Regular expressions in C# can save you when you need to do some complex pattern matching on strings. But as the language grows and evolves we continue to get new ways that we can use regular expressions in C#. That’s why I wanted to take a moment to consider C# Regex performance by running benchmarks across the various ways I’m familiar with using regular expressions.
When we have options that look the same, I like remaining curious and understanding what’s actually different. This turned out to be a fun exercise and I hope you have some learnings to take away just like I did!
PLEASE NOTE:
There was an error in the benchmarks that I created because of an assumption that was invalid. The contents of this article can still be very valuable, but there is an update to the benchmark code and an accurate analysis in this follow-up article:
https://www.devleader.ca/2024/04/12/csharp-regular-expression-benchmarks-how-to-avoid-my-mistakes/
What is a Regular Expression?
Regular expressions, often referred to as regex, are powerful tools used for pattern matching in text. They allow you to define a search pattern that can be used to find, replace, or manipulate specific parts of a string. Regular expressions provide a concise and flexible way to search for and identify specific patterns within text data.
I have used regular expressions for many different things in my career such as:
Pattern matching on user input
Scraping data from the web
Parsing data sources like logs or other files
Digital forensics data recovery
… and many more uses!
Regular expressions can be used for all sorts of advanced pattern matching. But if you want to get started you can check out these articles:
Aside from when I am trying to profile and optimize my applications and services, I like benchmarking things when I am curious. This happens especially when I find there are seemingly multiple ways to do the same thing — it makes me wonder TRULY what the differences are aside form just syntax and usability. Recently with collection initializers, I was completely blown away by the performance differences, so it’s always a good reminder to be curious.
Regular expressions in C# have several different flavors:
Static method call
Compiled flag or not
Source generators
Now, the compiled flag is supposed to give us a performance boost but what’s the overhead of using just the static method call since it makes things really convenient to call in code? And what the heck are these (relatively) new source generators for regular expressions in C#? Microsoft had some REALLY awesome documentation on compiled regular expressions and source generation — and I never would have stumbled upon this if I wasn’t curious to benchmark.
While we won’t be comparing all of the regex methods we have access to, I did want to see the performance difference for getting all matches in a body of text. Considering the scenarios listed above, I was curious to see which would pull forward given they all seem roughly the same on the surface.
I know you’re eager to jump STRAIGHT to the details, and while nothing is stopping you from scrolling to the bottom I’m hopeful you’ll pause in this section to understand the benchmarks first. When I post benchmarking investigations, my ultimate goal is not to persuade you to code differently, but rather to be curious about what you’re coding. I think we’ll see some obvious things to avoid in these benchmarks — but still, being curious is the goal.
I’ll be working with BenchmarkDotNet as usual for these benchmarks. If you’re interested in learning more about how to run your own benchmarks I have plenty of BenchmarkDotNet articles you can check out.
The Test Data For Benchmarking
While we’re not interested in absolute performance here — and you might be if you’re profiling and benchmarking your own application — we are interested in relative performance between our various Regex options. Aside from the options we have to use, some other considerations:
I call these things out because they are uncertain to me. The source data I feel shouldn’t be TOO big of an issue — but maybe different heuristics of matching don’t allow compiled or source-generated regular expressions to shine. Maybe the Regex pattern I’ve selected doesn’t allow compiled or source-generated regular expressions to have an advantage. Or maybe it does — and I should be able to call this out.
The point is that there are some variables that may influence results and I don’t fully understand how. But this is me being transparent, and if you know better feel free to share your insights!
I figured to make this as “fair” as possible, I would look for some patterns in real text: words that end in “ing” or in “ed”. To find some real text, I am using the data from Project Gutenberg — Specifically, this E-Book. It’s 2200+ lines of English text so there are plenty of matches on words for our pattern.
