Crate regex

source ·
Expand description

This crate provides routines for searching strings for matches of a regular expression (aka “regex”). The regex syntax supported by this crate is similar to other regex engines, but it lacks several features that are not known how to implement efficiently. This includes, but is not limited to, look-around and backreferences. In exchange, all regex searches in this crate have worst case O(m * n) time complexity, where m is proportional to the size of the regex and n is proportional to the size of the string being searched.

If you just want API documentation, then skip to the Regex type. Otherwise, here’s a quick example showing one way of parsing the output of a grep-like program:

use regex::Regex;

let re = Regex::new(r"(?m)^([^:]+):([0-9]+):(.+)$").unwrap();
let hay = "\
path/to/foo:54:Blue Harvest
path/to/bar:90:Something, Something, Something, Dark Side
path/to/baz:3:It's a Trap!

let mut results = vec![];
for (_, [path, lineno, line]) in re.captures_iter(hay).map(|c| c.extract()) {
    results.push((path, lineno.parse::<u64>()?, line));
assert_eq!(results, vec![
    ("path/to/foo", 54, "Blue Harvest"),
    ("path/to/bar", 90, "Something, Something, Something, Dark Side"),
    ("path/to/baz", 3, "It's a Trap!"),


The primary type in this crate is a Regex. Its most important methods are as follows:

  • Regex::new compiles a regex using the default configuration. A RegexBuilder permits setting a non-default configuration. (For example, case insensitive matching, verbose mode and others.)
  • Regex::is_match reports whether a match exists in a particular haystack.
  • Regex::find reports the byte offsets of a match in a haystack, if one exists. Regex::find_iter returns an iterator over all such matches.
  • Regex::captures returns a Captures, which reports both the byte offsets of a match in a haystack and the byte offsets of each matching capture group from the regex in the haystack. Regex::captures_iter returns an iterator over all such matches.

There is also a RegexSet, which permits searching for multiple regex patterns simultaneously in a single search. However, it currently only reports which patterns match and not the byte offsets of a match.

Otherwise, this top-level crate documentation is organized as follows:

  • Usage shows how to add the regex crate to your Rust project.
  • Examples provides a limited selection of regex search examples.
  • Performance provides a brief summary of how to optimize regex searching speed.
  • Unicode discusses support for non-ASCII patterns.
  • Syntax enumerates the specific regex syntax supported by this crate.
  • Untrusted input discusses how this crate deals with regex patterns or haystacks that are untrusted.
  • Crate features documents the Cargo features that can be enabled or disabled for this crate.
  • Other crates links to other crates in the regex family.


The regex crate is on and can be used by adding regex to your dependencies in your project’s Cargo.toml. Or more simply, just run cargo add regex.

Here is a complete example that creates a new Rust project, adds a dependency on regex, creates the source code for a regex search and then runs the program.

First, create the project in a new directory:

$ mkdir regex-example
$ cd regex-example
$ cargo init

Second, add a dependency on regex:

$ cargo add regex

Third, edit src/ Delete what’s there and replace it with this:

use regex::Regex;

fn main() {
    let re = Regex::new(r"Hello (?<name>\w+)!").unwrap();
    let Some(caps) = re.captures("Hello Murphy!") else {
        println!("no match!");
    println!("The name is: {}", &caps["name"]);

Fourth, run it with cargo run:

$ cargo run
   Compiling memchr v2.5.0
   Compiling regex-syntax v0.7.1
   Compiling aho-corasick v1.0.1
   Compiling regex v1.8.1
   Compiling regex-example v0.1.0 (/tmp/regex-example)
    Finished dev [unoptimized + debuginfo] target(s) in 4.22s
     Running `target/debug/regex-example`
The name is: Murphy

The first time you run the program will show more output like above. But subsequent runs shouldn’t have to re-compile the dependencies.


This section provides a few examples, in tutorial style, showing how to search a haystack with a regex. There are more examples throughout the API documentation.

Before starting though, it’s worth defining a few terms:

  • A regex is a Rust value whose type is Regex. We use re as a variable name for a regex.
  • A pattern is the string that is used to build a regex. We use pat as a variable name for a pattern.
  • A haystack is the string that is searched by a regex. We use hay as a variable name for a haystack.

Sometimes the words “regex” and “pattern” are used interchangeably.

General use of regular expressions in this crate proceeds by compiling a pattern into a regex, and then using that regex to search, split or replace parts of a haystack.

§Example: find a middle initial

We’ll start off with a very simple example: a regex that looks for a specific name but uses a wildcard to match a middle initial. Our pattern serves as something like a template that will match a particular name with any middle initial.

use regex::Regex;

// We use 'unwrap()' here because it would be a bug in our program if the
// pattern failed to compile to a regex. Panicking in the presence of a bug
// is okay.
let re = Regex::new(r"Homer (.)\. Simpson").unwrap();
let hay = "Homer J. Simpson";
let Some(caps) = re.captures(hay) else { return };
assert_eq!("J", &caps[1]);

There are a few things worth noticing here in our first example:

  • The . is a special pattern meta character that means “match any single character except for new lines.” (More precisely, in this crate, it means “match any UTF-8 encoding of any Unicode scalar value other than \n.”)
  • We can match an actual . literally by escaping it, i.e., \..
  • We use Rust’s raw strings to avoid needing to deal with escape sequences in both the regex pattern syntax and in Rust’s string literal syntax. If we didn’t use raw strings here, we would have had to use \\. to match a literal . character. That is, r"\." and "\\." are equivalent patterns.
  • We put our wildcard . instruction in parentheses. These parentheses have a special meaning that says, “make whatever part of the haystack matches within these parentheses available as a capturing group.” After finding a match, we access this capture group with &caps[1].

Otherwise, we execute a search using re.captures(hay) and return from our function if no match occurred. We then reference the middle initial by asking for the part of the haystack that matched the capture group indexed at 1. (The capture group at index 0 is implicit and always corresponds to the entire match. In this case, that’s Homer J. Simpson.)

