Backtracking

Backtracking occurs when a regular expression pattern contains optional quantifiers or alternation constructs, and the regular expression engine returns to a previous saved state to continue its search for a match. Backtracking is central to the power of regular expressions; it makes it possible for expressions to be powerful and flexible, and to match very complex patterns. At the same time, this power comes at a cost. Backtracking is often the single most important factor that affects the performance of the regular expression engine. Fortunately, the developer has control over the behavior of the regular expression engine and how it uses backtracking. This topic explains how backtracking works and how it can be controlled.

This topic contains the following sections:

• Linear Comparison Without Backtracking

• Backtracking with Optional Quantifiers or Alternation Constructs

• Backtracking with Nested Optional Quantifiers

• Controlling Backtracking

Linear Comparison Without Backtracking

If a regular expression pattern has no optional quantifiers or alternation constructs, the regular expression engine executes in linear time. That is, after the regular expression engine matches the first language element in the pattern with text in the input string, it tries to match the next language element in the pattern with the next character or group of characters in the input string. This continues until the match either succeeds or fails. In either case, the regular expression engine advances by one character at a time in the input string.

The following example provides an illustration. The regular expression e{2}\w\b looks for two occurrences of the letter "e" followed by any word character followed by a word boundary.

Imports System.Text.RegularExpressions

Module Example
   Public Sub Main()
      Dim input As String = "needing a reed" 
      Dim pattern As String = "e{2}\w\b" 
      For Each match As Match In Regex.Matches(input, pattern)
         Console.WriteLine("{0} found at position {1}", _
                           match.Value, match.Index)
      Next    
   End Sub 
End Module 
' The example displays the following output: 
'       eed found at position 11

using System;
using System.Text.RegularExpressions;

public class Example
{
   public static void Main()
   {
      string input = "needing a reed";
      string pattern = @"e{2}\w\b";
      foreach (Match match in Regex.Matches(input, pattern))
         Console.WriteLine("{0} found at position {1}", 
                           match.Value, match.Index);
   }
}
// The example displays the following output: 
//       eed found at position 11

Although this regular expression includes the quantifier {2}, it is evaluated in a linear manner. The regular expression engine does not backtrack because {2} is not an optional quantifier; it specifies an exact number and not a variable number of times that the previous subexpression must match. As a result, the regular expression engine tries to match the regular expression pattern with the input string as shown in the following table.

If a regular expression pattern includes no optional quantifiers or alternation constructs, the maximum number of comparisons required to match the regular expression pattern with the input string is roughly equivalent to the number of characters in the input string. In other words, the regular expression engine runs in linear time if it contains no optional quantifiers or alternation constructs.

Back to top

Backtracking with Optional Quantifiers or Alternation Constructs

When a regular expression includes optional quantifiers or alternation constructs, the evaluation of the input string is no longer linear. Pattern matching with an NFA engine is driven by the language elements in the regular expression and not by the characters to be matched in the input string. Therefore, the regular expression engine tries to fully match optional or alternative subexpressions. When it advances to the next language element in the subexpression and the match is unsuccessful, the regular expression engine can abandon a portion of its successful match and return to an earlier saved state in the interest of matching the regular expression as a whole with the input string. This process of returning to a previous saved state to find a match is known as backtracking.

For example, consider the regular expression pattern .*(es), which matches the characters "es" and all the characters that precede it. As the following example shows, if the input string is "Essential services are provided by regular expressions.", the pattern matches the whole string up to and including the "es" in "expressions".

