mirror of
https://github.com/42wim/matterbridge
synced 2024-11-19 03:25:33 +00:00
254 lines
6.8 KiB
Go
254 lines
6.8 KiB
Go
// Copyright (c) 2017-present Mattermost, Inc. All Rights Reserved.
|
|
// See License.txt for license information.
|
|
|
|
package markdown
|
|
|
|
import (
|
|
"regexp"
|
|
"strings"
|
|
"unicode"
|
|
"unicode/utf8"
|
|
)
|
|
|
|
// Based off of extensions/autolink.c from https://github.com/github/cmark
|
|
|
|
var (
|
|
DefaultUrlSchemes = []string{"http", "https", "ftp", "mailto", "tel"}
|
|
)
|
|
|
|
// Given a string with a w at the given position, tries to parse and return a range containing a www link.
|
|
// if one exists. If the text at the given position isn't a link, returns an empty string. Equivalent to
|
|
// www_match from the reference code.
|
|
func parseWWWAutolink(data string, position int) (Range, bool) {
|
|
// Check that this isn't part of another word
|
|
if position > 1 {
|
|
prevChar := data[position-1]
|
|
|
|
if !isWhitespaceByte(prevChar) && !isAllowedBeforeWWWLink(prevChar) {
|
|
return Range{}, false
|
|
}
|
|
}
|
|
|
|
// Check that this starts with www
|
|
if len(data)-position < 4 || !regexp.MustCompile(`^www\d{0,3}\.`).MatchString(data[position:]) {
|
|
return Range{}, false
|
|
}
|
|
|
|
end := checkDomain(data[position:], false)
|
|
if end == 0 {
|
|
return Range{}, false
|
|
}
|
|
|
|
end += position
|
|
|
|
// Grab all text until the end of the string or the next whitespace character
|
|
for end < len(data) && !isWhitespaceByte(data[end]) {
|
|
end += 1
|
|
}
|
|
|
|
// Trim trailing punctuation
|
|
end = trimTrailingCharactersFromLink(data, position, end)
|
|
if position == end {
|
|
return Range{}, false
|
|
}
|
|
|
|
return Range{position, end}, true
|
|
}
|
|
|
|
func isAllowedBeforeWWWLink(c byte) bool {
|
|
switch c {
|
|
case '*', '_', '~', ')':
|
|
return true
|
|
default:
|
|
return false
|
|
}
|
|
}
|
|
|
|
// Given a string with a : at the given position, tried to parse and return a range containing a URL scheme
|
|
// if one exists. If the text around the given position isn't a link, returns an empty string. Equivalent to
|
|
// url_match from the reference code.
|
|
func parseURLAutolink(data string, position int) (Range, bool) {
|
|
// Check that a :// exists. This doesn't match the clients that treat the slashes as optional.
|
|
if len(data)-position < 4 || data[position+1] != '/' || data[position+2] != '/' {
|
|
return Range{}, false
|
|
}
|
|
|
|
start := position - 1
|
|
for start > 0 && isAlphanumericByte(data[start-1]) {
|
|
start -= 1
|
|
}
|
|
|
|
// Ensure that the URL scheme is allowed and that at least one character after the scheme is valid.
|
|
scheme := data[start:position]
|
|
if !isSchemeAllowed(scheme) || !isValidHostCharacter(data[position+3:]) {
|
|
return Range{}, false
|
|
}
|
|
|
|
end := checkDomain(data[position+3:], true)
|
|
if end == 0 {
|
|
return Range{}, false
|
|
}
|
|
|
|
end += position
|
|
|
|
// Grab all text until the end of the string or the next whitespace character
|
|
for end < len(data) && !isWhitespaceByte(data[end]) {
|
|
end += 1
|
|
}
|
|
|
|
// Trim trailing punctuation
|
|
end = trimTrailingCharactersFromLink(data, start, end)
|
|
if start == end {
|
|
return Range{}, false
|
|
}
|
|
|
|
return Range{start, end}, true
|
|
}
|
|
|
|
func isSchemeAllowed(scheme string) bool {
|
|
// Note that this doesn't support the custom URL schemes implemented by the client
|
|
for _, allowed := range DefaultUrlSchemes {
|
|
if strings.EqualFold(allowed, scheme) {
|
|
return true
|
|
}
|
|
}
|
|
|
|
return false
|
|
}
|
|
|
|
// Given a string starting with a URL, returns the number of valid characters that make up the URL's domain.
|
|
// Returns 0 if the string doesn't start with a domain name. allowShort determines whether or not the domain
|
|
// needs to contain a period to be considered valid. Equivalent to check_domain from the reference code.