§Example: named capture groups

Continuing from our middle initial example above, we can tweak the pattern slightly to give a name to the group that matches the middle initial:

use regex::Regex;

// Note that (?P<middle>.) is a different way to spell the same thing.
let re = Regex::new(r"Homer (?<middle>.)\. Simpson").unwrap();
let hay = "Homer J. Simpson";
let Some(caps) = re.captures(hay) else { return };
assert_eq!("J", &caps["middle"]);

Giving a name to a group can be useful when there are multiple groups in a pattern. It makes the code referring to those groups a bit easier to understand.

§Example: validating a particular date format

This examples shows how to confirm whether a haystack, in its entirety, matches a particular date format:

use regex::Regex;

let re = Regex::new(r"^\d{4}-\d{2}-\d{2}$").unwrap();

Notice the use of the ^ and $ anchors. In this crate, every regex search is run with an implicit (?s:.)*? at the beginning of its pattern, which allows the regex to match anywhere in a haystack. Anchors, as above, can be used to ensure that the full haystack matches a pattern.

This crate is also Unicode aware by default, which means that \d might match more than you might expect it to. For example:

use regex::Regex;

let re = Regex::new(r"^\d{4}-\d{2}-\d{2}$").unwrap();

To only match an ASCII decimal digit, all of the following are equivalent:

  • [0-9]
  • (?-u:\d)
  • [[:digit:]]
  • [\d&&\p{ascii}]

§Example: finding dates in a haystack

In the previous example, we showed how one might validate that a haystack, in its entirety, corresponded to a particular date format. But what if we wanted to extract all things that look like dates in a specific format from a haystack? To do this, we can use an iterator API to find all matches (notice that we’ve removed the anchors and switched to looking for ASCII-only digits):

use regex::Regex;

let re = Regex::new(r"[0-9]{4}-[0-9]{2}-[0-9]{2}").unwrap();
let hay = "What do 1865-04-14, 1881-07-02, 1901-09-06 and 1963-11-22 have in common?";
// 'm' is a 'Match', and 'as_str()' returns the matching part of the haystack.
let dates: Vec<&str> = re.find_iter(hay).map(|m| m.as_str()).collect();
assert_eq!(dates, vec![

We can also iterate over Captures values instead of Match values, and that in turn permits accessing each component of the date via capturing groups:

use regex::Regex;

let re = Regex::new(r"(?<y>[0-9]{4})-(?<m>[0-9]{2})-(?<d>[0-9]{2})").unwrap();
let hay = "What do 1865-04-14, 1881-07-02, 1901-09-06 and 1963-11-22 have in common?";
// 'm' is a 'Match', and 'as_str()' returns the matching part of the haystack.
let dates: Vec<(&str, &str, &str)> = re.captures_iter(hay).map(|caps| {
    // The unwraps are okay because every capture group must match if the whole
    // regex matches, and in this context, we know we have a match.
    // Note that we use `"y").unwrap().as_str()` instead of
    // `&caps["y"]` because the lifetime of the former is the same as the
    // lifetime of `hay` above, but the lifetime of the latter is tied to the
    // lifetime of `caps` due to how the `Index` trait is defined.
    let year ="y").unwrap().as_str();
    let month ="m").unwrap().as_str();
    let day ="d").unwrap().as_str();
    (year, month, day)
assert_eq!(dates, vec![
    ("1865", "04", "14"),
    ("1881", "07", "02"),
    ("1901", "09", "06"),
    ("1963", "11", "22"),

§Example: simpler capture group extraction

One can use Captures::extract to make the code from the previous example a bit simpler in this case:

use regex::Regex;

let re = Regex::new(r"([0-9]{4})-([0-9]{2})-([0-9]{2})").unwrap();
let hay = "What do 1865-04-14, 1881-07-02, 1901-09-06 and 1963-11-22 have in common?";
let dates: Vec<(&str, &str, &str)> = re.captures_iter(hay).map(|caps| {
    let (_, [year, month, day]) = caps.extract();
    (year, month, day)
assert_eq!(dates, vec![
    ("1865", "04", "14"),
    ("1881", "07", "02"),
    ("1901", "09", "06"),
    ("1963", "11", "22"),

Captures::extract works by ensuring that the number of matching groups match the number of groups requested via the [year, month, day] syntax. If they do, then the substrings for each corresponding capture group are automatically returned in an appropriately sized array. Rust’s syntax for pattern matching arrays does the rest.

§Example: replacement with named capture groups

Building on the previous example, perhaps we’d like to rearrange the date formats. This can be done by finding each match and replacing it with something different. The Regex::replace_all routine provides a convenient way to do this, including by supporting references to named groups in the replacement string:

use regex::Regex;

let re = Regex::new(r"(?<y>\d{4})-(?<m>\d{2})-(?<d>\d{2})").unwrap();
let before = "1973-01-05, 1975-08-25 and 1980-10-18";
let after = re.replace_all(before, "$m/$d/$y");
assert_eq!(after, "01/05/1973, 08/25/1975 and 10/18/1980");

The replace methods are actually polymorphic in the replacement, which provides more flexibility than is seen here. (See the documentation for Regex::replace for more details.)

§Example: verbose mode

When your regex gets complicated, you might consider using something other than regex. But if you stick with regex, you can use the x flag to enable insignificant whitespace mode or “verbose mode.” In this mode, whitespace is treated as insignificant and one may write comments. This may make your patterns easier to comprehend.

use regex::Regex;

let re = Regex::new(r"(?x)
  (?P<y>\d{4}) # the year, including all Unicode digits
  (?P<m>\d{2}) # the month, including all Unicode digits
  (?P<d>\d{2}) # the day, including all Unicode digits

let before = "1973-01-05, 1975-08-25 and 1980-10-18";
let after = re.replace_all(before, "$m/$d/$y");
assert_eq!(after, "01/05/1973, 08/25/1975 and 10/18/1980");

If you wish to match against whitespace in this mode, you can still use \s, \n, \t, etc. For escaping a single space character, you can escape it directly with \ , use its hex character code \x20 or temporarily disable the x flag, e.g., (?-x: ).

