Fast(er) regular expression engines in Ruby

Fast(er) regular expression engines in Ruby(serpapi.com)

60 points by davidsojevic 1 year ago | 6 comments

> Another nuance was found in ruby, which cannot scan the haystack with invalid UTF-8 byte sequences.

This is extremely basic ruby: UTF-8 encoded strings must be valid UTF-8. This is not unique to ruby. If I recall correctly, python 3 does the same thing.

    2.7.1 :001 > haystack = "\xfc\xa1\xa1\xa1\xa1\xa1abc"
    2.7.1 :003 > haystack.force_encoding "ASCII-8BIT"
    => "\xFC\xA1\xA1\xA1\xA1\xA1abc" 
    2.7.1 :004 > haystack.scan(/.+/)
    => ["\xFC\xA1\xA1\xA1\xA1\xA1abc"]

This person is a senior engineer on their Team page. All they had to do was google "ArgumentError: invalid byte sequence in UTF-8". Or ask a coworker... the company has Ruby on Rails applications. headdesk

burntsushi 1 year ago | |

The nuance is specifically relevant here because neither of the other two regex engines benchmarked have this requirement. It's doubly relevant because that means running a regex search doesn't require a UTF-8 validation step, and is therefore likely beneficial from a perf perspective, dependening on the workload.

kayodelycaon 1 year ago | | |

That’s a good point. I hadn’t considered it because I’ve hit the validation error long before getting to search. It is possible to avoid string operations with careful coding prior to the search.

Edit: After a little testing, the strings can be read from and written to files without triggering validation. Presumably this applies to sockets as well.

DmitryOlshansky 1 year ago |

I wonder how std.regex of dlang would fare in such test. Sadly due to a tiny bit of D’s GC use it’s hard to provide as a library for other languages. If there is an interest I might take it through the tests.

yxhuvud 1 year ago |

Eww, pretending to support utf8 matchers while not supporting them at all was not pretty to see.

gitroom 1 year ago |

Honestly that part bugs me, fake support is worse than no support imo