|
|
func checkDomain(data string, allowShort bool) int {
|
|
foundUnderscore := false
|
|
foundPeriod := false
|
|
|
|
i := 1
|
|
for ; i < len(data)-1; i++ {
|
|
if data[i] == '_' {
|
|
foundUnderscore = true
|
|
break
|
|
} else if data[i] == '.' {
|
|
foundPeriod = true
|
|
} else if !isValidHostCharacter(data[i:]) && data[i] != '-' {
|
|
break
|
|
}
|
|
}
|
|
|
|
if foundUnderscore {
|
|
return 0
|
|
}
|
|
|
|
if allowShort {
|
|
// If allowShort is set, accept any string of valid domain characters
|
|
return i
|
|
}
|
|
|
|
// If allowShort isn't set, a valid domain just requires at least a single period. Note that this
|
|
// logic isn't entirely necessary because we already know the string starts with "www." when
|
|
// this is called from parseWWWAutolink
|
|
if foundPeriod {
|
|
return i
|
|
} else {
|
|
return 0
|
|
}
|
|
}
|
|
|
|
// Returns true if the provided link starts with a valid character for a domain name. Equivalent to
|
|
// is_valid_hostchar from the reference code.
|
|
func isValidHostCharacter(link string) bool {
|
|
c, _ := utf8.DecodeRuneInString(link)
|
|
if c == utf8.RuneError {
|
|
return false
|
|
}
|
|
|
|
return !unicode.IsSpace(c) && !unicode.IsPunct(c)
|
|
}
|
|
|
|
// Removes any trailing characters such as punctuation or stray brackets that shouldn't be part of the link.
|
|
// Returns a new end position for the link. Equivalent to autolink_delim from the reference code.
|
|
func trimTrailingCharactersFromLink(markdown string, start int, end int) int {
|
|
runes := []rune(markdown[start:end])
|
|
linkEnd := len(runes)
|
|
|
|
// Cut off the link before an open angle bracket if it contains one
|
|
for i, c := range runes {
|
|
if c == '<' {
|
|
linkEnd = i
|
|
break
|
|
}
|
|
}
|
|
|
|
for linkEnd > 0 {
|
|
c := runes[linkEnd-1]
|
|
|
|
if !canEndAutolink(c) {
|
|
// Trim trailing quotes, periods, etc
|
|
linkEnd = linkEnd - 1
|
|
} else if c == ';' {
|
|
// Trim a trailing HTML entity
|
|
newEnd := linkEnd - 2
|
|
|
|
for newEnd > 0 && ((runes[newEnd] >= 'a' && runes[newEnd] <= 'z') || (runes[newEnd] >= 'A' && runes[newEnd] <= 'Z')) {
|
|
newEnd -= 1
|
|
}
|
|
|
|
if newEnd < linkEnd-2 && runes[newEnd] == '&' {
|
|
linkEnd = newEnd
|
|
} else {
|
|
// This isn't actually an HTML entity, so just trim the semicolon
|
|
linkEnd = linkEnd - 1
|
|
}
|
|
} else if c == ')' {
|
|
// Only allow an autolink ending with a bracket if that bracket is part of a matching pair of brackets.
|
|
// If there are more closing brackets than opening ones, remove the extra bracket
|
|
|
|
numClosing := 0
|
|
numOpening := 0
|
|
|
|
// Examples (input text => output linked portion):
|
|
//
|
|
// http://www.pokemon.com/Pikachu_(Electric)
|
|
// => http://www.pokemon.com/Pikachu_(Electric)
|
|
//
|
|
// http://www.pokemon.com/Pikachu_((Electric)
|
|
// => http://www.pokemon.com/Pikachu_((Electric)
|
|
//
|
|
// http://www.pokemon.com/Pikachu_(Electric))
|
|
// => http://www.pokemon.com/Pikachu_(Electric)
|
|
//
|
|
// http://www.pokemon.com/Pikachu_((Electric))
|
|
// => http://www.pokemon.com/Pikachu_((Electric))
|
|
|
|
for i := 0; i < linkEnd; i++ {
|
|
if runes[i] == '(' {
|
|
numOpening += 1
|
|
} else if runes[i] == ')' {
|
|
numClosing += 1
|
|
}
|
|
}
|
|
|
|
if numClosing <= numOpening {
|
|
// There's fewer or equal closing brackets, so we've found the end of the link
|
|
break
|
|
}
|
|
|
|
linkEnd -= 1
|
|
} else {
|
|
// There's no special characters at the end of the link, so we're at the end
|
|
break
|
|
}
|
|
}
|
|
|
|
return start + len(string(runes[:linkEnd]))
|
|
}
|
|
|
|
func canEndAutolink(c rune) bool {
|
|
switch c {
|
|
case '?', '!', '.', ',', ':', '*', '_', '~', '\'', '"':
|
|
return false
|
|
default:
|
|
return true
|
|
}
|
|
}
|