§Example: match multiple regular expressions simultaneously

This demonstrates how to use a RegexSet to match multiple (possibly overlapping) regexes in a single scan of a haystack:

use regex::RegexSet;

let set = RegexSet::new(&[

// Iterate over and collect all of the matches. Each match corresponds to the
// ID of the matching pattern.
let matches: Vec<_> = set.matches("foobar").into_iter().collect();
assert_eq!(matches, vec![0, 2, 3, 4, 6]);

// You can also test whether a particular regex matched:
let matches = set.matches("foobar");


This section briefly discusses a few concerns regarding the speed and resource usage of regexes.

§Only ask for what you need

When running a search with a regex, there are generally three different types of information one can ask for:

  1. Does a regex match in a haystack?
  2. Where does a regex match in a haystack?
  3. Where do each of the capturing groups match in a haystack?

Generally speaking, this crate could provide a function to answer only #3, which would subsume #1 and #2 automatically. However, it can be significantly more expensive to compute the location of capturing group matches, so it’s best not to do it if you don’t need to.

Therefore, only ask for what you need. For example, don’t use Regex::find if you only need to test if a regex matches a haystack. Use Regex::is_match instead.

§Unicode can impact memory usage and search speed

This crate has first class support for Unicode and it is enabled by default. In many cases, the extra memory required to support it will be negligible and it typically won’t impact search speed. But it can in some cases.

With respect to memory usage, the impact of Unicode principally manifests through the use of Unicode character classes. Unicode character classes tend to be quite large. For example, \w by default matches around 140,000 distinct codepoints. This requires additional memory, and tends to slow down regex compilation. While a \w here and there is unlikely to be noticed, writing \w{100} will for example result in quite a large regex by default. Indeed, \w is considerably larger than its ASCII-only version, so if your requirements are satisfied by ASCII, it’s probably a good idea to stick to ASCII classes. The ASCII-only version of \w can be spelled in a number of ways. All of the following are equivalent:

  • [0-9A-Za-z_]
  • (?-u:\w)
  • [[:word:]]
  • [\w&&\p{ascii}]

With respect to search speed, Unicode tends to be handled pretty well, even when using large Unicode character classes. However, some of the faster internal regex engines cannot handle a Unicode aware word boundary assertion. So if you don’t need Unicode-aware word boundary assertions, you might consider using (?-u:\b) instead of \b, where the former uses an ASCII-only definition of a word character.

§Literals might accelerate searches

This crate tends to be quite good at recognizing literals in a regex pattern and using them to accelerate a search. If it is at all possible to include some kind of literal in your pattern, then it might make search substantially faster. For example, in the regex \w+@\w+, the engine will look for occurrences of @ and then try a reverse match for \w+ to find the start position.

§Avoid re-compiling regexes, especially in a loop

It is an anti-pattern to compile the same pattern in a loop since regex compilation is typically expensive. (It takes anywhere from a few microseconds to a few milliseconds depending on the size of the pattern.) Not only is compilation itself expensive, but this also prevents optimizations that reuse allocations internally to the regex engine.

In Rust, it can sometimes be a pain to pass regexes around if they’re used from inside a helper function. Instead, we recommend using crates like once_cell and lazy_static to ensure that patterns are compiled exactly once.

This example shows how to use once_cell:

use {

fn some_helper_function(haystack: &str) -> bool {
    static RE: Lazy<Regex> = Lazy::new(|| Regex::new(r"...").unwrap());

fn main() {

Specifically, in this example, the regex will be compiled when it is used for the first time. On subsequent uses, it will reuse the previously built Regex. Notice how one can define the Regex locally to a specific function.

§Sharing a regex across threads can result in contention

While a single Regex can be freely used from multiple threads simultaneously, there is a small synchronization cost that must be paid. Generally speaking, one shouldn’t expect to observe this unless the principal task in each thread is searching with the regex and most searches are on short haystacks. In this case, internal contention on shared resources can spike and increase latency, which in turn may slow down each individual search.

One can work around this by cloning each Regex before sending it to another thread. The cloned regexes will still share the same internal read-only portion of its compiled state (it’s reference counted), but each thread will get optimized access to the mutable space that is used to run a search. In general, there is no additional cost in memory to doing this. The only cost is the added code complexity required to explicitly clone the regex. (If you share the same Regex across multiple threads, each thread still gets its own mutable space, but accessing that space is slower.)


This section discusses what kind of Unicode support this regex library has. Before showing some examples, we’ll summarize the relevant points:

  • This crate almost fully implements “Basic Unicode Support” (Level 1) as specified by the Unicode Technical Standard #18. The full details of what is supported are documented in in the root of the regex crate repository. There is virtually no support for “Extended Unicode Support” (Level 2) from UTS#18.
  • The top-level Regex runs searches as if iterating over each of the codepoints in the haystack. That is, the fundamental atom of matching is a single codepoint.
  • bytes::Regex, in contrast, permits disabling Unicode mode for part of all of your pattern in all cases. When Unicode mode is disabled, then a search is run as if iterating over each byte in the haystack. That is, the fundamental atom of matching is a single byte. (A top-level Regex also permits disabling Unicode and thus matching as if it were one byte at a time, but only when doing so wouldn’t permit matching invalid UTF-8.)
  • When Unicode mode is enabled (the default), . will match an entire Unicode scalar value, even when it is encoded using multiple bytes. When Unicode mode is disabled (e.g., (?-u:.)), then . will match a single byte in all cases.
  • The character classes \w, \d and \s are all Unicode-aware by default. Use (?-u:\w), (?-u:\d) and (?-u:\s) to get their ASCII-only definitions.
  • Similarly, \b and \B use a Unicode definition of a “word” character. To get ASCII-only word boundaries, use (?-u:\b) and (?-u:\B). This also applies to the special word boundary assertions. (That is, \b{start}, \b{end}, \b{start-half}, \b{end-half}.)
  • ^ and $ are not Unicode-aware in multi-line mode. Namely, they only recognize \n (assuming CRLF mode is not enabled) and not any of the other forms of line terminators defined by Unicode.
  • Case insensitive searching is Unicode-aware and uses simple case folding.
  • Unicode general categories, scripts and many boolean properties are available by default via the \p{property name} syntax.
  • In all cases, matches are reported using byte offsets. Or more precisely, UTF-8 code unit offsets. This permits constant time indexing and slicing of the haystack.