Imports System.Text.RegularExpressions

Module Example
   Public Sub Main()
      Dim input As String = "Essential services are provided by regular expressions." 
      Dim pattern As String = ".*(es)"  
      Dim m As Match = Regex.Match(input, pattern, RegexOptions.IgnoreCase)
      If m.Success Then
         Console.WriteLine("'{0}' found at position {1}", _
                           m.Value, m.Index)
         Console.WriteLine("'es' found at position {0}", _
                           m.Groups(1).Index)                  
      End If      
   End Sub 
End Module 
'    'Essential services are provided by regular expres' found at position 0 
'    'es' found at position 47

using System;
using System.Text.RegularExpressions;

public class Example
{
   public static void Main()
   {
      string input = "Essential services are provided by regular expressions.";
      string pattern = ".*(es)"; 
      Match m = Regex.Match(input, pattern, RegexOptions.IgnoreCase);
      if (m.Success) {
         Console.WriteLine("'{0}' found at position {1}", 
                           m.Value, m.Index);
         Console.WriteLine("'es' found at position {0}", 
                           m.Groups[1].Index);      
      } 
   }
}
//    'Essential services are provided by regular expres' found at position 0 
//    'es' found at position 47

To do this, the regular expression engine uses backtracking as follows:

• It matches the .* (which matches zero, one, or more occurrences of any character) with the whole input string.

• It attempts to match "e" in the regular expression pattern. However, the input string has no remaining characters available to match.

• It backtracks to its last successful match, "Essential services are provided by regular expressions", and attempts to match "e" with the period at the end of the sentence. The match fails.

• It continues to backtrack to a previous successful match one character at a time until the tentatively matched substring is "Essential services are provided by regular expr". It then compares the "e" in the pattern to the second "e" in "expressions" and finds a match.

• It compares "s" in the pattern to the "s" that follows the matched "e" character (the first "s" in "expressions"). The match is successful.

When you use backtracking, matching the regular expression pattern with the input string, which is 55 characters long, requires 67 comparison operations. Interestingly, if the regular expression pattern included a lazy quantifier, .*?(es), matching the regular expression would require additional comparisons. In this case, instead of having to backtrack from the end of the string to the "r" in "expressions", the regular expression engine would have to backtrack all the way to the beginning of the string to match "Es" and would require 113 comparisons. Generally, if a regular expression pattern has a single alternation construct or a single optional quantifier, the number of comparison operations required to match the pattern is more than twice the number of characters in the input string.

Back to top

Backtracking with Nested Optional Quantifiers

The number of comparison operations required to match a regular expression pattern can increase exponentially if the pattern includes a large number of alternation constructs, if it includes nested alternation constructs, or, most commonly, if it includes nested optional quantifiers. For example, the regular expression pattern ^(a)+$ is designed to match a complete string that contains one or more "a" characters. The example provides two input strings of identical length, but only the first string matches the pattern. The System.Diagnostics.Stopwatch class is used to determine how long the match operation takes.

Imports System.Diagnostics
Imports System.Text.RegularExpressions

Module Example
   Public Sub Main()
      Dim pattern As String = "^(a)+$" 
      Dim inputs() As String = { "aaaaaa", "aaaaa!" }
      Dim rgx As New Regex(pattern)
      Dim sw As Stopwatch

      For Each input As String In inputs
         sw = Stopwatch.StartNew()   
         Dim match As Match = rgx.Match(input)
         sw.Stop()
         If match.Success Then
            Console.WriteLine("Matched {0} in {1}", match.Value, sw.Elapsed)
         Else
            Console.WriteLine("No match found in {0}", sw.Elapsed)
         End If       
      Next 
   End Sub 
End Module

using System;
using System.Diagnostics;
using System.Text.RegularExpressions;

public class Example
{
   public static void Main()
   {
      string pattern = "^(a)+$";
      string[] inputs = { "aaaaaa", "aaaaa!" };
      Regex rgx = new Regex(pattern);
      Stopwatch sw;

      foreach (string input in inputs) {
         sw = Stopwatch.StartNew();   
         Match match = rgx.Match(input);
         sw.Stop();
         if (match.Success)
            Console.WriteLine("Matched {0} in {1}", match.Value, sw.Elapsed);
         else
            Console.WriteLine("No match found in {0}", sw.Elapsed);
      }
   }
}

As the output from the example shows, the regular expression engine took about twice as long to find that an input string did not match the pattern as it did to identify a matching string. This is because an unsuccessful match always represents a worst-case scenario. The regular expression engine must use the regular expression to follow all possible paths through the data before it can conclude that the match is unsuccessful, and the nested parentheses create many additional paths through the data. The regular expression engine concludes that the second string did not match the pattern by doing the following:

• It checks that it was at the beginning of the string, and then matches the first five characters in the string with the pattern a+. It then determines that there are no additional groups of "a" characters in the string. Finally, it tests for the end of the string. Because one additional character remains in the string, the match fails. This failed match requires 9 comparisons. The regular expression engine also saves state information from its matches of "a" (which we will call match 1), "aa" (match 2), "aaa" (match 3), and "aaaa" (match 4).

