|
|
to_return := newMatch(numGroups + 1)
|
|
if start.groupBegin {
|
|
to_return[start.groupNum].startIdx = offset
|
|
}
|
|
if ok, openGrps, closeGrps := zeroMatchPossible(str, offset, start); ok {
|
|
for _, gIdx := range openGrps {
|
|
to_return[gIdx].startIdx = offset
|
|
}
|
|
for _, gIdx := range closeGrps {
|
|
to_return[gIdx].endIdx = offset
|
|
}
|
|
to_return[0] = Group{offset, offset}
|
|
return true, to_return, offset + 1
|
|
}
|
|
}
|
|
return false, []Group{}, offset + 1
|
|
}
|
|
|
|
// Hold a list of match indices for the current run. When we
|
|
// can no longer find a match, the match with the largest range is
|
|
// chosen as the match for the entire string.
|
|
// This allows us to pick the longest possible match (which is how greedy matching works).
|
|
// COMMENT ABOVE IS CURRENTLY NOT UP-TO-DATE
|
|
tempIndices := newMatch(numGroups + 1)
|
|
|
|
foundPath := false
|
|
startIdx := offset
|
|
endIdx := offset
|
|
currentStates := make([]*State, 0)
|
|
tempStates := make([]*State, 0) // Used to store states that should be used in next loop iteration
|
|
i := offset // Index in string
|
|
startingFrom := i // Store starting index
|
|
|
|
// If the first state is an assertion, makes sure the assertion
|
|
// is true before we do _anything_ else.
|
|
if start.assert != NONE {
|
|
if start.checkAssertion(str, offset) == false {
|
|
i++
|
|
return false, []Group{}, i
|
|
}
|
|
}
|
|
// Increment until we hit a character matching the start state (assuming not 0-state)
|
|
if start.isEmpty == false {
|
|
for i < len(str) && !start.contentContains(str, i) {
|
|
i++
|
|
}
|
|
startIdx = i
|
|
startingFrom = i
|
|
i++ // Advance to next character (if we aren't at a 0-state, which doesn't match anything), so that we can check for transitions. If we advance at a 0-state, we will never get a chance to match the first character
|
|
}
|
|
|
|
// Check if the start state begins a group - if so, add the start index to our list
|
|
if start.groupBegin {
|
|
tempIndices[start.groupNum].startIdx = i
|
|
}
|
|
|
|
currentStates = append(currentStates, start)
|
|
|
|
// Main loop
|
|
for i < len(str) {
|
|
foundPath = false
|
|
|
|
zeroStates := make([]*State, 0)
|
|
// Keep taking zero-states, until there are no more left to take
|
|
// Objective: If any of our current states have transitions to 0-states, replace them with the 0-state. Do this until there are no more transitions to 0-states, or there are no more unique 0-states to take.
|
|
zeroStates, isZero, openParenGroups, closeParenGroups := takeZeroState(currentStates)
|
|
for _, val := range openParenGroups {
|
|
tempIndices[val].startIdx = i
|
|
}
|
|
for _, val := range closeParenGroups {
|
|
tempIndices[val].endIdx = i
|
|
}
|
|
tempStates = append(tempStates, zeroStates...)
|
|
num_appended := 0
|
|
for isZero == true {
|
|
zeroStates, isZero, openParenGroups, closeParenGroups = takeZeroState(tempStates)
|
|
for _, val := range openParenGroups {
|
|
tempIndices[val].startIdx = i
|
|
}
|
|
for _, val := range closeParenGroups {
|
|
tempIndices[val].endIdx = i
|
|
}
|
|
tempStates, num_appended = unique_append(tempStates, zeroStates...)
|
|
if num_appended == 0 { // Break if we haven't appended any more unique values
|
|
break
|
|
}
|
|
}
|
|
|
|
currentStates, _ = unique_append(currentStates, tempStates...)
|
|
tempStates = nil
|
|
|
|
// Take any transitions corresponding to current character
|
|
numStatesMatched := 0 // The number of states which had at least 1 match for this round
|
|
assertionFailed := false // Whether or not an assertion failed for this round
|
|
lastStateInList := false // Whether or not a last state was in our list of states
|
|
lastLookaroundInList := false // Whether or not a last state (that is a lookaround) was in our list of states
|
|
for _, state := range currentStates {
|
|
matches, numMatches := state.matchesFor(str, i)
|
|
if numMatches > 0 {
|
|
numStatesMatched++
|
|
tempStates = append(tempStates, matches...)
|
|
foundPath = true
|
|
}
|
|
if numMatches < 0 {
|
|
assertionFailed = true
|
|
}
|
|
if state.isLast {
|
|
if state.isLookaround() {
|
|
lastLookaroundInList = true
|
|
}
|
|
lastStateInList = true
|
|
}
|
|
}
|
|
|
|
if assertionFailed && numStatesMatched == 0 { // Nothing has matched and an assertion has failed
|
|
// If I'm being completely honest, I'm not sure why I have to check specifically for a _lookaround_
|
|
// state. The explanation below is my attempt to explain this behavior.
