LinkTed/proc-macro-regex

A proc macro regex library to match an arbitrary string or byte array to a regular expression.

proc-macro-regex

A proc macro regex library to match an arbitrary string or byte array to a regular expression.

Usage

Add this to your Cargo.toml:

[dependencies]
proc-macro-regex = "~1.1.0"

Example

The macro regex! creates a function of the given name which takes a string or byte array and returns true if the argument matches the regex, otherwise false.

use proc_macro_regex::regex;

/// Create the function with the signature:
/// fn regex_email(s: &str) -> bool; 
regex!(regex_email "^[a-zA-Z0-9_.+-]+@[a-zA-Z0-9-]+\\.[a-zA-Z0-9-.]+$");

fn main () {
   println!("Returns true  == {}", regex_email("example@example.org"));
   println!("Returns false == {}", regex_email("example.example.org"));
}

The given regex works the same as in the regex crate. If the ^ is at the beginning of the regex and $ at the end then the whole string is checked, otherwise is check if the string contains the regex.

How it works

The macro creates a deterministic finite automaton (DFA), which parse the given input. Depending on the size of the DFA or the character of the regex, a lookup table or a code base implementation (binary search) is generated. If the size of the lookup table would be bigger than 65536 bytes (can be changed) then a code base implementation (binary search) is used. Additionally, if the regex contains any Unicode (no ASCII) character then a code base implementation (binary search) is used, too.

The following macro generates the following code:

regex!(example_1 "abc");

Generates:

fn example_1(s: &str) -> bool {
    static TABLE: [[u8; 256]; 3usize] = [ ... ];
    let mut state = 0;
    for c in s.bytes() {
        state = TABLE[state as usize][c as usize];
        if state == u8::MAX {
            return true;
        }
    }
    false
}

To tell the macro that the lookup table is not allowed to be bigger than 256 bytes, a third argument can be given. Therefore, a code base implementation (binary search) of the DFA is generated.

regex!(example_2 "abc" 256);

Generates:

fn example_2(s: &str) -> bool {
    let mut state = 0;
    for c in s.bytes() {
        state = if state < 1usize {
            match c {
                97u8 => 1usize,
                _ => 0usize,
            }
        } else {
            if state == 1usize {
                match c {
                    97u8 => 1usize,
                    98u8 => 2usize,
                    _ => 0usize,
                }
            } else {
                match c {
                    97u8 => 1usize,
                    99u8 => return true,
                    _ => 0usize,
                }
            }
        };
    }
    false
}

To change the visibility of the function, add the keywords at the beginning of the arguments.

regex!(pub example_2 "abc" 256);

Generates:

pub fn example_3(s: &str) -> bool {
   // same as in example_1 (see above)
}

To parse a byte array instead of string, pass a byte string.

regex!(example_4 b"abc");

Generates:

fn example_4(s: &[u8]) -> bool {
   // same as in example_1 (see above)
}

The generated code should work with #![no_std], too.

proc-macro-regex vs regex

Advantages:

• Compile-time (no runtime initialization, no lazy-static)
• Generated code that does not contain any dependencies
• No heap allocation
• Approximately 12%-68% faster for no trivia regex ¹

Disadvantages:

• Currently, no group captures
• No runtime regex generation

Performance

This is the performance comparison between this crate and the regex crate. If you want to test it by yourself, run cargo bench --bench compare.

Name	proc-macro-regex	regex	Result
E-Mail	743.95 MiB/s	441.67 MiB/s	68.44 %
URL	584.62 MiB/s	519.00 MiB/s	12.64 %
IPv6	746.92 MiB/s	473.38 MiB/s	57.78 %

This was compiled with rustc 1.53.0-nightly (392ba2ba1 2021-04-17).

License

This project is licensed under the BSD-3-Clause license.

Contribution

Any contribution intentionally submitted for inclusion in proc-macro-regex by you, shall be licensed as BSD-3-Clause, without any additional terms or conditions.

Footnotes

1. It were tested with regex in benches/compare.rs. For pattern/word matching it is slower because the regex library uses aho-corasick. (See Performance) ↩