Fast(er) regular expression engines in Ruby

https://serpapi.com/blog/faster-regular-expression-engines-in-ruby/

60•davidsojevic•1y ago

Comments

yxhuvud•1y ago

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

gitroom•1y ago

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

kayodelycaon•1y ago

> 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•1y 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•1y 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•1y 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.

GLM 5.2 beats Claude in our benchmarks

Age verification is just a precursor to automated attribution of speech

Historical memory prices 1960-2026

Better Images of AI

5k menus from the New York Public Library’s Buttolph Collection (1880-1920)

Knowledge Distillation of Black-Box Large Language Models (2024)

I used Claude Code to get a second opinion on my MRI

Tell Congress: Don't Force Age Checks Online

Deciphering Basmala

AI boom risks global financial crash, warn central bankers

TOP500 at ISC’26: We have a New Number 1 Supercomputer

The Boeing 747 begins its final descent

Show HN: Zanagrams

Professor denounces mass AI fraud on an exam at Brown

Librepods: AirPods liberated

Working around dragons with the Lemote Yeeloong laptop and OpenBSD

Daisugi, the Japanese technique of growing trees out of other trees (2020)

The Baffling World of Masayoshi Son's Presentations (2020)

Idler Magazine

Show HN: DRM-Free Books

Researchers have developed pixels that can emit and analyse light together

Tokenmaxxing is dead, long live tokenmaxxing

The KIDS Act would require age checks to get online

A way to exclude sensitive files issue still open for OpenAI Codex

Examining circuit boards from the Space Shuttle's I/O Processor

The curious case of the disappearing Polish S (2015)

Show HN: Bash4LLM+ – A lightweight, dependency-free Bash wrapper for LLM APIs

Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch

Model Training as Code

The MUMPS 76 Primer – anniversary edition