0.1.47 changed matching of `\r`

[DanielKeep]

Given the following program:

extern crate regex;
fn main() {
    let re = regex::Regex::new(r"^(.*?)(\n|\r\n?|$)").unwrap();
    println!("{:?}", re.find("ab\rcd"));
}

and rustc nightly 2bd875d3d4a9f06a4a18325d6e4a68b5234a09f6 (2016-01-19), version 0.1.46 of regex outputs:

Some((0, 3))

Version 0.1.47 outputs:

Some((0, 5))

I'm not sure what's going on here, but my understanding is that the new interpretation is wrong.

[DanielKeep]

Also, I just tested r"^(.*?)(\n|\r\n|\r|$)", and it produces Some((0, 3)) on both 0.1.46 and 0.1.47.

[BurntSushi]

This is indeed a bug! I'm shocked this wasn't caught in the test suite.

I've got a smaller example: ab?|$ on text az should return (0, 1) (and does on 0.1.46), but returns Some((2, 2)) on 0.1.47. Specifically, the compilation on 0.1.46 is:

000 Save(0)
001 Split(2, 6)
002 'a'
003 Split(4, 5)
004 'b'
005 Jump(7)
006 EndText
007 Save(1)
008 Match

And the compilation on 0.1.47 is:

000 Save(0)
001 Split(2, 5)
002 'a'
003 Split(4, 5)
004 'b' (goto: 6)
005 EndText
006 Save(1)
007 Match

The second compilation is wrong. The Split(4, 5) instruction should be Split(4, 6). Splitting to 5 means that either a b or a $ is required after matching an a, but of course, an a is sufficient for a match. This means the alternation fails until the end of the input, at which point $ matches.

In the case of your regex, this causes the alternation to fail and the preceding .*? ends up matching everything.

Your workaround fixed things because it removed the inner alternation which is provoking this bug.

Working on a fix!

[BurntSushi]

Fixed in 0.1.48 on crates.io. Thanks for the bug report @DanielKeep!

[DanielKeep]

Don't mention it; thanks for regex. :)