`Regex::is_match` unexpectedly returning false with regex-syntax v0.8.0

[edmorley]

Hi 😄

First, thank you for a fantastic crate!

What version of regex are you using?

$ cargo tree
testcasev0.1.0 (/Users/emorley/src/testcase)
└── regex v1.10.0
    ├── aho-corasick v1.1.2
    │   └── memchr v2.6.4
    ├── memchr v2.6.4
    ├── regex-automata v0.4.1
    │   ├── aho-corasick v1.1.2 (*)
    │   ├── memchr v2.6.4
    │   └── regex-syntax v0.8.0
    └── regex-syntax v0.8.0

Describe the bug at a high level.

Using regex-syntax v0.8.0, a Regex::is_match that previously matched in regex-syntax v0.7.4 no longer matches.

Whilst the issue no longer reproduces as part of the regex-syntax v0.8.1 release (I would guess due to the revert in that release: #1102), I thought you might still wish to see the testcase in case it covers something that the newly added fuzz tests do not, ready for the future relanding of #1051.

What are the steps to reproduce the behavior?

1. cargo new testcase && cd $_
2. cargo add regex@1.10.0
3. cargo update -p regex-syntax --precise 0.8.0
4. Add this to the end of main.rs:

   #[cfg(test)]
   mod tests {
       #[test]
       fn testcase() {
           assert!(regex::Regex::new(r"^[[:alnum:]./-]+$")
               .unwrap()
               .is_match("a-b"));
       }
   }

5. cargo test
6. cargo update -p regex-syntax --precise 0.8.1
7. cargo test

What is the actual behavior?

• cargo test at step 5 (ie: when using regex-syntax v0.8.0) fails with the output below
• cargo test at step 7 (ie: when using regex-syntax v0.8.1) passes

---- tests::testcase stdout ----
thread 'tests::testcase' panicked at src/main.rs:9:9:
assertion failed: regex::Regex::new(r\"^[[:alnum:]./-]+$\").unwrap().is_match(\"a-b\")

What is the expected behavior?

cargo test passes regardless of regex-syntax version.

Notes

• I'm using Rust 1.74.0-beta.1 on macOS on ARM
• If I replace the [:alnum:] in the regex with eg a-z then the issue stops reproducing.

[BurntSushi]

Thanks! I'll add this as a test case. There are probably a lot of them unfortunately. I clearly had too much confidence in the test suite. The issue is in the HIR:

$ regex-cli-interval-opts debug hir '^[[:alnum:]./-]+$'
    parse time:  16.109µs
translate time:  8.883µs

Concat(
    [
        Look(
            Start,
        ),
        Repetition(
            Repetition {
                min: 1,
                max: None,
                greedy: true,
                sub: Class(
                    {
                        '.'..='.',
                        '-'..='9',
                        'A'..='Z',
                        'a'..='z',
                    },
                ),
            },
        ),
        Look(
            End,
        ),
    ],
)

$ regex-cli debug hir '^[[:alnum:]./-]+$'
    parse time:  14.264µs
translate time:  7.067µs

Concat(
    [
        Look(
            Start,
        ),
        Repetition(
            Repetition {
                min: 1,
                max: None,
                greedy: true,
                sub: Class(
                    {
                        '-'..='9',
                        'A'..='Z',
                        'a'..='z',
                    },
                ),
            },
        ),
        Look(
            End,
        ),
    ],
)

The first command is with regex-syntax 0.8.0 while the latter is 0.8.1. You can see that the ranges are out of order. For example, . should just be part of the '-'..='9' range.

[edmorley]

Thank you for taking a look and also explaining the issue :-)