Forked from sketchytech/Common Swift String Extensions
Created
October 13, 2017 21:24
-
-
Save wuftymerguftyguff/b9c45c573d4699e91f7cb4ceb5e85576 to your computer and use it in GitHub Desktop.
Added and amended to use pure Swift where possible
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import Foundation | |
extension String | |
{ | |
// Works in Xcode but not Playgrounds because of a bug with .insert() | |
mutating func insertString(string:String,ind:Int) { | |
var insertIndex = advance(self.startIndex, ind, self.endIndex) | |
for c in string { | |
self.insert(c, atIndex: insertIndex) | |
insertIndex = advance(insertIndex, 1) | |
} | |
} | |
// new replace method using replaceRange (replaces all instances of string) | |
mutating func replace(string:String, replacement:String) { | |
let ranges = self.rangesOfString(string) | |
// if the string isn't found return unchanged string | |
for r in ranges { | |
self.replaceRange(r, with: replacement) | |
} | |
} | |
// Swift pure containsString | |
func containsString(findStr:String) -> Bool { | |
var arr = [Range<String.Index>]() | |
var startInd = self.startIndex | |
var i = 0 | |
// test first of all whether the string is likely to appear at all | |
if contains(self, first(findStr)!) { | |
startInd = find(self,first(findStr)!)! | |
} | |
else { | |
return false | |
} | |
// set starting point for search based on the finding of the first character | |
i = distance(self.startIndex, startInd) | |
while i<=countElements(self)-countElements(findStr) { | |
if self[advance(self.startIndex, i)..<advance(self.startIndex, i+countElements(findStr))] == findStr { | |
arr.append(Range(start:advance(self.startIndex, i),end:advance(self.startIndex, i+countElements(findStr)))) | |
return true | |
} | |
i++ | |
} | |
return false | |
} | |
// pure Swift - removeAtIndex no longer required, since the addition of following methods to String type | |
// mutating func removeAtIndex(i: String.Index) -> Character | |
// mutating func removeRange(subRange: Range<String.Index>) | |
// mutating func removeAll(keepCapacity: Bool = default) | |
// insert() method written in pure Swift, overloads the new String type method of the same name | |
func insert(string:String,ind:Int) -> String { | |
var insertIndex = advance(self.startIndex, ind, self.endIndex) | |
var returnString = toString(self) | |
for c in string { | |
returnString.insert(c, atIndex: insertIndex) | |
insertIndex = advance(insertIndex, 1) | |
} | |
return returnString | |
} | |
// rangesOfString: written in pure Swift (no Cocoa) | |
func rangesOfString(findStr:String) -> [Range<String.Index>] { | |
var arr = [Range<String.Index>]() | |
var startInd = self.startIndex | |
// check first that the first character of search string exists | |
if contains(self, first(findStr)!) { | |
// if so set this as the place to start searching | |
startInd = find(self,first(findStr)!)! | |
} | |
else { | |
// if not return empty array | |
return arr | |
} | |
var i = distance(self.startIndex, startInd) | |
while i<=countElements(self)-countElements(findStr) { | |
if self[advance(self.startIndex, i)..<advance(self.startIndex, i+countElements(findStr))] == findStr { | |
arr.append(Range(start:advance(self.startIndex, i),end:advance(self.startIndex, i+countElements(findStr)))) | |
i = i+countElements(findStr)-1 | |
// check again for first occurrence of character (this reduces number of times loop will run | |
if contains(self[advance(self.startIndex, i)..<self.endIndex], first(findStr)!) { | |
// if so set this as the place to start searching | |
i = distance(self.startIndex,find(self[advance(self.startIndex, i)..<self.endIndex],first(findStr)!)!) + i | |
countElements(findStr) | |
} | |
else { | |
return arr | |
} | |
} | |
i++ | |
} | |
return arr | |
} | |
func stringByReplacingOccurrencesOfString(string:String, replacement:String) -> String { | |
// get ranges first using rangesOfString: method, then glue together the string using ranges of existing string and old string | |
let ranges = self.rangesOfString(string) | |
// if the string isn't found return unchanged string | |
if ranges.isEmpty { | |
return self | |
} | |
// using toString to make a copy so that self isn't altered | |
var newString = toString(self) | |
for r in ranges { | |
newString.replaceRange(r, with: replacement) | |
} | |
return newString | |
} | |
// Added String splitting methods that return arrays (Pure Swift) | |
func splitStringByCharacters() -> [Character] { | |