• It returns to the previously saved match 4. It determines that there is one additional "a" character to assign to an additional captured group. Finally, it tests for the end of the string. Because one additional character remains in the string, the match fails. This failed match requires 4 comparisons. So far, a total of 13 comparisons have been performed.

• It returns to the previously saved match 3. It determines that there are two additional "a" characters to assign to an additional captured group. However, the end-of-string test fails. It then returns to match3 and tries to match the two additional "a" characters in two additional captured groups. The end-of-string test still fails. These failed matches require 12 comparisons. So far, a total of 25 comparisons have been performed.

Comparison of the input string with the regular expression continues in this way until the regular expression engine has tried all possible combinations of matches, and then concludes that there is no match. Because of the nested quantifiers, this comparison is an O(2n) or an exponential operation, where n is the number of characters in the input string. This means that in the worst case, an input string of 30 characters requires approximately 1,073,741,824 comparisons, and an input string of 40 characters requires approximately 1,099,511,627,776 comparisons. If you use strings of these or even greater lengths, regular expression methods can take an extremely long time to complete when they process input that does not match the regular expression pattern.

Back to top

Controlling Backtracking

Backtracking lets you create powerful, flexible regular expressions. However, as the previous section showed, these benefits may be coupled with unacceptably poor performance. The .NET Framework supports three regular expression language elements that limit or suppress backtracking and that support complex regular expressions with little or no performance penalty: nonbacktracking subexpressions, lookbehind assertions, and lookahead assertions. For more information about each language element, see Grouping Constructs.

Nonbacktracking Subexpression

The (?> subexpression) language element suppresses backtracking in a subexpression. It is useful for preventing the performance problems associated with failed matches.

The following example illustrates how suppressing backtracking improves performance when using nested quantifiers. It measures the time required for the regular expression engine to determine that an input string does not match two regular expressions. The first regular expression uses backtracking to attempt to match a string that contains one or more occurrences of one or more hexadecimal digits, followed by a colon, followed by one or more hexadecimal digits, followed by two colons. The second regular expression is identical to the first, except that it disables backtracking. As the output from the example shows, the performance improvement from disabling backtracking is significant.

Imports System.Diagnostics
Imports System.Text.RegularExpressions

Module Example
   Public Sub Main()
      Dim input As String = "b51:4:1DB:9EE1:5:27d60:f44:D4:cd:E:5:0A5:4a:D24:41Ad:" 
      Dim matched As Boolean 
      Dim sw As Stopwatch

      Console.WriteLine("With backtracking:")
      Dim backPattern As String = "^(([0-9a-fA-F]{1,4}:)*([0-9a-fA-F]{1,4}))*(::)$"
      sw = Stopwatch.StartNew()
      matched = Regex.IsMatch(input, backPattern)
      sw.Stop()
      Console.WriteLine("Match: {0} in {1}", Regex.IsMatch(input, backPattern), sw.Elapsed)
      Console.WriteLine()

      Console.WriteLine("Without backtracking:")
      Dim noBackPattern As String = "^((?>[0-9a-fA-F]{1,4}:)*(?>[0-9a-fA-F]{1,4}))*(::)$"
      sw = Stopwatch.StartNew()
      matched = Regex.IsMatch(input, noBackPattern)
      sw.Stop()
      Console.WriteLine("Match: {0} in {1}", Regex.IsMatch(input, noBackPattern), sw.Elapsed)
   End Sub 
End Module 
' The example displays the following output: 
'       With backtracking: 
'       Match: False in 00:00:27.4282019 
'        
'       Without backtracking: 
'       Match: False in 00:00:00.0001391

using System;
using System.Diagnostics;
using System.Text.RegularExpressions;

public class Example
{
   public static void Main()
   {
      string input = "b51:4:1DB:9EE1:5:27d60:f44:D4:cd:E:5:0A5:4a:D24:41Ad:";
      bool matched;
      Stopwatch sw;