|
|
// If you replace 'lastLookaroundInList' with 'lastStateInList', one of the test cases fails.
|
|
//
|
|
// One of the states in our list was a last state and a lookaround. In this case, we
|
|
// don't abort upon failure of the assertion, because we have found
|
|
// another path to a final state.
|
|
// Even if the last state _was_ an assertion, we can use the previously
|
|
// saved indices to find a match.
|
|
if lastLookaroundInList {
|
|
break
|
|
} else {
|
|
if i == startingFrom {
|
|
i++
|
|
}
|
|
return false, []Group{}, i
|
|
}
|
|
}
|
|
if lastStateInList { // A last-state was in the list of states. add the matchIndex to our MatchIndex list
|
|
endIdx = i
|
|
tempIndices[0] = Group{startIdx, endIdx}
|
|
}
|
|
|
|
// Check if we can find a zero-length match
|
|
if foundPath == false {
|
|
if ok, _, _ := zeroMatchPossible(str, i, currentStates...); ok {
|
|
if tempIndices[0].isValid() == false {
|
|
tempIndices[0] = Group{startIdx, startIdx}
|
|
}
|
|
}
|
|
// If we haven't moved in the string, increment the counter by 1
|
|
// to ensure we don't keep trying the same string over and over.
|
|
// if i == startingFrom {
|
|
startIdx++
|
|
// i++
|
|
// }
|
|
if tempIndices.numValidGroups() > 0 && tempIndices[0].isValid() {
|
|
if tempIndices[0].startIdx == tempIndices[0].endIdx { // If we have a zero-length match, we have to shift the index at which we start. Otherwise we keep looking at the same paert of the string over and over.
|
|
return true, tempIndices, tempIndices[0].endIdx + 1
|
|
} else {
|
|
return true, tempIndices, tempIndices[0].endIdx
|
|
}
|
|
}
|
|
return false, []Group{}, startIdx
|
|
}
|
|
currentStates = make([]*State, len(tempStates))
|
|
copy(currentStates, tempStates)
|
|
tempStates = nil
|
|
|
|
i++
|
|
}
|
|
|
|
// End-of-string reached. Go to any 0-states, until there are no more 0-states to go to. Then check if any of our states are in the end position.
|
|
// This is the exact same algorithm used inside the loop, so I should probably put it in a function.
|
|
zeroStates, isZero, openParenGroups, closeParenGroups := takeZeroState(currentStates)
|
|
for _, val := range openParenGroups {
|
|
tempIndices[val].startIdx = i
|
|
}
|
|
for _, val := range closeParenGroups {
|
|
tempIndices[val].endIdx = i
|
|
}
|
|
tempStates = append(tempStates, zeroStates...)
|
|
num_appended := 0 // Number of unique states addded to tempStates
|
|
for isZero == true {
|
|
zeroStates, isZero, openParenGroups, closeParenGroups = takeZeroState(tempStates)
|
|
for _, val := range openParenGroups {
|
|
tempIndices[val].startIdx = i
|
|
}
|
|
for _, val := range closeParenGroups {
|
|
tempIndices[val].endIdx = i
|
|
}
|
|
tempStates, num_appended = unique_append(tempStates, zeroStates...)
|
|
if num_appended == 0 { // Break if we haven't appended any more unique values
|
|
break
|
|
}
|
|
}
|
|
|
|
currentStates = append(currentStates, tempStates...)
|
|
tempStates = nil
|
|
|
|
for _, state := range currentStates {
|
|
// Only add the match if the start index is in bounds. If the state has an assertion,
|
|
// make sure the assertion checks out.
|
|
if state.isLast && startIdx < len(str) {
|
|
if state.assert == NONE || state.checkAssertion(str, i) {
|
|
endIdx = i
|
|
tempIndices[0] = Group{startIdx, endIdx}
|
|
}
|
|
}
|
|
}
|
|
|
|
if tempIndices.numValidGroups() > 0 {
|
|
if tempIndices[0].startIdx == tempIndices[0].endIdx { // If we have a zero-length match, we have to shift the index at which we start. Otherwise we keep looking at the same paert of the string over and over.
|
|
return true, tempIndices, tempIndices[0].endIdx + 1
|
|
} else {
|
|
return true, tempIndices, tempIndices[0].endIdx
|
|
}
|
|
}
|
|
if startIdx == startingFrom { // Increment starting index if we haven't moved in the string. Prevents us from matching the same part of the string over and over.
|
|
startIdx++
|
|
}
|
|
return false, []Group{}, startIdx
|
|
}
|