RegExp unicode mode differs from Bun runtime

[kevgeoleo]

Version: Deno 2.2.5

Hi,

I would like to report an issue I encountered in Deno while I was running the following code snippet:

console.log(new RegExp("\\ud800\udc00+", "u").exec("\u{10000}\u{10000}"));

Deno gives null as output whereas Bun gives [ "𐀀", index: 0, input: "𐀀𐀀", groups: undefined ]

Regards,
Kevin

[dsherret]

I think Deno is following the spec here and Bun is not. For example, this test262 passes in Deno, but fails in Bun: https://github.com/tc39/test262/blob/ce7e72d2107f99d165f4259571f10aa75753d997/test/staging/sm/RegExp/unicode-raw.js#L56 -- I'm not super familiar with the spec or the unicode modifier so can't say for sure though.

Chrome, Firefox, Node and Deno all seem to output null in this case.

[0f-0b]

The string "\\ud800\udc00+" consists of, in order, a \u escape of a high surrogate, a literal low surrogate, and a repetition operator +. In Unicode mode, a \u escape followed by a literal character do not form a surrogate pair, and an unpaired surrogate in the source string matches an unpaired surrogate in the input string, so this regex matches an unpaired high surrogate followed by one or more unpaired low surrogates, which is impossible. That's why exec on this regex always returns null.

The bug lies in JSC's JIT compiler. If you set the environment variable JSC_useRegExpJIT to 0 when running Bun, the bug would not happen.

[dsherret]

@0f-0b thanks for explaining. Going to close.