How to split a comma separated String while ignoring escaped commas?

The regular expression

[^\\],

means "match a character which is not a backslash followed by a comma" - this is why patterns such as t, are matching, because t is a character which is not a backslash.

I think you need to use some sort of negative lookbehind, to capture a , which is not preceded by a \ without capturing the preceding character, something like

(?<!\\),

(BTW, note that I have purposefully not doubly-escaped the backslashes to make this more readable)


Try:

String array[] = str.split("(?<!\\\\),");

Basically this is saying split on a comma, except where that comma is preceded by two backslashes. This is called a negative lookbehind zero-width assertion.


For future reference, here is the complete method i ended up with:

public static String[] commaDelimitedListToStringArray(String str, String escapeChar) {
    // these characters need to be escaped in a regular expression
    String regularExpressionSpecialChars = "/.*+?|()[]{}\\";

    String escapedEscapeChar = escapeChar;

    // if the escape char for our comma separated list needs to be escaped 
    // for the regular expression, escape it using the \ char
    if(regularExpressionSpecialChars.indexOf(escapeChar) != -1) 
        escapedEscapeChar = "\\" + escapeChar;

    // see http://stackoverflow.com/questions/820172/how-to-split-a-comma-separated-string-while-ignoring-escaped-commas
    String[] temp = str.split("(?<!" + escapedEscapeChar + "),", -1);

    // remove the escapeChar for the end result
    String[] result = new String[temp.length];
    for(int i=0; i<temp.length; i++) {
        result[i] = temp[i].replaceAll(escapedEscapeChar + ",", ",");
    }

    return result;
}

As matt b said, [^\\], will interpret the character preceding the comma as a part of the delimiter.

"test\\\\\\,test\\\\,test\\,test,test"
  -(split)->
["test\\\\\\,test\\\\,test\\,tes" , "test"]

As drvdijk said, (?<!\\), will misinterpret escaped backslashes.

"test\\\\\\,test\\\\,test\\,test,test"
  -(split)->
["test\\\\\\,test\\\\,test\\,test" , "test"]
  -(unescape commas)->
["test\\\\,test\\,test,test" , "test"]

I would expect being able to escape backslashes as well...

"test\\\\\\,test\\\\,test\\,test,test"
  -(split)->
["test\\\\\\,test\\\\" , "test\\,test" , "test"]
  -(unescape commas and backslashes)->
["test\\,test\\" , "test,test" , "test"]

drvdijk suggested (?<=(?<!\\\\)(\\\\\\\\){0,100}), which works well for lists with elements ending with up to 100 backslashes. This is far enough... but why a limit? Is there a more efficient way (isn't lookbehind greedy)? What about invalid strings?

I searched for a while for a generic solution, then I wrote the thing myself... The idea is to split following a pattern that matches the list elements (instead of matching the delimiter).

My answer does not take the escape character as a parameter.

public static List<String> commaDelimitedListStringToStringList(String list) {
    // Check the validity of the list
    // ex: "te\\st" is not valid, backslash should be escaped
    if (!list.matches("^(([^\\\\,]|\\\\,|\\\\\\\\)*(,|$))+")) {
        // Could also raise an exception
        return null;
    }
    // Matcher for the list elements
    Matcher matcher = Pattern
            .compile("(?<=(^|,))([^\\\\,]|\\\\,|\\\\\\\\)*(?=(,|$))")
            .matcher(list);
    ArrayList<String> result = new ArrayList<String>();
    while (matcher.find()) {
        // Unescape the list element
        result.add(matcher.group().replaceAll("\\\\([\\\\,])", "$1"));
    }
    return result;
}

Description for the pattern (unescaped):

(?<=(^|,)) forward is start of string or a ,

([^\\,]|\\,|\\\\)* the element composed of \,, \\ or characters wich are neither \ nor ,

(?=(,|$)) behind is end of string or a ,

The pattern may be simplified.

Even with the 3 parsings (matches + find + replaceAll), this method seems faster than the one suggested by drvdijk. It can still be optimized by writing a specific parser.

Also, what is the need of having an escape character if only one character is special, it could simply be doubled...

public static List<String> commaDelimitedListStringToStringList2(String list) {
    if (!list.matches("^(([^,]|,,)*(,|$))+")) {
        return null;
    }
    Matcher matcher = Pattern.compile("(?<=(^|,))([^,]|,,)*(?=(,|$))")
                    .matcher(list);
    ArrayList<String> result = new ArrayList<String>();
    while (matcher.find()) {
        result.add(matcher.group().replaceAll(",,", ","));
    }
    return result;
}