      Console.WriteLine("With backtracking:");
      string backPattern = "^(([0-9a-fA-F]{1,4}:)*([0-9a-fA-F]{1,4}))*(::)$";
      sw = Stopwatch.StartNew();
      matched = Regex.IsMatch(input, backPattern);
      sw.Stop();
      Console.WriteLine("Match: {0} in {1}", Regex.IsMatch(input, backPattern), sw.Elapsed);
      Console.WriteLine();

      Console.WriteLine("Without backtracking:");
      string noBackPattern = "^((?>[0-9a-fA-F]{1,4}:)*(?>[0-9a-fA-F]{1,4}))*(::)$";
      sw = Stopwatch.StartNew();
      matched = Regex.IsMatch(input, noBackPattern);
      sw.Stop();
      Console.WriteLine("Match: {0} in {1}", Regex.IsMatch(input, noBackPattern), sw.Elapsed);
   }
}
// The example displays output like the following: 
//       With backtracking: 
//       Match: False in 00:00:27.4282019 
//        
//       Without backtracking: 
//       Match: False in 00:00:00.0001391

Lookbehind Assertions

The .NET Framework includes two language elements, (?<=subexpression) and (?<!subexpression), that match the previous character or characters in the input string. Both language elements are zero-width assertions; that is, they determine whether the character or characters that immediately precede the current character can be matched by subexpression, without advancing or backtracking.

(?<=subexpression) is a positive lookbehind assertion; that is, the character or characters before the current position must match subexpression. (?<!subexpression) is a negative lookbehind assertion; that is, the character or characters before the current position must not match subexpression. Both positive and negative lookbehind assertions are most useful when subexpression is a subset of the previous subexpression.

The following example uses two equivalent regular expression patterns that validate the user name in an e-mail address. The first pattern is subject to poor performance because of excessive backtracking. The second pattern modifies the first regular expression by replacing a nested quantifier with a positive lookbehind assertion. The output from the example displays the execution time of the Regex.IsMatch method.

Module Example
   Public Sub Main()
      Dim sw As Stopwatch
      Dim input As String = "aaaaaaaaaaaaaaaaaaaa" 
      Dim result As Boolean 

      Dim pattern As String = "^[0-9A-Z]([-.\w]*[0-9A-Z])?@"
      sw = Stopwatch.StartNew()
      result = Regex.IsMatch(input, pattern, RegexOptions.IgnoreCase)
      sw.Stop()
      Console.WriteLine("Match: {0} in {1}", result, sw.Elapsed)

      Dim behindPattern As String = "^[0-9A-Z][-.\w]*(?<=[0-9A-Z])@"
      sw = Stopwatch.StartNew()
      result = Regex.IsMatch(input, behindPattern, RegexOptions.IgnoreCase)
      sw.Stop()
      Console.WriteLine("Match with Lookbehind: {0} in {1}", result, sw.Elapsed)
   End Sub 
End Module 
' The example displays output similar to the following: 
'       Match: True in 00:00:00.0017549 
'       Match with Lookbehind: True in 00:00:00.0000659

using System;
using System.Diagnostics;
using System.Text.RegularExpressions;

public class Example
{
   public static void Main()
   {
      Stopwatch sw;
      string input = "aaaaaaaaaaaaaaaaaaaa";
      bool result;

      string pattern = @"^[0-9A-Z]([-.\w]*[0-9A-Z])?@";
      sw = Stopwatch.StartNew();
      result = Regex.IsMatch(input, pattern, RegexOptions.IgnoreCase);
      sw.Stop();
      Console.WriteLine("Match: {0} in {1}", result, sw.Elapsed);

      string behindPattern = @"^[0-9A-Z][-.\w]*(?<=[0-9A-Z])@";
      sw = Stopwatch.StartNew();
      result = Regex.IsMatch(input, behindPattern, RegexOptions.IgnoreCase);
      sw.Stop();
      Console.WriteLine("Match with Lookbehind: {0} in {1}", result, sw.Elapsed);
   }
}
// The example displays output similar to the following: 
//       Match: True in 00:00:00.0017549 
//       Match with Lookbehind: True in 00:00:00.0000659

The first regular expression pattern, ^[0-9A-Z]([-.\w]*[0-9A-Z])*@, is defined as shown in the following table.

The second regular expression pattern, ^[0-9A-Z][-.\w]*(?<=[0-9A-Z])@, uses a positive lookbehind assertion. It is defined as shown in the following table.

Lookahead Assertions

The .NET Framework includes two language elements, (?=subexpression) and (?!subexpression), that match the next character or characters in the input string. Both language elements are zero-width assertions; that is, they determine whether the character or characters that immediately follow the current character can be matched by subexpression, without advancing or backtracking.

(?=subexpression) is a positive lookahead assertion; that is, the character or characters after the current position must match subexpression. (?!subexpression) is a negative lookahead assertion; that is, the character or characters after the current position must not match subexpression. Both positive and negative lookahead assertions are most useful when subexpression is a subset of the next subexpression.

The following example uses two equivalent regular expression patterns that validate a fully qualified type name. The first pattern is subject to poor performance because of excessive backtracking. The second modifies the first regular expression by replacing a nested quantifier with a positive lookahead assertion. The output from the example displays the execution time of the Regex.IsMatch method.

Imports System.Diagnostics
Imports System.Text.RegularExpressions

Module Example
   Public Sub Main()
      Dim input As String = "aaaaaaaaaaaaaaaaaaaaaa." 
      Dim result As Boolean 
      Dim sw As Stopwatch

      Dim pattern As String = "^(([A-Z]\w*)+\.)*[A-Z]\w*$"
      sw = Stopwatch.StartNew()
      result = Regex.IsMatch(input, pattern, RegexOptions.IgnoreCase)
      sw.Stop()
      Console.WriteLine("{0} in {1}", result, sw.Elapsed)

      Dim aheadPattern As String = "^((?=[A-Z])\w+\.)*[A-Z]\w*$"
      sw = Stopwatch.StartNew()
      result = Regex.IsMatch(input, aheadPattern, RegexOptions.IgnoreCase)
      sw.Stop()
      Console.WriteLine("{0} in {1}", result, sw.Elapsed)
   End Sub 
End Module 
' The example displays the following output: 
'       False in 00:00:03.8003793 
'       False in 00:00:00.0000866

using System;
using System.Diagnostics;
using System.Text.RegularExpressions;

public class Example
{
   public static void Main()
   {
      string input = "aaaaaaaaaaaaaaaaaaaaaa.";
      bool result;
      Stopwatch sw;

      string pattern = @"^(([A-Z]\w*)+\.)*[A-Z]\w*$";
      sw = Stopwatch.StartNew();
      result = Regex.IsMatch(input, pattern, RegexOptions.IgnoreCase);
      sw.Stop();
      Console.WriteLine("{0} in {1}", result, sw.Elapsed);

      string aheadPattern = @"^((?=[A-Z])\w+\.)*[A-Z]\w*$";
      sw = Stopwatch.StartNew();
      result = Regex.IsMatch(input, aheadPattern, RegexOptions.IgnoreCase);
      sw.Stop();
      Console.WriteLine("{0} in {1}", result, sw.Elapsed);
   }
}
// The example displays the following output: 
//       False in 00:00:03.8003793 
//       False in 00:00:00.0000866

The first regular expression pattern, ^(([A-Z]\w*)+\.)*[A-Z]\w*$, is defined as shown in the following table.

The second regular expression pattern, ^((?=[A-Z])\w+\.)*[A-Z]\w*$, uses a positive lookahead assertion. It is defined as shown in the following table.

Back to top

See Also

Concepts

Regular Expression Language - Quick Reference

Reference

Quantifiers

Alternation Constructs

Grouping Constructs

Other Resources

.NET Framework Regular Expressions

Change History