Patterns themselves are only interpreted as a sequence of Unicode scalar values. This means you can use Unicode characters directly in your pattern:

use regex::Regex;

let re = Regex::new(r"(?i)Δ+").unwrap();
let m = re.find("ΔδΔ").unwrap();
assert_eq!((0, 6), (m.start(), m.end()));
// alternatively:
assert_eq!(0..6, m.range());

As noted above, Unicode general categories, scripts, script extensions, ages and a smattering of boolean properties are available as character classes. For example, you can match a sequence of numerals, Greek or Cherokee letters:

use regex::Regex;

let re = Regex::new(r"[\pN\p{Greek}\p{Cherokee}]+").unwrap();
let m = re.find("abcΔᎠβⅠᏴγδⅡxyz").unwrap();
assert_eq!(3..23, m.range());

While not specific to Unicode, this library also supports character class set operations. Namely, one can nest character classes arbitrarily and perform set operations on them. Those set operations are union (the default), intersection, difference and symmetric difference. These set operations tend to be most useful with Unicode character classes. For example, to match any codepoint that is both in the Greek script and in the Letter general category:

use regex::Regex;

let re = Regex::new(r"[\p{Greek}&&\pL]+").unwrap();
let subs: Vec<&str> = re.find_iter("ΔδΔ𐅌ΔδΔ").map(|m| m.as_str()).collect();
assert_eq!(subs, vec!["ΔδΔ", "ΔδΔ"]);

// If we just matches on Greek, then all codepoints would match!
let re = Regex::new(r"\p{Greek}+").unwrap();
let subs: Vec<&str> = re.find_iter("ΔδΔ𐅌ΔδΔ").map(|m| m.as_str()).collect();
assert_eq!(subs, vec!["ΔδΔ𐅌ΔδΔ"]);

§Opt out of Unicode support

The bytes::Regex type that can be used to search &[u8] haystacks. By default, haystacks are conventionally treated as UTF-8 just like it is with the main Regex type. However, this behavior can be disabled by turning off the u flag, even if doing so could result in matching invalid UTF-8. For example, when the u flag is disabled, . will match any byte instead of any Unicode scalar value.

Disabling the u flag is also possible with the standard &str-based Regex type, but it is only allowed where the UTF-8 invariant is maintained. For example, (?-u:\w) is an ASCII-only \w character class and is legal in an &str-based Regex, but (?-u:\W) will attempt to match any byte that isn’t in (?-u:\w), which in turn includes bytes that are invalid UTF-8. Similarly, (?-u:\xFF) will attempt to match the raw byte \xFF (instead of U+00FF), which is invalid UTF-8 and therefore is illegal in &str-based regexes.

Finally, since Unicode support requires bundling large Unicode data tables, this crate exposes knobs to disable the compilation of those data tables, which can be useful for shrinking binary size and reducing compilation times. For details on how to do that, see the section on crate features.


The syntax supported in this crate is documented below.

Note that the regular expression parser and abstract syntax are exposed in a separate crate, regex-syntax.

§Matching one character

.             any character except new line (includes new line with s flag)
[0-9]         any ASCII digit
\d            digit (\p{Nd})
\D            not digit
\pX           Unicode character class identified by a one-letter name
\p{Greek}     Unicode character class (general category or script)
\PX           Negated Unicode character class identified by a one-letter name
\P{Greek}     negated Unicode character class (general category or script)

§Character classes

[xyz]         A character class matching either x, y or z (union).
[^xyz]        A character class matching any character except x, y and z.
[a-z]         A character class matching any character in range a-z.
[[:alpha:]]   ASCII character class ([A-Za-z])
[[:^alpha:]]  Negated ASCII character class ([^A-Za-z])
[x[^xyz]]     Nested/grouping character class (matching any character except y and z)
[a-y&&xyz]    Intersection (matching x or y)
[0-9&&[^4]]   Subtraction using intersection and negation (matching 0-9 except 4)
[0-9--4]      Direct subtraction (matching 0-9 except 4)
[a-g~~b-h]    Symmetric difference (matching `a` and `h` only)
[\[\]]        Escaping in character classes (matching [ or ])
[a&&b]        An empty character class matching nothing

Any named character class may appear inside a bracketed [...] character class. For example, [\p{Greek}[:digit:]] matches any ASCII digit or any codepoint in the Greek script. [\p{Greek}&&\pL] matches Greek letters.

Precedence in character classes, from most binding to least:

  1. Ranges: [a-cd] == [[a-c]d]
  2. Union: [ab&&bc] == [[ab]&&[bc]]
  3. Intersection, difference, symmetric difference. All three have equivalent precedence, and are evaluated in left-to-right order. For example, [\pL--\p{Greek}&&\p{Uppercase}] == [[\pL--\p{Greek}]&&\p{Uppercase}].
  4. Negation: [^a-z&&b] == [^[a-z&&b]].


xy    concatenation (x followed by y)
x|y   alternation (x or y, prefer x)

This example shows how an alternation works, and what it means to prefer a branch in the alternation over subsequent branches.

use regex::Regex;

let haystack = "samwise";
// If 'samwise' comes first in our alternation, then it is
// preferred as a match, even if the regex engine could
// technically detect that 'sam' led to a match earlier.
let re = Regex::new(r"samwise|sam").unwrap();
assert_eq!("samwise", re.find(haystack).unwrap().as_str());
// But if 'sam' comes first, then it will match instead.
// In this case, it is impossible for 'samwise' to match
// because 'sam' is a prefix of it.
let re = Regex::new(r"sam|samwise").unwrap();
assert_eq!("sam", re.find(haystack).unwrap().as_str());


x*        zero or more of x (greedy)
x+        one or more of x (greedy)
x?        zero or one of x (greedy)
x*?       zero or more of x (ungreedy/lazy)
x+?       one or more of x (ungreedy/lazy)
x??       zero or one of x (ungreedy/lazy)
x{n,m}    at least n x and at most m x (greedy)
x{n,}     at least n x (greedy)
x{n}      exactly n x
x{n,m}?   at least n x and at most m x (ungreedy/lazy)
x{n,}?    at least n x (ungreedy/lazy)
x{n}?     exactly n x