return map(self){return $0} | |
} | |
func splitStringByLines() -> [String] { | |
return split(self, {contains("\u{2028}\n\r", $0) | |
}, allowEmptySlices: false) | |
} | |
func splitStringByWords() -> [String] { | |
return split(self, {contains(" .,!:;()[]{}<>?\"'\u{2028}\u{2029}\n\r", $0)}, allowEmptySlices: false) | |
} | |
func splitStringByParagraphs() -> [String] { | |
return split(self, {contains("\u{2029}\n\r", $0) | |
}, allowEmptySlices: false) | |
} | |
func splitStringBySentences() -> [String] { | |
let arr:[Character] = ["\u{2026}",".","?", "!"] | |
var startInd = self.startIndex | |
var strArr = [String]() | |
for b in enumerate(self) { | |
for a in arr { | |
if a == b.element { | |
var endInd = advance(self.startIndex,b.index,self.endIndex) | |
//TODO: add method to allow for multiple punctuation at end of sentence, e.g. ??? or !!! | |
var str = self[startInd...endInd] | |
// removes initial spaces and returns from sentence | |
if contains(" \u{2028}\u{2029}\n\r",first(str)!) { | |
str = dropFirst(str) | |
} | |
strArr.append(str) | |
startInd = advance(endInd,1,self.endIndex) | |
} | |
} | |
} | |
return strArr | |
} | |
func regexMatchesInString(regexString:String) -> [String] { | |
var arr = [String]() | |
var rang = startIndex..<endIndex | |
var foundRange:Range<String.Index>? | |
repeat | |
{ | |
foundRange = rangeOfString(regexString, options: NSStringCompareOptions.RegularExpressionSearch, range: rang, locale: nil) | |
if let fR = foundRange { | |
arr.append(substringWithRange(fR)) | |
rang.startIndex = fR.endIndex | |
} | |
} | |
while foundRange != nil | |
return arr | |
} | |
var length: Int { | |
get { | |
return countElements(self) | |
} | |
} | |
subscript (i: Int) -> Character | |
{ | |
get { | |
let index = advance(startIndex, i) | |
return self[index] | |
} | |
} | |
subscript (r: Range<Int>) -> String | |
{ | |
get { | |
let startIndex = advance(self.startIndex, r.startIndex) | |
let endIndex = advance(self.startIndex, r.endIndex - 1) | |
return self[Range(start: startIndex, end: endIndex)] | |
} | |
} | |
func subString(startIndex: Int, length: Int) -> String | |
{ | |
var start = advance(self.startIndex, startIndex) | |
var end = advance(self.startIndex, startIndex + length) | |
return self.substringWithRange(Range<String.Index>(start: start, end: end)) | |
} | |
func indexOf(target: String) -> Int | |
{ | |
var range = self.rangeOfString(target) | |
if let range = range { | |
return distance(self.startIndex, range.startIndex) | |
} else { | |
return -1 | |
} | |
} | |
func indexOf(target: String, startIndex: Int) -> Int | |
{ | |
var startRange = advance(self.startIndex, startIndex) | |
var range = self.rangeOfString(target, options: NSStringCompareOptions.LiteralSearch, range: Range<String.Index>(start: startRange, end: self.endIndex)) | |
if let range = range { | |
return distance(self.startIndex, range.startIndex) | |
} else { | |
return -1 | |
} | |
} | |
func lastIndexOf(target: String) -> Int | |
{ | |
var index = -1 | |
var stepIndex = self.indexOf(target) | |
while stepIndex > -1 | |
{ | |
index = stepIndex | |
if stepIndex + target.length < self.length { | |
stepIndex = indexOf(target, startIndex: stepIndex + target.length) | |
} else { | |
stepIndex = -1 | |
} | |
} | |
return index | |
} | |
// Updated isMatch to remove reliance on NSRegularExpression | |
func isMatch(regex: String, options: NSStringCompareOptions?) -> Bool | |
{ | |
let match = self.rangeOfString(regex, options: options ?? nil, range: Range(start: self.startIndex, end: self.endIndex), locale: nil) | |
return match != nil ? true : false | |
} | |
// getMatches updated to remove reliance on NSRegularExpression | |
func getMatches(str: String, options: NSStringCompareOptions?) -> [Range<String.Index>] { | |
var arr = [Range<String.Index>]() | |
var rang = startIndex..<endIndex | |
var foundRange:Range<String.Index>? | |
repeat | |
{ | |
foundRange = rangeOfString(str, options: options ?? [], range: rang, locale: nil) | |
if let fR = foundRange { | |
arr.append(fR) | |
rang.startIndex = fR.endIndex | |
} | |
} | |
while foundRange != nil | |
return arr | |
} | |
private var vowels: [String] | |
{ | |
return ["a", "e", "i", "o", "u"] | |
} | |
private var consonants: [String] | |
{ | |
return ["b", "c", "d", "f", "g", "h", "j", "k", "l", "m", "n", "p", "q", "r", "s", "t", "v", "w", "x", "z"] | |
} | |
func pluralize(count: Int) -> String | |
{ | |
if count == 1 { | |
return self | |
} else { | |
var lastChar = self.subString(self.length - 1, length: 1) | |
