Kleingrep is a regular expression engine, providing a library and command-line tool written in Go.
It aims to provide a more featureful engine, compared to the one in Go's
[regexp](https://pkg.go.dev/regexp), while retaining some semblance of efficiency.
The engine does __not__ use backtracking, relying on the NFA-based method described in
[Russ Cox's articles](https://swtch.com/~rsc/regexp). As such, it is immune to catastrophic backtracking.
It also includes features not present in regexp, such as lookarounds and backreferences.
### Syntax
The syntax is, for the most part, a superset of Go's regexp. A full overview of the syntax can be found [here](https://pkg.go.dev/gitea.twomorecents.org/Rockingcool/kleingrep/regex#hdr-Syntax).
__For more information, see https://pkg.go.dev/gitea.twomorecents.org/Rockingcool/kleingrep/regex__.
### How it works
I've written about the inner workings of the engine [on my blog](https://twomorecents.org/writing-regex-engine/index.html).
if*recursiveFlag||len(flag.Args())>2{// If we have 2 args, then we're only searching 1 file. We should only print the filename if there's more than 1 file.
// Newline after every match - only if -o is enabled and -v is disabled.
if*onlyFlag&&!(*invertFlag){
formatchIdxNum,idx:=rangematchIndices{
ifmatchIdxNum<len(matchIndices)-1{// Only print a newline afte printing a match, if there are multiple matches on the line, and we aren't on the last one. This is because the newline that gets added at the end will take care of that.
ifi+1==idx[0].EndIdx{// End index is one more than last index of match
fmt.Fprintf(out,"\n")
break
}
}
}
}
}else{
if!(*onlyFlag){
fmt.Fprintf(out,"%c",c)
}
}
}
}
err=out.Flush()
iferr!=nil{
panic(err)
}
// If the last character in the string wasn't a newline, AND we either have don't -o set or we do (and we've matched something), then print a newline
i+=(numDigits-1)// I have to move back a step, so that I can add a concatenation operator if necessary, and so that the increment at the bottom of the loop works as intended
}elseifunicode.IsDigit(re_runes[i]){// Any other number - backreference
numDigits:=1
fori+numDigits<len(re_runes)&&unicode.IsDigit(re_runes[i+numDigits]){// Skip while we see a digit
ifi<len(re_runes)&&(re_runes[i]!='('&&re_runes[i]!=nonCapLparenRune&&re_runes[i]!='|'&&re_runes[i]!='\\')||(i>0&&re_runes[i-1]=='\\'){// Every character should be concatenated if it is escaped
parenIndices:=make([]Group,0)// I really shouldn't be using Group here, because that's strictly for matching purposes, but its a convenient way to store the indices of the opening and closing parens.
parenIndices=append(parenIndices,Group{0,0})// I append a weird value here, because the 0-th group doesn't have any parens. This way, the 1st group will be at index 1, 2nd at 2 ...
i+=numDigitsParsed-1// Shift forward by the number of digits that were parsed. Move back one character, because the loop increment will move us back to the next character automatically
for(i+numDigitsParsed)<len(re_postfix)-1&&isOctal(re_postfix[i+numDigitsParsed])&&numDigitsParsed<=4{// The '-1' exists, because even in the worst case (the character class extends till the end), the last character must be a closing bracket (and nothing else)
for(i+numDigitsParsed)<len(re_postfix)-1&&isOctal(re_postfix[i+numDigitsParsed])&&numDigitsParsed<=3{// The '-1' exists, because even in the worst case (the character class extends till the end), the last character must be a closing bracket (and nothing else)
}elseif!currentState.isAlternation&&!currentState.isKleene&&!currentState.isQuestion&&!currentState.isBackreference &&!currentState.groupBegin &&!currentState.groupEnd&¤tState.assert==noneAssert{// Normal character
}elseif!currentState.isAlternation&&!currentState.isKleene&&!currentState.isQuestion&&!currentState.groupBegin &&!currentState.groupEnd&¤tState.assert==noneAssert{// Normal character
isKleenebool// Identifies whether current node is a 0-state representing Kleene star
isQuestionbool// Identifies whether current node is a 0-state representing the question operator
isAlternationbool// Identifies whether current node is a 0-state representing an alternation
isLazybool// Only for split states - Identifies whether or not to flip the order of branches (try one branch before the other)
splitState*nfaState// Only for alternation states - the 'other' branch of the alternation ('next' is the first)
assertassertType// Type of assertion of current node - NONE means that the node doesn't assert anything
allCharsbool// Whether or not the state represents all characters (eg. a 'dot' metacharacter). A 'dot' node doesn't store any contents directly, as it would take up too much space
@ -45,11 +44,9 @@ type nfaState struct {
groupBeginbool// Whether or not the node starts a capturing group
groupEndbool// Whether or not the node ends a capturing group
groupNumint// Which capturing group the node starts / ends
isBackreferencebool// Whether or not current node is backreference
referredGroupint// If current node is a backreference, the node that it points to
// The following properties depend on the current match - I should think about resetting them for every match.
threadGroups []Group// Assuming that a state is part of a 'thread' in the matching process, this array stores the indices of capturing groups in the current thread. As matches are found for this state, its groups will be copied over.
threadBackref int// If current node is a backreference, how many characters to look forward into the referred group
zeroMatchFoundbool// Whether or not the state has been used for a zero-length match - only relevant for zero states
threadGroups []Group// Assuming that a state is part of a 'thread' in the matching process, this array stores the indices of capturing groups in the current thread. As matches are found for this state, its groups will be copied over.
varinfinite_repsint=-1// Represents infinite reps eg. the end range in {5,}
// This represents a node in the postfix representation of the expression
typepostfixNodestruct{
nodetypenodeType
contents[]rune// Contents of the node
startRepsint// Minimum number of times the node should be repeated - used with numeric specifiers
endRepsint// Maximum number of times the node should be repeated - used with numeric specifiers
allCharsbool// Whether or not the current node represents all characters (eg. dot metacharacter)
except[]postfixNode// For inverted character classes, we match every unicode character _except_ a few. In this case, allChars is true and the exceptions are placed here.
lookaroundSignint// ONLY USED WHEN nodetype == ASSERTION. Whether we have a positive or negative lookaround.
lookaroundDirint// Lookbehind or lookahead
nodeContents[]postfixNode// ONLY USED WHEN nodetype == CHARCLASS. Holds all the nodes inside the given CHARCLASS node.
referencedGroupint// ONLY USED WHEN nodetype == backreferenceNode. Holds the group which this one refers to. After parsing is done, the expression will be rewritten eg. (a)\1 will become (a)(a). So the return value of ShuntingYard() shouldn't contain a backreferenceNode.
isLazybool// ONLY USED WHEN nodetype == kleene or question
nodetypenodeType
contents[]rune// Contents of the node
startRepsint// Minimum number of times the node should be repeated - used with numeric specifiers
endRepsint// Maximum number of times the node should be repeated - used with numeric specifiers
allCharsbool// Whether or not the current node represents all characters (eg. dot metacharacter)
except[]postfixNode// For inverted character classes, we match every unicode character _except_ a few. In this case, allChars is true and the exceptions are placed here.
lookaroundSignint// ONLY USED WHEN nodetype == ASSERTION. Whether we have a positive or negative lookaround.
lookaroundDirint// Lookbehind or lookahead
nodeContents[]postfixNode// ONLY USED WHEN nodetype == CHARCLASS. Holds all the nodes inside the given CHARCLASS node.
}
// Converts the given list of postfixNodes to one node of type CHARCLASS.
// groupIncrement := 0 // If we have a backreference before the group its referring to, then the group its referring to will have its group number incremented.
// rtv = slices.Concat(rtv, nodes[parenIndices[node.referencedGroup].StartIdx:parenIndices[node.referencedGroup].EndIdx+1]) // Add all the nodes in the group to rtv
// numGroups += 1
// if i < parenIndices[node.referencedGroup].StartIdx {
{`a*?`,nil,`-`,nil},// non-greedy operators are not supported
// Numeric range tests - this is a feature that I added, and doesn't exist
// in any other mainstream regex engine
@ -538,30 +520,6 @@ var reTests = []struct {
{`<389-400`,nil,`-`,nil},
{`<389-400>`,nil,`391`,[]Group{{0,3}}},
{`\b<1-10000>\b`,nil,`America declared independence in 1776.`,[]Group{{33,37}}},
{`\p{Tamil}+`,nil,`உயிரெழுத்து`,[]Group{{0,11}}},// Each letter and matra is counted as a separate rune, so 'u', 'ya', 'e (matra), 'ra', 'e (matra)', 'zha', (oo (matra), 'tha', 'ith', 'tha', 'oo (matra)'.