§Empty matches

^               the beginning of a haystack (or start-of-line with multi-line mode)
$               the end of a haystack (or end-of-line with multi-line mode)
\A              only the beginning of a haystack (even with multi-line mode enabled)
\z              only the end of a haystack (even with multi-line mode enabled)
\b              a Unicode word boundary (\w on one side and \W, \A, or \z on other)
\B              not a Unicode word boundary
\b{start}, \<   a Unicode start-of-word boundary (\W|\A on the left, \w on the right)
\b{end}, \>     a Unicode end-of-word boundary (\w on the left, \W|\z on the right))
\b{start-half}  half of a Unicode start-of-word boundary (\W|\A on the left)
\b{end-half}    half of a Unicode end-of-word boundary (\W|\z on the right)

The empty regex is valid and matches the empty string. For example, the empty regex matches abc at positions 0, 1, 2 and 3. When using the top-level Regex on &str haystacks, an empty match that splits a codepoint is guaranteed to never be returned. However, such matches are permitted when using a bytes::Regex. For example:

let re = regex::Regex::new(r"").unwrap();
let ranges: Vec<_> = re.find_iter("💩").map(|m| m.range()).collect();
assert_eq!(ranges, vec![0..0, 4..4]);

let re = regex::bytes::Regex::new(r"").unwrap();
let ranges: Vec<_> = re.find_iter("💩".as_bytes()).map(|m| m.range()).collect();
assert_eq!(ranges, vec![0..0, 1..1, 2..2, 3..3, 4..4]);

Note that an empty regex is distinct from a regex that can never match. For example, the regex [a&&b] is a character class that represents the intersection of a and b. That intersection is empty, which means the character class is empty. Since nothing is in the empty set, [a&&b] matches nothing, not even the empty string.

§Grouping and flags

(exp)          numbered capture group (indexed by opening parenthesis)
(?P<name>exp)  named (also numbered) capture group (names must be alpha-numeric)
(?<name>exp)   named (also numbered) capture group (names must be alpha-numeric)
(?:exp)        non-capturing group
(?flags)       set flags within current group
(?flags:exp)   set flags for exp (non-capturing)

Capture group names must be any sequence of alpha-numeric Unicode codepoints, in addition to ., _, [ and ]. Names must start with either an _ or an alphabetic codepoint. Alphabetic codepoints correspond to the Alphabetic Unicode property, while numeric codepoints correspond to the union of the Decimal_Number, Letter_Number and Other_Number general categories.

Flags are each a single character. For example, (?x) sets the flag x and (?-x) clears the flag x. Multiple flags can be set or cleared at the same time: (?xy) sets both the x and y flags and (?x-y) sets the x flag and clears the y flag.

All flags are by default disabled unless stated otherwise. They are:

i     case-insensitive: letters match both upper and lower case
m     multi-line mode: ^ and $ match begin/end of line
s     allow . to match \n
R     enables CRLF mode: when multi-line mode is enabled, \r\n is used
U     swap the meaning of x* and x*?
u     Unicode support (enabled by default)
x     verbose mode, ignores whitespace and allow line comments (starting with `#`)

Note that in verbose mode, whitespace is ignored everywhere, including within character classes. To insert whitespace, use its escaped form or a hex literal. For example, \ or \x20 for an ASCII space.

Flags can be toggled within a pattern. Here’s an example that matches case-insensitively for the first part but case-sensitively for the second part:

use regex::Regex;

let re = Regex::new(r"(?i)a+(?-i)b+").unwrap();
let m = re.find("AaAaAbbBBBb").unwrap();
assert_eq!(m.as_str(), "AaAaAbb");

Notice that the a+ matches either a or A, but the b+ only matches b.

Multi-line mode means ^ and $ no longer match just at the beginning/end of the input, but also at the beginning/end of lines:

use regex::Regex;

let re = Regex::new(r"(?m)^line \d+").unwrap();
let m = re.find("line one\nline 2\n").unwrap();
assert_eq!(m.as_str(), "line 2");

Note that ^ matches after new lines, even at the end of input:

use regex::Regex;

let re = Regex::new(r"(?m)^").unwrap();
let m = re.find_iter("test\n").last().unwrap();
assert_eq!((m.start(), m.end()), (5, 5));

When both CRLF mode and multi-line mode are enabled, then ^ and $ will match either \r and \n, but never in the middle of a \r\n:

use regex::Regex;

let re = Regex::new(r"(?mR)^foo$").unwrap();
let m = re.find("\r\nfoo\r\n").unwrap();
assert_eq!(m.as_str(), "foo");

Unicode mode can also be selectively disabled, although only when the result would not match invalid UTF-8. One good example of this is using an ASCII word boundary instead of a Unicode word boundary, which might make some regex searches run faster:

use regex::Regex;

let re = Regex::new(r"(?-u:\b).+(?-u:\b)").unwrap();
let m = re.find("$$abc$$").unwrap();
assert_eq!(m.as_str(), "abc");

§Escape sequences

Note that this includes all possible escape sequences, even ones that are documented elsewhere.

\*              literal *, applies to all ASCII except [0-9A-Za-z<>]
\a              bell (\x07)
\f              form feed (\x0C)
\t              horizontal tab
\n              new line
\r              carriage return
\v              vertical tab (\x0B)
\A              matches at the beginning of a haystack
\z              matches at the end of a haystack
\b              word boundary assertion
\B              negated word boundary assertion
\b{start}, \<   start-of-word boundary assertion
\b{end}, \>     end-of-word boundary assertion
\b{start-half}  half of a start-of-word boundary assertion
\b{end-half}    half of a end-of-word boundary assertion
\123            octal character code, up to three digits (when enabled)
\x7F            hex character code (exactly two digits)
\x{10FFFF}      any hex character code corresponding to a Unicode code point
\u007F          hex character code (exactly four digits)
\u{7F}          any hex character code corresponding to a Unicode code point
\U0000007F      hex character code (exactly eight digits)
\U{7F}          any hex character code corresponding to a Unicode code point
\p{Letter}      Unicode character class
\P{Letter}      negated Unicode character class
\d, \s, \w      Perl character class
\D, \S, \W      negated Perl character class

§Perl character classes (Unicode friendly)

These classes are based on the definitions provided in UTS#18:

\d     digit (\p{Nd})
\D     not digit
\s     whitespace (\p{White_Space})
\S     not whitespace
\w     word character (\p{Alphabetic} + \p{M} + \d + \p{Pc} + \p{Join_Control})
\W     not word character

§ASCII character classes

These classes are based on the definitions provided in UTS#18:

[[:alnum:]]    alphanumeric ([0-9A-Za-z])
[[:alpha:]]    alphabetic ([A-Za-z])
[[:ascii:]]    ASCII ([\x00-\x7F])
[[:blank:]]    blank ([\t ])
[[:cntrl:]]    control ([\x00-\x1F\x7F])
[[:digit:]]    digits ([0-9])
[[:graph:]]    graphical ([!-~])
[[:lower:]]    lower case ([a-z])
[[:print:]]    printable ([ -~])
[[:punct:]]    punctuation ([!-/:-@\[-`{-~])
[[:space:]]    whitespace ([\t\n\v\f\r ])
[[:upper:]]    upper case ([A-Z])
[[:word:]]     word characters ([0-9A-Za-z_])
[[:xdigit:]]   hex digit ([0-9A-Fa-f])

§Untrusted input

This crate is meant to be able to run regex searches on untrusted haystacks without fear of ReDoS. This crate also, to a certain extent, supports untrusted patterns.

This crate differs from most (but not all) other regex engines in that it doesn’t use unbounded backtracking to run a regex search. In those cases, one generally cannot use untrusted patterns or untrusted haystacks because it can be very difficult to know whether a particular pattern will result in catastrophic backtracking or not.

We’ll first discuss how this crate deals with untrusted inputs and then wrap it up with a realistic discussion about what practice really looks like.


Outside of clearly documented cases, most APIs in this crate are intended to never panic regardless of the inputs given to them. For example, Regex::new, Regex::is_match, Regex::find and Regex::captures should never panic. That is, it is an API promise that those APIs will never panic no matter what inputs are given to them. With that said, regex engines are complicated beasts, and providing a rock solid guarantee that these APIs literally never panic is essentially equivalent to saying, “there are no bugs in this library.” That is a bold claim, and not really one that can be feasibly made with a straight face.

Don’t get the wrong impression here. This crate is extensively tested, not just with unit and integration tests, but also via fuzz testing. For example, this crate is part of the OSS-fuzz project. Panics should be incredibly rare, but it is possible for bugs to exist, and thus possible for a panic to occur. If you need a rock solid guarantee against panics, then you should wrap calls into this library with std::panic::catch_unwind.

It’s also worth pointing out that this library will generally panic when other regex engines would commit undefined behavior. When undefined behavior occurs, your program might continue as if nothing bad has happened, but it also might mean your program is open to the worst kinds of exploits. In contrast, the worst thing a panic can do is a denial of service.

§Untrusted patterns

The principal way this crate deals with them is by limiting their size by default. The size limit can be configured via RegexBuilder::size_limit. The idea of a size limit is that compiling a pattern into a Regex will fail if it becomes “too big.” Namely, while most resources consumed by compiling a regex are approximately proportional (albeit with some high constant factors in some cases, such as with Unicode character classes) to the length of the pattern itself, there is one particular exception to this: counted repetitions. Namely, this pattern:


Is equivalent to this pattern:


In both of these cases, the actual pattern string is quite small, but the resulting Regex value is quite large. Indeed, as the first pattern shows, it isn’t enough to locally limit the size of each repetition because they can be stacked in a way that results in exponential growth.

To provide a bit more context, a simplified view of regex compilation looks like this:

  • The pattern string is parsed into a structured representation called an AST. Counted repetitions are not expanded and Unicode character classes are not looked up in this stage. That is, the size of the AST is proportional to the size of the pattern with “reasonable” constant factors. In other words, one can reasonably limit the memory used by an AST by limiting the length of the pattern string.
  • The AST is translated into an HIR. Counted repetitions are still not expanded at this stage, but Unicode character classes are embedded into the HIR. The memory usage of a HIR is still proportional to the length of the original pattern string, but the constant factors—mostly as a result of Unicode character classes—can be quite high. Still though, the memory used by an HIR can be reasonably limited by limiting the length of the pattern string.
  • The HIR is compiled into a Thompson NFA. This is the stage at which something like \w{5} is rewritten to \w\w\w\w\w. Thus, this is the stage at which RegexBuilder::size_limit is enforced. If the NFA exceeds the configured size, then this stage will fail.

The size limit helps avoid two different kinds of exorbitant resource usage:

  • It avoids permitting exponential memory usage based on the size of the pattern string.
  • It avoids long search times. This will be discussed in more detail in the next section, but worst case search time is dependent on the size of the regex. So keeping regexes limited to a reasonable size is also a way of keeping search times reasonable.

Finally, it’s worth pointing out that regex compilation is guaranteed to take worst case O(m) time, where m is proportional to the size of regex. The size of the regex here is after the counted repetitions have been expanded.

Advice for those using untrusted regexes: limit the pattern length to something small and expand it as needed. Configure RegexBuilder::size_limit to something small and then expand it as needed.

§Untrusted haystacks

The main way this crate guards against searches from taking a long time is by using algorithms that guarantee a O(m * n) worst case time and space bound. Namely:

  • m is proportional to the size of the regex, where the size of the regex includes the expansion of all counted repetitions. (See the previous section on untrusted patterns.)
  • n is proportional to the length, in bytes, of the haystack.

In other words, if you consider m to be a constant (for example, the regex pattern is a literal in the source code), then the search can be said to run in “linear time.” Or equivalently, “linear time with respect to the size of the haystack.”

But the m factor here is important not to ignore. If a regex is particularly big, the search times can get quite slow. This is why, in part, RegexBuilder::size_limit exists.