var secondToLastChar = self.subString(self.length - 2, length: 1) | |
var prefix = "", suffix = "" | |
if lastChar.lowercaseString == "y" && vowels.filter({x in x == secondToLastChar}).count == 0 { | |
prefix = self[0...self.length - 1] | |
suffix = "ies" | |
} else if lastChar.lowercaseString == "s" || (lastChar.lowercaseString == "o" && consonants.filter({x in x == secondToLastChar}).count > 0) { | |
prefix = self[0...self.length] | |
suffix = "es" | |
} else { | |
prefix = self[0...self.length] | |
suffix = "s" | |
} | |
return prefix + (lastChar != lastChar.uppercaseString ? suffix : suffix.uppercaseString) | |
} | |
} | |
// for fun, not sure whether this has a practical application | |
func removeDuplicates(array:String) -> String { | |
var arr = array | |
var indArr = [Int]() | |
var tempArr = arr | |
var i = 0 | |
for a in arr { | |
if contains(prefix(arr, i), a) { | |
indArr.append(i) | |
} | |
i++ | |
} | |
var ind = 0 | |
for i in indArr { | |
arr.removeAtIndex(i-ind) | |
ind++ | |
} | |
return arr | |
} | |
} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
// sketchyTech added and updated begins | |
"Hello".regexMatchesInString("[^Hh]{1,}") | |
"Hello".isMatch("[x-z]{1,}", options: NSStringCompareOptions.RegularExpressionSearch) | |
"Hello".getMatches("[x-z]{1,}", options: NSStringCompareOptions.RegularExpressionSearch) | |
// pure Swift methods | |
"Hello".splitStringByCharacters() | |
"Hello. This is a new method.".splitStringByWords() | |
"Hello. This is a new method.".splitStringBySentences() | |
"Hello. This is a new method.\n I hope you enjoy using it.".splitStringByParagraphs() | |
"a very good hello, hello".rangesOfString("hello") | |
"a very good hello, hello".stringByReplacingOccurrencesOfString("hello",replacement:"bye") | |
"Hello World!".insert("Swift ", ind: 6) | |
// update replace method to use new replaceRange method | |
"a very good hello, hello".replace("hello",replacement:"bye") | |
// for fun (also pure Swift) | |
let str = "hello world" | |
str.removeDuplicates() | |
// insertString() | |
var a = "hello world" | |
a.insertString("big ", ind:6) | |
REMOVED: removeAtIndex | |
// pure Swift - removeAtIndex no longer required, since the addition of following methods to String type | |
// mutating func removeAtIndex(i: String.Index) -> Character | |
// mutating func removeRange(subRange: Range<String.Index>) | |
// mutating func removeAll(keepCapacity: Bool = default) | |
REMOVED: contains() and replaced with containsString() | |
str.containsString("play") | |
// sketchyTech added and updated ends | |
"ReplaceMe".replace("Me", withString: "You") == "ReplaceYou" | |
"MeReplace".replace("Me", withString: "You") == "YouReplace" | |
"ReplaceMeNow".replace("Me", withString: "You") == "ReplaceYouNow" | |
"0123456789"[0] == "0" | |
"0123456789"[5] == "5" | |
"0123456789"[9] == "9" | |
"0123456789"[5...6] == "5" | |
"0123456789"[0...1] == "0" | |
"0123456789"[8...9] == "8" | |
"0123456789"[1...5] == "1234" | |
"Reply"[0...4] == "Repl" | |
"Hello, playground"[0...5] == "Hello" | |
"Coolness"[4...7] == "nes" | |
"Awesome".indexOf("nothin") == -1 | |
"Awesome".indexOf("Awe") == 0 | |
"Awesome".indexOf("some") == 3 | |
"Awesome".indexOf("e", startIndex: 3) == 6 | |
"Awesome".lastIndexOf("e") == 6 | |
"Cool".lastIndexOf("o") == 2 | |
var emailRegex = "[a-z_\\-\\.]+@[a-z_\\-\\.]{3,}" | |
"[email protected]".isMatch(emailRegex, options: NSStringCompareOptions.CaseInsensitiveSearch) == true | |
"email-test.com".isMatch(emailRegex, options: NSStringCompareOptions.InsensitiveSearch) == false | |
var testText = "[email protected], [email protected], [email protected]" | |
var matches = testText.getMatches(emailRegex, options: NSStringCompareOptions.InsensitiveSearch) | |
matches.count == 3 | |
testText.subString(matches[0].range.location, length: matches[0].range.length) == "[email protected]" | |
testText.subString(matches[1].range.location, length: matches[1].range.length) == "[email protected]" | |
testText.subString(matches[2].range.location, length: matches[2].range.length) == "[email protected]" | |
"Reply".pluralize(0) == "Replies" | |
"Reply".pluralize(1) == "Reply" | |
"Reply".pluralize(2) == "Replies" | |
"REPLY".pluralize(3) == "REPLIES" | |
"Horse".pluralize(2) == "Horses" | |
"Boy".pluralize(2) == "Boys" | |
"Cut".pluralize(2) == "Cuts" | |
"Boss".pluralize(2) == "Bosses" | |
"Domino".pluralize(2) == "Dominoes" |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment