mirror of
https://github.com/go-gitea/gitea.git
synced 2024-12-21 22:14:50 +08:00
ad2642a8aa
* Implementation for calculating language statistics Impement saving code language statistics to database Implement rendering langauge stats Add primary laguage to show in repository list Implement repository stats indexer queue Add indexer test Refactor to use queue module * Do not timeout for queues
669 lines
17 KiB
Go
669 lines
17 KiB
Go
package rubex
|
|
|
|
/*
|
|
#cgo CFLAGS: -I/usr/local/include
|
|
#cgo LDFLAGS: -L/usr/local/lib -lonig
|
|
#include <stdlib.h>
|
|
#include <oniguruma.h>
|
|
#include "chelper.h"
|
|
*/
|
|
import "C"
|
|
|
|
import (
|
|
"bytes"
|
|
"errors"
|
|
"fmt"
|
|
"io"
|
|
"log"
|
|
"runtime"
|
|
"strconv"
|
|
"sync"
|
|
"unicode/utf8"
|
|
"unsafe"
|
|
)
|
|
|
|
type strRange []int
|
|
|
|
const numMatchStartSize = 4
|
|
const numReadBufferStartSize = 256
|
|
|
|
var mutex sync.Mutex
|
|
|
|
type MatchData struct {
|
|
count int
|
|
indexes [][]int32
|
|
}
|
|
|
|
type NamedGroupInfo map[string]int
|
|
|
|
type Regexp struct {
|
|
pattern string
|
|
regex C.OnigRegex
|
|
region *C.OnigRegion
|
|
encoding C.OnigEncoding
|
|
errorInfo *C.OnigErrorInfo
|
|
errorBuf *C.char
|
|
matchData *MatchData
|
|
namedGroupInfo NamedGroupInfo
|
|
}
|
|
|
|
// NewRegexp creates and initializes a new Regexp with the given pattern and option.
|
|
func NewRegexp(pattern string, option int) (re *Regexp, err error) {
|
|
return initRegexp(&Regexp{pattern: pattern, encoding: C.ONIG_ENCODING_UTF8}, option)
|
|
}
|
|
|
|
// NewRegexpASCII is equivalent to NewRegexp, but with the encoding restricted to ASCII.
|
|
func NewRegexpASCII(pattern string, option int) (re *Regexp, err error) {
|
|
return initRegexp(&Regexp{pattern: pattern, encoding: C.ONIG_ENCODING_ASCII}, option)
|
|
}
|
|
|
|
func initRegexp(re *Regexp, option int) (*Regexp, error) {
|
|
var err error
|
|
patternCharPtr := C.CString(re.pattern)
|
|
defer C.free(unsafe.Pointer(patternCharPtr))
|
|
mutex.Lock()
|
|
defer mutex.Unlock()
|
|
errorCode := C.NewOnigRegex(patternCharPtr, C.int(len(re.pattern)), C.int(option), &re.regex, &re.region, &re.encoding, &re.errorInfo, &re.errorBuf)
|
|
if errorCode != C.ONIG_NORMAL {
|
|
err = errors.New(C.GoString(re.errorBuf))
|
|
} else {
|
|
err = nil
|
|
numCapturesInPattern := int(C.onig_number_of_captures(re.regex)) + 1
|
|
re.matchData = &MatchData{}
|
|
re.matchData.indexes = make([][]int32, numMatchStartSize)
|
|
for i := 0; i < numMatchStartSize; i++ {
|
|
re.matchData.indexes[i] = make([]int32, numCapturesInPattern*2)
|
|
}
|
|
re.namedGroupInfo = re.getNamedGroupInfo()
|
|
runtime.SetFinalizer(re, (*Regexp).Free)
|
|
}
|
|
return re, err
|
|
}
|
|
|
|
func Compile(str string) (*Regexp, error) {
|
|
return NewRegexp(str, ONIG_OPTION_DEFAULT)
|
|
}
|
|
|
|
func MustCompile(str string) *Regexp {
|
|
regexp, error := NewRegexp(str, ONIG_OPTION_DEFAULT)
|
|
if error != nil {
|
|
panic("regexp: compiling " + str + ": " + error.Error())
|
|
}
|
|
return regexp
|
|
}
|
|
|
|
func CompileWithOption(str string, option int) (*Regexp, error) {
|
|
return NewRegexp(str, option)
|
|
}
|
|
|
|
func MustCompileWithOption(str string, option int) *Regexp {
|
|
regexp, error := NewRegexp(str, option)
|
|
if error != nil {
|
|
panic("regexp: compiling " + str + ": " + error.Error())
|
|
}
|
|
return regexp
|
|
}
|
|
|
|
// MustCompileASCII is equivalent to MustCompile, but with the encoding restricted to ASCII.
|
|
func MustCompileASCII(str string) *Regexp {
|
|
regexp, error := NewRegexpASCII(str, ONIG_OPTION_DEFAULT)
|
|
if error != nil {
|
|
panic("regexp: compiling " + str + ": " + error.Error())
|
|
}
|
|
return regexp
|
|
}
|
|
|
|
func (re *Regexp) Free() {
|
|
mutex.Lock()
|
|
if re.regex != nil {
|
|
C.onig_free(re.regex)
|
|
re.regex = nil
|
|
}
|
|
if re.region != nil {
|
|
C.onig_region_free(re.region, 1)
|
|
re.region = nil
|
|
}
|
|
mutex.Unlock()
|
|
if re.errorInfo != nil {
|
|
C.free(unsafe.Pointer(re.errorInfo))
|
|
re.errorInfo = nil
|
|
}
|
|
if re.errorBuf != nil {
|
|
C.free(unsafe.Pointer(re.errorBuf))
|
|
re.errorBuf = nil
|
|
}
|
|
}
|
|
|
|
func (re *Regexp) getNamedGroupInfo() (namedGroupInfo NamedGroupInfo) {
|
|
numNamedGroups := int(C.onig_number_of_names(re.regex))
|
|
//when any named capture exisits, there is no numbered capture even if there are unnamed captures
|
|
if numNamedGroups > 0 {
|
|
namedGroupInfo = make(map[string]int)
|
|
//try to get the names
|
|
bufferSize := len(re.pattern) * 2
|
|
nameBuffer := make([]byte, bufferSize)
|
|
groupNumbers := make([]int32, numNamedGroups)
|
|
bufferPtr := unsafe.Pointer(&nameBuffer[0])
|
|
numbersPtr := unsafe.Pointer(&groupNumbers[0])
|
|
length := int(C.GetCaptureNames(re.regex, bufferPtr, (C.int)(bufferSize), (*C.int)(numbersPtr)))
|
|
if length > 0 {
|
|
namesAsBytes := bytes.Split(nameBuffer[:length], ([]byte)(";"))
|
|
if len(namesAsBytes) != numNamedGroups {
|
|
log.Fatalf("the number of named groups (%d) does not match the number names found (%d)\n", numNamedGroups, len(namesAsBytes))
|
|
}
|
|
for i, nameAsBytes := range namesAsBytes {
|
|
name := string(nameAsBytes)
|
|
namedGroupInfo[name] = int(groupNumbers[i])
|
|
}
|
|
} else {
|
|
log.Fatalf("could not get the capture group names from %q", re.String())
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
func (re *Regexp) groupNameToId(name string) (id int) {
|
|
if re.namedGroupInfo == nil {
|
|
id = ONIGERR_UNDEFINED_NAME_REFERENCE
|
|
} else {
|
|
id = re.namedGroupInfo[name]
|
|
}
|
|
return
|
|
}
|
|
|
|
func (re *Regexp) processMatch(numCaptures int) (match []int32) {
|
|
if numCaptures <= 0 {
|
|
panic("cannot have 0 captures when processing a match")
|
|
}
|
|
matchData := re.matchData
|
|
return matchData.indexes[matchData.count][:numCaptures*2]
|
|
}
|
|
|
|
func (re *Regexp) ClearMatchData() {
|
|
matchData := re.matchData
|
|
matchData.count = 0
|
|
}
|
|
|
|
func (re *Regexp) find(b []byte, n int, offset int) (match []int) {
|
|
if n == 0 {
|
|
b = []byte{0}
|
|
}
|
|
ptr := unsafe.Pointer(&b[0])
|
|
matchData := re.matchData
|
|
capturesPtr := unsafe.Pointer(&(matchData.indexes[matchData.count][0]))
|
|
numCaptures := int32(0)
|
|
numCapturesPtr := unsafe.Pointer(&numCaptures)
|
|
pos := int(C.SearchOnigRegex((ptr), C.int(n), C.int(offset), C.int(ONIG_OPTION_DEFAULT), re.regex, re.region, re.errorInfo, (*C.char)(nil), (*C.int)(capturesPtr), (*C.int)(numCapturesPtr)))
|
|
if pos >= 0 {
|
|
if numCaptures <= 0 {
|
|
panic("cannot have 0 captures when processing a match")
|
|
}
|
|
match2 := matchData.indexes[matchData.count][:numCaptures*2]
|
|
match = make([]int, len(match2))
|
|
for i := range match2 {
|
|
match[i] = int(match2[i])
|
|
}
|
|
numCapturesInPattern := int32(C.onig_number_of_captures(re.regex)) + 1
|
|
if numCapturesInPattern != numCaptures {
|
|
log.Fatalf("expected %d captures but got %d\n", numCapturesInPattern, numCaptures)
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
func getCapture(b []byte, beg int, end int) []byte {
|
|
if beg < 0 || end < 0 {
|
|
return nil
|
|
}
|
|
return b[beg:end]
|
|
}
|
|
|
|
func (re *Regexp) match(b []byte, n int, offset int) bool {
|
|
re.ClearMatchData()
|
|
if n == 0 {
|
|
b = []byte{0}
|
|
}
|
|
ptr := unsafe.Pointer(&b[0])
|
|
pos := int(C.SearchOnigRegex((ptr), C.int(n), C.int(offset), C.int(ONIG_OPTION_DEFAULT), re.regex, re.region, re.errorInfo, (*C.char)(nil), (*C.int)(nil), (*C.int)(nil)))
|
|
return pos >= 0
|
|
}
|
|
|
|
func (re *Regexp) findAll(b []byte, n int) (matches [][]int) {
|
|
re.ClearMatchData()
|
|
|
|
if n < 0 {
|
|
n = len(b)
|
|
}
|
|
matchData := re.matchData
|
|
offset := 0
|
|
for offset <= n {
|
|
if matchData.count >= len(matchData.indexes) {
|
|
length := len(matchData.indexes[0])
|
|
matchData.indexes = append(matchData.indexes, make([]int32, length))
|
|
}
|
|
if match := re.find(b, n, offset); len(match) > 0 {
|
|
matchData.count += 1
|
|
//move offset to the ending index of the current match and prepare to find the next non-overlapping match
|
|
offset = match[1]
|
|
//if match[0] == match[1], it means the current match does not advance the search. we need to exit the loop to avoid getting stuck here.