Advice for those searching untrusted haystacks: As long as your regexes are not enormous, you should expect to be able to search untrusted haystacks without fear. If you aren’t sure, you should benchmark it. Unlike backtracking engines, if your regex is so big that it’s likely to result in slow searches, this is probably something you’ll be able to observe regardless of what the haystack is made up of.

§Iterating over matches

One thing that is perhaps easy to miss is that the worst case time complexity bound of O(m * n) applies to methods like Regex::is_match, Regex::find and Regex::captures. It does not apply to Regex::find_iter or Regex::captures_iter. Namely, since iterating over all matches can execute many searches, and each search can scan the entire haystack, the worst case time complexity for iterators is O(m * n^2).

One example of where this occurs is when a pattern consists of an alternation, where an earlier branch of the alternation requires scanning the entire haystack only to discover that there is no match. It also requires a later branch of the alternation to have matched at the beginning of the search. For example, consider the pattern .*[^A-Z]|[A-Z] and the haystack AAAAA. The first search will scan to the end looking for matches of .*[^A-Z] even though a finite automata engine (as in this crate) knows that [A-Z] has already matched the first character of the haystack. This is due to the greedy nature of regex searching. That first search will report a match at the first A only after scanning to the end to discover that no other match exists. The next search then begins at the second A and the behavior repeats.

There is no way to avoid this. This means that if both patterns and haystacks are untrusted and you’re iterating over all matches, you’re susceptible to worst case quadratic time complexity. One possible way to mitigate this is to drop down to the lower level regex-automata crate and use its meta::Regex iterator APIs. There, you can configure the search to operate in “earliest” mode by passing a Input::new(haystack).earliest(true) to meta::Regex::find_iter (for example). By enabling this mode, you give up the normal greedy match semantics of regex searches and instead ask the regex engine to immediately stop as soon as a match has been found. Enabling this mode will thus restore the worst case O(m * n) time complexity bound, but at the cost of different semantics.

§Untrusted inputs in practice

While providing a O(m * n) worst case time bound on all searches goes a long way toward preventing ReDoS, that doesn’t mean every search you can possibly run will complete without burning CPU time. In general, there are a few ways for the m * n time bound to still bite you:

  • You are searching an exceptionally long haystack. No matter how you slice it, a longer haystack will take more time to search. This crate may often make very quick work of even long haystacks because of its literal optimizations, but those aren’t available for all regexes.
  • Unicode character classes can cause searches to be quite slow in some cases. This is especially true when they are combined with counted repetitions. While the regex size limit above will protect you from the most egregious cases, the default size limit still permits pretty big regexes that can execute more slowly than one might expect.
  • While routines like Regex::find and Regex::captures guarantee worst case O(m * n) search time, routines like Regex::find_iter and Regex::captures_iter actually have worst case O(m * n^2) search time. This is because find_iter runs many searches, and each search takes worst case O(m * n) time. Thus, iteration of all matches in a haystack has worst case O(m * n^2). A good example of a pattern that exhibits this is (?:A+){1000}| or even .*[^A-Z]|[A-Z].

In general, unstrusted haystacks are easier to stomach than untrusted patterns. Untrusted patterns give a lot more control to the caller to impact the performance of a search. In many cases, a regex search will actually execute in average case O(n) time (i.e., not dependent on the size of the regex), but this can’t be guaranteed in general. Therefore, permitting untrusted patterns means that your only line of defense is to put a limit on how big m (and perhaps also n) can be in O(m * n). n is limited by simply inspecting the length of the haystack while m is limited by both applying a limit to the length of the pattern and a limit on the compiled size of the regex via RegexBuilder::size_limit.

It bears repeating: if you’re accepting untrusted patterns, it would be a good idea to start with conservative limits on m and n, and then carefully increase them as needed.

§Crate features

By default, this crate tries pretty hard to make regex matching both as fast as possible and as correct as it can be. This means that there is a lot of code dedicated to performance, the handling of Unicode data and the Unicode data itself. Overall, this leads to more dependencies, larger binaries and longer compile times. This trade off may not be appropriate in all cases, and indeed, even when all Unicode and performance features are disabled, one is still left with a perfectly serviceable regex engine that will work well in many cases. (Note that code is not arbitrarily reducible, and for this reason, the regex-lite crate exists to provide an even more minimal experience by cutting out Unicode and performance, but still maintaining the linear search time bound.)

This crate exposes a number of features for controlling that trade off. Some of these features are strictly performance oriented, such that disabling them won’t result in a loss of functionality, but may result in worse performance. Other features, such as the ones controlling the presence or absence of Unicode data, can result in a loss of functionality. For example, if one disables the unicode-case feature (described below), then compiling the regex (?i)a will fail since Unicode case insensitivity is enabled by default. Instead, callers must use (?i-u)a to disable Unicode case folding. Stated differently, enabling or disabling any of the features below can only add or subtract from the total set of valid regular expressions. Enabling or disabling a feature will never modify the match semantics of a regular expression.

Most features below are enabled by default. Features that aren’t enabled by default are noted.

§Ecosystem features

  • std - When enabled, this will cause regex to use the standard library. In terms of APIs, std causes error types to implement the std::error::Error trait. Enabling std will also result in performance optimizations, including SIMD and faster synchronization primitives. Notably, disabling the std feature will result in the use of spin locks. To use a regex engine without std and without spin locks, you’ll need to drop down to the regex-automata crate.
  • logging - When enabled, the log crate is used to emit messages about regex compilation and search strategies. This is disabled by default. This is typically only useful to someone working on this crate’s internals, but might be useful if you’re doing some rabbit hole performance hacking. Or if you’re just interested in the kinds of decisions being made by the regex engine.

