Started rewrite of matching algorithm, got concatenation and alternation done, kleene and zero-state stuff is next

remotes/origin/implementPCREMatchingRules
Aadhavan Srinivasan 2 months ago
parent 5563a70568
commit 753e973d82

@ -1,7 +1,6 @@
package regex package regex
import ( import (
"container/heap"
"fmt" "fmt"
"slices" "slices"
"sort" "sort"
@ -267,16 +266,15 @@ func findAllSubmatchHelper(start *nfaState, str []rune, offset int, numGroups in
// chosen as the match for the entire string. // chosen as the match for the entire string.
// This allows us to pick the longest possible match (which is how greedy matching works). // This allows us to pick the longest possible match (which is how greedy matching works).
// COMMENT ABOVE IS CURRENTLY NOT UP-TO-DATE // COMMENT ABOVE IS CURRENTLY NOT UP-TO-DATE
tempIndices := newMatch(numGroups + 1) // tempIndices := newMatch(numGroups + 1)
foundPath := false // foundPath := false
startIdx := offset //startIdx := offset
endIdx := offset //endIdx := offset
currentStates := &priorityQueue{} currentStates := make([]*nfaState, 0)
heap.Init(currentStates) // tempStates := make([]*nfaState, 0) // Used to store states that should be used in next loop iteration
tempStates := make([]*nfaState, 0) // Used to store states that should be used in next loop iteration
i := offset // Index in string i := offset // Index in string
startingFrom := i // Store starting index //startingFrom := i // Store starting index
// If the first state is an assertion, makes sure the assertion // If the first state is an assertion, makes sure the assertion
// is true before we do _anything_ else. // is true before we do _anything_ else.
@ -287,214 +285,266 @@ func findAllSubmatchHelper(start *nfaState, str []rune, offset int, numGroups in
} }
} }
// Increment until we hit a character matching the start state (assuming not 0-state) // Increment until we hit a character matching the start state (assuming not 0-state)
if start.isEmpty == false { // if start.isEmpty == false {
for i < len(str) && !start.contentContains(str, i) { // for i < len(str) && !start.contentContains(str, i) {
i++ // i++
} // }
startIdx = i // startIdx = i
startingFrom = 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 // 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
} // }
start.threadGroups = newMatch(numGroups + 1) // start.threadGroups = newMatch(numGroups + 1)
// Check if the start state begins a group - if so, add the start index to our list // Check if the start state begins a group - if so, add the start index to our list
if start.groupBegin { //if start.groupBegin {
start.threadGroups[start.groupNum].StartIdx = i // start.threadGroups[start.groupNum].StartIdx = i
// tempIndices[start.groupNum].startIdx = i // tempIndices[start.groupNum].startIdx = i
} //}
start.threadSP = i start.threadSP = i
heap.Push(currentStates, newPriorQueueItem(start)) currentStates = append(currentStates, start)
var foundMatch bool
// Main loop // Main loop
for currentStates.Len() > 0 { for len(currentStates) > 0 {
currentState := heap.Pop(currentStates) currentState, _ := pop(&currentStates)
foundPath = false idx := currentState.threadSP
foundMatch = false
zeroStates := make([]*nfaState, 0)
// Keep taking zero-states, until there are no more left to take if currentState.threadGroups == nil {
// 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. currentState.threadGroups = newMatch(numGroups + 1)
topStateItem := currentStates.peek() currentState.threadGroups[0].StartIdx = idx
topState := topStateItem.(*priorQueueItem).state }
zeroStates, isZero := takeZeroState([]*nfaState{topState}, numGroups, i) if currentState.groupBegin {
tempStates = append(tempStates, zeroStates...) currentState.threadGroups[currentState.groupNum].StartIdx = idx
num_appended := 0 } else if currentState.groupEnd {
for isZero == true { currentState.threadGroups[currentState.groupNum].EndIdx = idx
zeroStates, isZero = takeZeroState(tempStates, numGroups, i) } else if currentState.isKleene {
tempStates, num_appended = uniqueAppend(tempStates, zeroStates...) // Append the
if num_appended == 0 { // Break if we haven't appended any more unique values } else if currentState.isAlternation {
break rightState := currentState.rightState
rightState.threadGroups = currentState.threadGroups
rightState.threadSP = currentState.threadSP
currentStates = append(currentStates, currentState.rightState)
leftState := currentState.leftState
leftState.threadGroups = currentState.threadGroups
leftState.threadSP = currentState.threadSP
currentStates = append(currentStates, currentState.leftState)
continue
} else if currentState.contentContains(str, idx) {
foundMatch = true
allMatches := make([]*nfaState, 0)
for _, v := range currentState.transitions {
allMatches = append(allMatches, v...)