|
|
if match[0] == match[1] {
|
|
if offset < n && offset >= 0 {
|
|
//there are more bytes, so move offset by a word
|
|
_, width := utf8.DecodeRune(b[offset:])
|
|
offset += width
|
|
} else {
|
|
//search is over, exit loop
|
|
break
|
|
}
|
|
}
|
|
} else {
|
|
break
|
|
}
|
|
}
|
|
matches2 := matchData.indexes[:matchData.count]
|
|
matches = make([][]int, len(matches2))
|
|
for i, v := range matches2 {
|
|
matches[i] = make([]int, len(v))
|
|
for j, v2 := range v {
|
|
matches[i][j] = int(v2)
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
func (re *Regexp) FindIndex(b []byte) []int {
|
|
re.ClearMatchData()
|
|
match := re.find(b, len(b), 0)
|
|
if len(match) == 0 {
|
|
return nil
|
|
}
|
|
return match[:2]
|
|
}
|
|
|
|
func (re *Regexp) Find(b []byte) []byte {
|
|
loc := re.FindIndex(b)
|
|
if loc == nil {
|
|
return nil
|
|
}
|
|
return getCapture(b, loc[0], loc[1])
|
|
}
|
|
|
|
func (re *Regexp) FindString(s string) string {
|
|
b := []byte(s)
|
|
mb := re.Find(b)
|
|
if mb == nil {
|
|
return ""
|
|
}
|
|
return string(mb)
|
|
}
|
|
|
|
func (re *Regexp) FindStringIndex(s string) []int {
|
|
b := []byte(s)
|
|
return re.FindIndex(b)
|
|
}
|
|
|
|
func (re *Regexp) FindAllIndex(b []byte, n int) [][]int {
|
|
matches := re.findAll(b, n)
|
|
if len(matches) == 0 {
|
|
return nil
|
|
}
|
|
return matches
|
|
}
|
|
|
|
func (re *Regexp) FindAll(b []byte, n int) [][]byte {
|
|
matches := re.FindAllIndex(b, n)
|
|
if matches == nil {
|
|
return nil
|
|
}
|
|
matchBytes := make([][]byte, 0, len(matches))
|
|
for _, match := range matches {
|
|
matchBytes = append(matchBytes, getCapture(b, match[0], match[1]))
|
|
}
|
|
return matchBytes
|
|
}
|
|
|
|
func (re *Regexp) FindAllString(s string, n int) []string {
|
|
b := []byte(s)
|
|
matches := re.FindAllIndex(b, n)
|
|
if matches == nil {
|
|
return nil
|
|
}
|
|
matchStrings := make([]string, 0, len(matches))
|
|
for _, match := range matches {
|
|
m := getCapture(b, match[0], match[1])
|
|
if m == nil {
|
|
matchStrings = append(matchStrings, "")
|
|
} else {
|
|
matchStrings = append(matchStrings, string(m))
|
|
}
|
|
}
|
|
return matchStrings
|
|
|
|
}
|
|
|
|
func (re *Regexp) FindAllStringIndex(s string, n int) [][]int {
|
|
b := []byte(s)
|
|
return re.FindAllIndex(b, n)
|
|
}
|
|
|
|
func (re *Regexp) findSubmatchIndex(b []byte) (match []int) {
|
|
re.ClearMatchData()
|
|
match = re.find(b, len(b), 0)
|
|
return
|
|
}
|
|
|
|
func (re *Regexp) FindSubmatchIndex(b []byte) []int {
|
|
match := re.findSubmatchIndex(b)
|
|
if len(match) == 0 {
|
|
return nil
|
|
}
|
|
return match
|
|
}
|
|
|
|
func (re *Regexp) FindSubmatch(b []byte) [][]byte {
|
|
match := re.findSubmatchIndex(b)
|
|
if match == nil {
|
|
return nil
|
|
}
|
|
length := len(match) / 2
|
|
if length == 0 {
|
|
return nil
|
|
}
|
|
results := make([][]byte, 0, length)
|
|
for i := 0; i < length; i++ {
|
|
results = append(results, getCapture(b, match[2*i], match[2*i+1]))
|
|
}
|
|
return results
|
|
}
|
|
|
|
func (re *Regexp) FindStringSubmatch(s string) []string {
|
|
b := []byte(s)
|
|
match := re.findSubmatchIndex(b)
|
|
if match == nil {
|
|
return nil
|
|
}
|
|
length := len(match) / 2
|
|
if length == 0 {
|
|
return nil
|
|
}
|
|
|
|
results := make([]string, 0, length)
|
|
for i := 0; i < length; i++ {
|
|
cap := getCapture(b, match[2*i], match[2*i+1])
|
|
if cap == nil {
|
|
results = append(results, "")
|
|
} else {
|
|
results = append(results, string(cap))
|
|
}
|
|
}
|
|
return results
|
|
}
|
|
|
|
func (re *Regexp) FindStringSubmatchIndex(s string) []int {
|
|
b := []byte(s)
|
|
return re.FindSubmatchIndex(b)
|
|
}
|
|
|
|
func (re *Regexp) FindAllSubmatchIndex(b []byte, n int) [][]int {
|
|
matches := re.findAll(b, n)
|
|
if len(matches) == 0 {
|
|
return nil
|
|
}
|
|
return matches
|
|
}
|
|
|
|
func (re *Regexp) FindAllSubmatch(b []byte, n int) [][][]byte {
|
|
matches := re.findAll(b, n)
|
|
if len(matches) == 0 {
|
|
return nil
|
|
}
|
|
allCapturedBytes := make([][][]byte, 0, len(matches))
|
|
for _, match := range matches {
|
|
length := len(match) / 2
|
|
capturedBytes := make([][]byte, 0, length)
|
|
for i := 0; i < length; i++ {
|
|
capturedBytes = append(capturedBytes, getCapture(b, match[2*i], match[2*i+1]))
|
|
}
|
|
allCapturedBytes = append(allCapturedBytes, capturedBytes)
|
|
}
|
|
|
|
return allCapturedBytes
|
|
}
|
|
|
|
func (re *Regexp) FindAllStringSubmatch(s string, n int) [][]string {
|
|
b := []byte(s)
|
|
matches := re.findAll(b, n)
|
|
if len(matches) == 0 {
|
|
return nil
|
|
}
|
|
allCapturedStrings := make([][]string, 0, len(matches))
|
|
for _, match := range matches {
|
|
length := len(match) / 2
|
|
capturedStrings := make([]string, 0, length)
|
|
for i := 0; i < length; i++ {
|
|
cap := getCapture(b, match[2*i], match[2*i+1])
|
|
if cap == nil {
|
|
capturedStrings = append(capturedStrings, "")
|
|
} else {
|
|
capturedStrings = append(capturedStrings, string(cap))
|
|
}
|
|
}
|
|
allCapturedStrings = append(allCapturedStrings, capturedStrings)
|
|
}
|
|
return allCapturedStrings
|
|
}
|
|
|
|
func (re *Regexp) FindAllStringSubmatchIndex(s string, n int) [][]int {
|
|
b := []byte(s)
|
|
return re.FindAllSubmatchIndex(b, n)
|
|
}
|
|
|
|
func (re *Regexp) Match(b []byte) bool {
|
|
return re.match(b, len(b), 0)
|
|
}
|
|
|
|
func (re *Regexp) MatchString(s string) bool {
|
|
b := []byte(s)
|
|
return re.Match(b)
|
|
}
|
|
|
|
func (re *Regexp) NumSubexp() int {
|
|
return (int)(C.onig_number_of_captures(re.regex))
|
|
}
|
|
|
|
func (re *Regexp) getNamedCapture(name []byte, capturedBytes [][]byte) []byte {
|
|
nameStr := string(name)
|
|
capNum := re.groupNameToId(nameStr)
|
|
if capNum < 0 || capNum >= len(capturedBytes) {
|
|
panic(fmt.Sprintf("capture group name (%q) has error\n", nameStr))
|
|
}
|
|
return capturedBytes[capNum]
|
|
}
|
|
|
|
func (re *Regexp) getNumberedCapture(num int, capturedBytes [][]byte) []byte {
|
|
//when named capture groups exist, numbered capture groups returns ""
|
|
if re.namedGroupInfo == nil && num <= (len(capturedBytes)-1) && num >= 0 {
|
|
return capturedBytes[num]
|
|
}
|
|
return ([]byte)("")
|
|
}
|
|
|
|
func fillCapturedValues(repl []byte, _ []byte, capturedBytes map[string][]byte) []byte {
|
|
replLen := len(repl)
|
|
newRepl := make([]byte, 0, replLen*3)
|
|
inEscapeMode := false
|
|
inGroupNameMode := false
|
|
groupName := make([]byte, 0, replLen)
|
|
for index := 0; index < replLen; index += 1 {
|
|
ch := repl[index]
|
|
if inGroupNameMode && ch == byte('<') {
|
|
} else if inGroupNameMode && ch == byte('>') {
|
|
inGroupNameMode = false
|
|
groupNameStr := string(groupName)
|
|
capBytes := capturedBytes[groupNameStr]
|
|
newRepl = append(newRepl, capBytes...)
|
|
groupName = groupName[:0] //reset the name
|
|
} else if inGroupNameMode {
|
|
groupName = append(groupName, ch)
|
|
} else if inEscapeMode && ch <= byte('9') && byte('1') <= ch {
|
|
capNumStr := string(ch)
|
|
capBytes := capturedBytes[capNumStr]
|
|
newRepl = append(newRepl, capBytes...)
|
|
} else if inEscapeMode && ch == byte('k') && (index+1) < replLen && repl[index+1] == byte('<') {
|
|
inGroupNameMode = true
|
|
inEscapeMode = false
|
|
index += 1 //bypass the next char '<'
|
|
} else if inEscapeMode {
|
|
newRepl = append(newRepl, '\\')
|
|
newRepl = append(newRepl, ch)
|
|
} else if ch != '\\' {
|
|
newRepl = append(newRepl, ch)
|
|
}
|
|
if ch == byte('\\') || inEscapeMode {
|
|
inEscapeMode = !inEscapeMode
|
|
}
|
|
}
|
|
return newRepl
|
|
}
|
|
|
|
func (re *Regexp) replaceAll(src, repl []byte, replFunc func([]byte, []byte, map[string][]byte) []byte) []byte {
|
|
srcLen := len(src)
|
|
matches := re.findAll(src, srcLen)
|
|
if len(matches) == 0 {
|
|
return src
|
|
}
|
|
dest := make([]byte, 0, srcLen)
|
|
for i, match := range matches {
|
|
length := len(match) / 2
|
|
capturedBytes := make(map[string][]byte)
|
|
if re.namedGroupInfo == nil {
|
|
for j := 0; j < length; j++ {
|
|
capturedBytes[strconv.Itoa(j)] = getCapture(src, match[2*j], match[2*j+1])
|
|
}
|
|
} else {
|
|
for name, j := range re.namedGroupInfo {
|
|
capturedBytes[name] = getCapture(src, match[2*j], match[2*j+1])
|
|
}
|
|
}
|
|
matchBytes := getCapture(src, match[0], match[1])
|
|
newRepl := replFunc(repl, matchBytes, capturedBytes)
|
|
prevEnd := 0
|
|
if i > 0 {
|
|
prevMatch := matches[i-1][:2]
|
|
prevEnd = prevMatch[1]
|
|
}
|
|
if match[0] > prevEnd && prevEnd >= 0 && match[0] <= srcLen {
|
|
dest = append(dest, src[prevEnd:match[0]]...)
|
|
}
|
|
dest = append(dest, newRepl...)
|
|
}
|
|
lastEnd := matches[len(matches)-1][1]
|
|
if lastEnd < srcLen && lastEnd >= 0 {
|
|
dest = append(dest, src[lastEnd:]...)
|
|
}
|
|
return dest
|
|
}
|
|
|
|
func (re *Regexp) ReplaceAll(src, repl []byte) []byte {
|
|
return re.replaceAll(src, repl, fillCapturedValues)
|
|
}
|
|
|
|
func (re *Regexp) ReplaceAllFunc(src []byte, repl func([]byte) []byte) []byte {
|
|
return re.replaceAll(src, []byte(""), func(_ []byte, matchBytes []byte, _ map[string][]byte) []byte {
|
|
return repl(matchBytes)
|
|
})
|
|
}
|
|
|
|
func (re *Regexp) ReplaceAllString(src, repl string) string {
|
|
return string(re.ReplaceAll([]byte(src), []byte(repl)))
|
|
}
|
|
|
|
func (re *Regexp) ReplaceAllStringFunc(src string, repl func(string) string) string {
|
|
srcB := []byte(src)
|
|
destB := re.replaceAll(srcB, []byte(""), func(_ []byte, matchBytes []byte, _ map[string][]byte) []byte {
|
|
return []byte(repl(string(matchBytes)))
|
|
})
|
|
return string(destB)
|
|
}
|
|
|
|
func (re *Regexp) String() string {
|
|
return re.pattern
|
|
}
|
|
|
|
func grow_buffer(b []byte, offset int, n int) []byte {
|
|
if offset+n > cap(b) {
|
|
buf := make([]byte, 2*cap(b)+n)
|
|
copy(buf, b[:offset])
|
|
return buf
|
|
}
|
|
return b
|
|
}
|
|
|
|
func fromReader(r io.RuneReader) []byte {
|
|
b := make([]byte, numReadBufferStartSize)
|
|
offset := 0
|
|
var err error = nil
|
|
for err == nil {
|
|
rune, runeWidth, err := r.ReadRune()
|
|
if err == nil {
|
|
b = grow_buffer(b, offset, runeWidth)
|
|
writeWidth := utf8.EncodeRune(b[offset:], rune)
|
|
if runeWidth != writeWidth {
|
|
panic("reading rune width not equal to the written rune width")
|
|
}
|
|
offset += writeWidth
|
|
} else {
|
|
break
|
|
}
|
|
}
|
|
return b[:offset]
|
|
}
|
|
|
|
func (re *Regexp) FindReaderIndex(r io.RuneReader) []int {
|
|
b := fromReader(r)
|
|
return re.FindIndex(b)
|
|
}
|
|
|
|
func (re *Regexp) FindReaderSubmatchIndex(r io.RuneReader) []int {
|
|
b := fromReader(r)
|
|
return re.FindSubmatchIndex(b)
|
|
}
|
|
|
|
func (re *Regexp) MatchReader(r io.RuneReader) bool {
|
|
b := fromReader(r)
|
|
return re.Match(b)
|
|
}
|
|
|
|
func (re *Regexp) LiteralPrefix() (prefix string, complete bool) {
|
|
//no easy way to implement this
|
|
return "", false
|
|
}
|
|
|
|
func MatchString(pattern string, s string) (matched bool, error error) {
|
|
re, err := Compile(pattern)
|
|
if err != nil {
|
|
return false, err
|
|
}
|
|
return re.MatchString(s), nil
|
|
}
|
|
|
|
func (re *Regexp) Gsub(src, repl string) string {
|
|
srcBytes := ([]byte)(src)
|
|
replBytes := ([]byte)(repl)
|
|
replaced := re.replaceAll(srcBytes, replBytes, fillCapturedValues)
|
|
return string(replaced)
|
|
}
|
|
|
|
func (re *Regexp) GsubFunc(src string, replFunc func(string, map[string]string) string) string {
|
|
srcBytes := ([]byte)(src)
|
|
replaced := re.replaceAll(srcBytes, nil, func(_ []byte, matchBytes []byte, capturedBytes map[string][]byte) []byte {
|
|
capturedStrings := make(map[string]string)
|
|
for name, capBytes := range capturedBytes {
|
|
capturedStrings[name] = string(capBytes)
|
|
}
|
|
matchString := string(matchBytes)
|
|
return ([]byte)(replFunc(matchString, capturedStrings))
|
|
})
|
|
return string(replaced)
|
|
}
|