§Performance features

  • perf - Enables all performance related features except for perf-dfa-full. This feature is enabled by default is intended to cover all reasonable features that improve performance, even if more are added in the future.
  • perf-dfa - Enables the use of a lazy DFA for matching. The lazy DFA is used to compile portions of a regex to a very fast DFA on an as-needed basis. This can result in substantial speedups, usually by an order of magnitude on large haystacks. The lazy DFA does not bring in any new dependencies, but it can make compile times longer.
  • perf-dfa-full - Enables the use of a full DFA for matching. Full DFAs are problematic because they have worst case O(2^n) construction time. For this reason, when this feature is enabled, full DFAs are only used for very small regexes and a very small space bound is used during determinization to avoid the DFA from blowing up. This feature is not enabled by default, even as part of perf, because it results in fairly sizeable increases in binary size and compilation time. It can result in faster search times, but they tend to be more modest and limited to non-Unicode regexes.
  • perf-onepass - Enables the use of a one-pass DFA for extracting the positions of capture groups. This optimization applies to a subset of certain types of NFAs and represents the fastest engine in this crate for dealing with capture groups.
  • perf-backtrack - Enables the use of a bounded backtracking algorithm for extracting the positions of capture groups. This usually sits between the slowest engine (the PikeVM) and the fastest engine (one-pass DFA) for extracting capture groups. It’s used whenever the regex is not one-pass and is small enough.
  • perf-inline - Enables the use of aggressive inlining inside match routines. This reduces the overhead of each match. The aggressive inlining, however, increases compile times and binary size.
  • perf-literal - Enables the use of literal optimizations for speeding up matches. In some cases, literal optimizations can result in speedups of several orders of magnitude. Disabling this drops the aho-corasick and memchr dependencies.
  • perf-cache - This feature used to enable a faster internal cache at the cost of using additional dependencies, but this is no longer an option. A fast internal cache is now used unconditionally with no additional dependencies. This may change in the future.

§Unicode features

  • unicode - Enables all Unicode features. This feature is enabled by default, and will always cover all Unicode features, even if more are added in the future.
  • unicode-age - Provide the data for the Unicode Age property. This makes it possible to use classes like \p{Age:6.0} to refer to all codepoints first introduced in Unicode 6.0
  • unicode-bool - Provide the data for numerous Unicode boolean properties. The full list is not included here, but contains properties like Alphabetic, Emoji, Lowercase, Math, Uppercase and White_Space.
  • unicode-case - Provide the data for case insensitive matching using Unicode’s “simple loose matches” specification.
  • unicode-gencat - Provide the data for Unicode general categories. This includes, but is not limited to, Decimal_Number, Letter, Math_Symbol, Number and Punctuation.
  • unicode-perl - Provide the data for supporting the Unicode-aware Perl character classes, corresponding to \w, \s and \d. This is also necessary for using Unicode-aware word boundary assertions. Note that if this feature is disabled, the \s and \d character classes are still available if the unicode-bool and unicode-gencat features are enabled, respectively.
  • unicode-script - Provide the data for Unicode scripts and script extensions. This includes, but is not limited to, Arabic, Cyrillic, Hebrew, Latin and Thai.
  • unicode-segment - Provide the data necessary to provide the properties used to implement the Unicode text segmentation algorithms. This enables using classes like \p{gcb=Extend}, \p{wb=Katakana} and \p{sb=ATerm}.

§Other crates

This crate has two required dependencies and several optional dependencies. This section briefly describes them with the goal of raising awareness of how different components of this crate may be used independently.

It is somewhat unusual for a regex engine to have dependencies, as most regex libraries are self contained units with no dependencies other than a particular environment’s standard library. Indeed, for other similarly optimized regex engines, most or all of the code in the dependencies of this crate would normally just be unseparable or coupled parts of the crate itself. But since Rust and its tooling ecosystem make the use of dependencies so easy, it made sense to spend some effort de-coupling parts of this crate and making them independently useful.

We only briefly describe each crate here.

  • regex-lite is not a dependency of regex, but rather, a standalone zero-dependency simpler version of regex that prioritizes compile times and binary size. In exchange, it eschews Unicode support and performance. Its match semantics are as identical as possible to the regex crate, and for the things it supports, its APIs are identical to the APIs in this crate. In other words, for a lot of use cases, it is a drop-in replacement.
  • regex-syntax provides a regular expression parser via Ast and Hir types. It also provides routines for extracting literals from a pattern. Folks can use this crate to do analysis, or even to build their own regex engine without having to worry about writing a parser.
  • regex-automata provides the regex engines themselves. One of the downsides of finite automata based regex engines is that they often need multiple internal engines in order to have similar or better performance than an unbounded backtracking engine in practice. regex-automata in particular provides public APIs for a PikeVM, a bounded backtracker, a one-pass DFA, a lazy DFA, a fully compiled DFA and a meta regex engine that combines all them together. It also has native multi-pattern support and provides a way to compile and serialize full DFAs such that they can be loaded and searched in a no-std no-alloc environment. regex-automata itself doesn’t even have a required dependency on regex-syntax!
  • memchr provides low level SIMD vectorized routines for quickly finding the location of single bytes or even substrings in a haystack. In other words, it provides fast memchr and memmem routines. These are used by this crate in literal optimizations.
  • aho-corasick provides multi-substring search. It also provides SIMD vectorized routines in the case where the number of substrings to search for is relatively small. The regex crate also uses this for literal optimizations.


  • Search for regex matches in &[u8] haystacks.


  • A low level representation of the byte offsets of each capture group.
  • An iterator over all non-overlapping capture matches in a haystack.
  • An iterator over the names of all capture groups in a regex.
  • Represents the capture groups for a single match.
  • Represents a single match of a regex in a haystack.
  • An iterator over all non-overlapping matches in a haystack.
  • A helper type for forcing literal string replacement.
  • A compiled regular expression for searching Unicode haystacks.
  • A configurable builder for a Regex.
  • Match multiple, possibly overlapping, regexes in a single search.
  • A configurable builder for a RegexSet.
  • A by-reference adaptor for a Replacer.
  • A set of matches returned by a regex set.
  • An owned iterator over the set of matches from a regex set.
  • A borrowed iterator over the set of matches from a regex set.
  • An iterator over all substrings delimited by a regex match.
  • An iterator over at most N substrings delimited by a regex match.
  • An iterator over all group matches in a Captures value.


  • An error that occurred during parsing or compiling a regular expression.


  • A trait for types that can be used to replace matches in a haystack.


  • Escapes all regular expression meta characters in pattern.