}
slices.Reverse(allMatches)
for _, m := range allMatches {
m.threadGroups = currentState.threadGroups
if currentState.assert == noneAssert {
m.threadSP = idx + 1
} else {
m.threadSP = idx
} }
} }
if isZero == true { currentStates = append(currentStates, allMatches...)
currentStates.Pop()
} }
for _, state := range tempStates { if currentState.isLast && foundMatch { // Last state reached
heap.Push(currentStates, newPriorQueueItem(state)) currentState.threadGroups[0].EndIdx = idx + 1
} return true, currentState.threadGroups, idx + 1
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
var lastStatePtr *nfaState = nil // Pointer to the last-state, if it was found
lastLookaroundInList := false // Whether or not a last state (that is a lookaround) was in our list of states
for numStatesMatched == 0 && lastStateInList == false {
if currentStates.Len() == 0 {
break
}
stateItem := heap.Pop(currentStates)
state := stateItem.(*priorQueueItem).state
matches, numMatches := state.matchesFor(str, i)
if numMatches > 0 {
numStatesMatched++
tempStates = append([]*nfaState(nil), matches...)
foundPath = true
for _, m := range matches {
if m.threadGroups == nil {
m.threadGroups = newMatch(numGroups + 1)
} }
m.threadSP = state.threadSP + 1
copy(m.threadGroups, state.threadGroups)
} }
} return false, []Group{}, i + 1
if numMatches < 0 { // zeroStates := make([]*nfaState, 0)
assertionFailed = true // // 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.
if state.isLast { // topStateItem := currentStates.peek()
if state.isLookaround() { // topState := topStateItem.(*priorQueueItem).state
lastLookaroundInList = true // zeroStates, isZero := takeZeroState([]*nfaState{topState}, numGroups, i)
} // tempStates = append(tempStates, zeroStates...)
lastStateInList = true // num_appended := 0
lastStatePtr = state // for isZero == true {
} // zeroStates, isZero = takeZeroState(tempStates, numGroups, i)
} // tempStates, num_appended = uniqueAppend(tempStates, zeroStates...)
// if num_appended == 0 { // Break if we haven't appended any more unique values
if assertionFailed && numStatesMatched == 0 { // Nothing has matched and an assertion has failed // break
// 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. // if isZero == true {
// currentStates.Pop()
// }
// //
// One of the states in our list was a last state and a lookaround. In this case, we // for _, state := range tempStates {
// don't abort upon failure of the assertion, because we have found // heap.Push(currentStates, newPriorQueueItem(state))
// another path to a final state. // }
// Even if the last state _was_ an assertion, we can use the previously // tempStates = nil
// saved indices to find a match. //
if lastLookaroundInList { // // Take any transitions corresponding to current character
break // numStatesMatched := 0 // The number of states which had at least 1 match for this round
} else { // assertionFailed := false // Whether or not an assertion failed for this round
if i == startingFrom { // lastStateInList := false // Whether or not a last state was in our list of states
i++ // var lastStatePtr *nfaState = nil // Pointer to the last-state, if it was found
} // lastLookaroundInList := false // Whether or not a last state (that is a lookaround) was in our list of states
return false, []Group{}, i // for numStatesMatched == 0 && lastStateInList == false {
} // if currentStates.Len() == 0 {
} // break
// Check if we can find a state in our list that is: // }
// a. A last-state // stateItem := heap.Pop(currentStates)
// b. Empty // state := stateItem.(*priorQueueItem).state
// c. Doesn't assert anything // matches, numMatches := state.matchesFor(str, i)
for _, stateItem := range *currentStates { // if numMatches > 0 {
s := stateItem.state // numStatesMatched++
if s.isLast && s.isEmpty && s.assert == noneAssert { // tempStates = append([]*nfaState(nil), matches...)
lastStatePtr = s // foundPath = true
lastStateInList = true // for _, m := range matches {
} // if m.threadGroups == nil {
} // m.threadGroups = newMatch(numGroups + 1)
if lastStateInList && numStatesMatched == 0 { // A last-state was in the list of states. add the matchIndex to our MatchIndex list // }
for j := 1; j < numGroups+1; j++ { // m.threadSP = state.threadSP + 1
tempIndices[j] = lastStatePtr.threadGroups[j] // copy(m.threadGroups, state.threadGroups)
} // }
endIdx = i // }
tempIndices[0] = Group{startIdx, endIdx} // if numMatches < 0 {
if tempIndices[0].StartIdx == tempIndices[0].EndIdx { // assertionFailed = true
return true, tempIndices, tempIndices[0].EndIdx + 1 // }
} else { // if state.isLast {
return true, tempIndices, tempIndices[0].EndIdx // if state.isLookaround() {
} // lastLookaroundInList = true
} // }
// lastStateInList = true
// Check if we can find a zero-length match // lastStatePtr = state
if foundPath == false { // }
currentStatesList := funcMap(*currentStates, func(item *priorQueueItem) *nfaState { // }
return item.state //
}) // if assertionFailed && numStatesMatched == 0 { // Nothing has matched and an assertion has failed
if ok := zeroMatchPossible(str, i, numGroups, currentStatesList...); ok { // // If I'm being completely honest, I'm not sure why I have to check specifically for a _lookaround_
if tempIndices[0].IsValid() == false { // // state. The explanation below is my attempt to explain this behavior.