The C# Regex Benchmark Code
There’s nothing super unique about these benchmarks compared to others I normally create. However, I’ll list out some points of interest so that you can pick them out in the code:
I’m using [Params] to load the source file in case you want to try these benchmarks across different datasets
The source data is read in during the global setup
I need to cache some Regex instances for some of the benchmarks, which is done in the global setup
You can find the benchmark code here on GitHub and in the code below:
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;
using System.Reflection;
using System.Text.RegularExpressions;
BenchmarkRunner.Run(
Assembly.GetExecutingAssembly(),
args: args);
[MemoryDiagnoser]
[MediumRunJob]
public partial class RegexBenchmarks
{
private const string RegexPattern = @"\b\w*(ing|ed)\b";
private string? _sourceText;
private Regex? _regex;
private Regex? _regexCompiled;
private Regex? _generatedRegex;
private Regex? _generatedRegexCompiled;
[GeneratedRegex(RegexPattern, RegexOptions.None, "en-US")]
private static partial Regex GetGeneratedRegex();
[GeneratedRegex(RegexPattern, RegexOptions.Compiled, "en-US")]
private static partial Regex GetGeneratedRegexCompiled();
[Params("pg73346.txt")]
public string? SourceFileName { get; set; }
[GlobalSetup]
public void Setup()
{
_sourceText = File.ReadAllText(SourceFileName!);
_regex = new(RegexPattern);
_regexCompiled = new(RegexPattern, RegexOptions.Compiled);
_generatedRegex = GetGeneratedRegex();
_generatedRegexCompiled = GetGeneratedRegexCompiled();
}
[Benchmark(Baseline = true)]
public MatchCollection Static()
{
return Regex.Matches(_sourceText!, RegexPattern!);
}
[Benchmark]
public MatchCollection New()
{
Regex regex = new(RegexPattern!);
return regex.Matches(_sourceText!);
}
[Benchmark]
public MatchCollection New_Compiled()
{
Regex regex = new(RegexPattern!, RegexOptions.Compiled);
return regex.Matches(_sourceText!);
}
[Benchmark]
public MatchCollection Cached()
{
return _regex!.Matches(_sourceText!);
}
[Benchmark]
public MatchCollection Cached_Compiled()
{
return _regexCompiled!.Matches(_sourceText!);
}
[Benchmark]
public MatchCollection Generated()
{
return GetGeneratedRegex().Matches(_sourceText!);
}
[Benchmark]
public MatchCollection Generated_Cached()
{
return _generatedRegex!.Matches(_sourceText!);
}
[Benchmark]
public MatchCollection Generated_Compiled()
{
return GetGeneratedRegexCompiled().Matches(_sourceText!);
}
[Benchmark]
public MatchCollection Generated_Cached_Compiled()
{
return _generatedRegexCompiled!.Matches(_sourceText!);
}
}
For our benchmarks, we’ll treat the static method on the Regex class as the baseline — just so we have something to anchor to when looking at the performance results.
With the BenchmarkDotNet code out of the way, let’s see the results:
From above we can see that creating a new Regex instance every time you want to perform a match is 100x slower than the static method. This is incredible — you should NOT do this if performance is important to you! But it gets worse… If you do this AND provide the compile flag, it’s almost 1000x as bad, 10x worse than newing it up every time. These are two things you should avoid doing.
We can see with the cached variations that follow that we can effectively invert the situation, giving us a slight boost over the static method. While these runtimes are very fast, it looks like in these situations it’s nearly 30% faster. But temper your expectations as I’m not convinced this scales with different data sets and different patterns!
The source-generated C# regular expressions are also much faster than the static method — but seem roughly on par with the two prior benchmarks. While the source-generated regular expressions do cache and calling the source-generated method should be no overhead, there are two benchmark variations that suggest it’s marginally faster to keep your own cache. These could simply be outliers though given how close the results are.
You can check out the video for a full walk-through on these C# regex benchmarks:
The takeaways for optimizing C# Regex performance: Stop declaring regular expressions in C# right before you go to use them! And even worse, stop declaring them with the compiled flag if you’re declaring them right before using them! These two things will crush your performance.
Otherwise, it seems like the Regex class with static method across the board is pretty safe, but you get the most benefit out of compiling and caching your regex. And according to Microsoft, can be improved even further with the C# Regex source generators in many situations.
If you found this useful and you’re looking for more learning opportunities, consider subscribing to my free weekly software engineering newsletter and check out my free videos on YouTube! Meet other like-minded software engineers and join my Discord community!
Want More Dev Leader Content?
Follow along on this platform if you haven’t already!
Subscribe to my free weekly software engineering and dotnet-focused newsletter. I include exclusive articles and early access to videos:
SUBSCRIBE FOR FREE
Looking for courses? Check out my offerings:
VIEW COURSES
E-Books & other resources:
VIEW RESOURCES
Watch hundreds of full-length videos on my YouTube channel:
VISIT CHANNEL
Visit my website for hundreds of articles on various software engineering topics (including code snippets):
VISIT WEBSITE
Check out the repository with many code examples from my articles and videos on GitHub:
VIEW REPOSITORY