tempIndices[0] = Group{startIdx, startIdx} // // 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
// If we haven't moved in the string, increment the counter by 1 // // don't abort upon failure of the assertion, because we have found
// to ensure we don't keep trying the same string over and over. // // 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 { // if i == startingFrom {
startIdx++
// i++ // i++
// } // }
if tempIndices.numValidGroups() > 0 && tempIndices[0].IsValid() { // return false, []Group{}, i
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 { // // Check if we can find a state in our list that is:
return true, tempIndices, tempIndices[0].EndIdx // // a. A last-state
} // // b. Empty
} // // c. Doesn't assert anything
return false, []Group{}, startIdx // for _, stateItem := range *currentStates {
} // s := stateItem.state
currentStates = &priorityQueue{} // if s.isLast && s.isEmpty && s.assert == noneAssert {
slices.Reverse(tempStates) // lastStatePtr = s
for _, state := range tempStates { // lastStateInList = true
heap.Push(currentStates, newPriorQueueItem(state)) // }
} // }
tempStates = nil // if lastStateInList && numStatesMatched == 0 { // A last-state was in the list of states. add the matchIndex to our MatchIndex list
// for j := 1; j < numGroups+1; j++ {
i++ // tempIndices[j] = lastStatePtr.threadGroups[j]
} // }
// endIdx = 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. // tempIndices[0] = Group{startIdx, endIdx}
// This is the exact same algorithm used inside the loop, so I should probably put it in a function. // if tempIndices[0].StartIdx == tempIndices[0].EndIdx {
if currentStates.Len() > 0 { // return true, tempIndices, tempIndices[0].EndIdx + 1
topStateItem := currentStates.peek() // } else {
topState := topStateItem.(*priorQueueItem).state // return true, tempIndices, tempIndices[0].EndIdx
zeroStates, isZero := takeZeroState([]*nfaState{topState}, numGroups, i) // }
tempStates = append(tempStates, zeroStates...) // }
num_appended := 0 // Number of unique states addded to tempStates //
for isZero == true { // // Check if we can find a zero-length match
zeroStates, isZero = takeZeroState(tempStates, numGroups, i) // if foundPath == false {
tempStates, num_appended = uniqueAppend(tempStates, zeroStates...) // currentStatesList := funcMap(*currentStates, func(item *priorQueueItem) *nfaState {
if num_appended == 0 { // Break if we haven't appended any more unique values // return item.state
break // })
} // if ok := zeroMatchPossible(str, i, numGroups, currentStatesList...); ok {
} // if tempIndices[0].IsValid() == false {
} // tempIndices[0] = Group{startIdx, startIdx}
// }
for _, state := range tempStates { // }
heap.Push(currentStates, newPriorQueueItem(state)) // // 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.
tempStates = nil // // if i == startingFrom {
// startIdx++
for _, stateItem := range *currentStates { // // i++
state := stateItem.state // // }
// Only add the match if the start index is in bounds. If the state has an assertion, // if tempIndices.numValidGroups() > 0 && tempIndices[0].IsValid() {
// make sure the assertion checks out. // 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.
if state.isLast && i <= len(str) { // return true, tempIndices, tempIndices[0].EndIdx + 1
if state.assert == noneAssert || state.checkAssertion(str, i) { // } else {
for j := 1; j < numGroups+1; j++ { // return true, tempIndices, tempIndices[0].EndIdx
tempIndices[j] = state.threadGroups[j] // }
} // }
endIdx = i // return false, []Group{}, startIdx
tempIndices[0] = Group{startIdx, endIdx} // }
} // currentStates = &priorityQueue{}
} // slices.Reverse(tempStates)
} // for _, state := range tempStates {
// heap.Push(currentStates, newPriorQueueItem(state))
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. // tempStates = nil
return true, tempIndices, tempIndices[0].EndIdx + 1 //
} else { // i++
return true, tempIndices, tempIndices[0].EndIdx // }
} //
} // // 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.
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. // // This is the exact same algorithm used inside the loop, so I should probably put it in a function.
startIdx++ //
} // if currentStates.Len() > 0 {
return false, []Group{}, startIdx // topStateItem := currentStates.peek()
// topState := topStateItem.(*priorQueueItem).state
// zeroStates, isZero := takeZeroState([]*nfaState{topState}, numGroups, i)
// tempStates = append(tempStates, zeroStates...)
// num_appended := 0 // Number of unique states addded to tempStates
// for isZero == true {
// zeroStates, isZero = takeZeroState(tempStates, numGroups, i)
// tempStates, num_appended = uniqueAppend(tempStates, zeroStates...)
// if num_appended == 0 { // Break if we haven't appended any more unique values
// break
// }
// }
// }
//
// for _, state := range tempStates {
// heap.Push(currentStates, newPriorQueueItem(state))
// }
//
// tempStates = nil
//
// for _, stateItem := range *currentStates {
// state := stateItem.state
// // 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 && i <= len(str) {
// if state.assert == noneAssert || state.checkAssertion(str, i) {
// for j := 1; j < numGroups+1; j++ {
// tempIndices[j] = state.threadGroups[j]
// }
// 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
} }

Loading…
Cancel
Save