iiinova — where your digital self keeps evolving

Here are the slides for my talk "Running Native Extensions and Ruby Code in Parallel Threads" at #EuRuKo 2026: eregon.me/blog/assets/...

074

TruffleRuby runs Doom at, 🥁, 662 FPS! From the talk of @khasinski.bsky.social at #EuRuKo

1122

"A language that is kind to humans is kind to AI too" Matz at #EuRuKo

050

So Matz is saying you can make Campfire (a Rails app) 5 times faster and use a twelvth of the memory by using github.com/rubys/roundh... + github.com/matz/spinel

1122

So part of the reason for Spinel (AOT compiler from Ruby to C) is lower memory usage, because RAM got expensive, and of course performance #EuRuKo

041

Benoit Daloze @eregon.me · 02/09/2026

"When I created Ruby 33 years ago" Matz at #EuRuKo

000

> It’s been 8 years since the first just-in-time compiler was introduced in Ruby. > For the first time, the garbage collector is no longer a black box to a Ruby JIT compiler. It would be good to be accurate here and replace Ruby by CRuby, e.g. TruffleRuby was already JITing allocations 10 years ago

040

Benoit Daloze @eregon.me · 31/08/2026

Any reason to not take the train for VIE->Brno? Because of the one change at Wien Hbf?

100

Reposted by Benoit Daloze

GraalVM @graalvm.org · 25/08/2026

GraalVM 25.3 is here!🎉 It brings performance and security improvements, along with updated language runtimes 🚀 Learn more: medium.com/graalvm/graa... Download: www.graalvm.org/downloads/

medium.com

GraalVM 25.3 is here! 🎉

Today, we’re pleased to announce the release of GraalVM 25.3!

0216

Benoit Daloze @eregon.me · 18/08/2026

The idea is that the native stack is gone since we join the Thread (or at least it won't be scanned by the GC) and the registers will reflect the current Thread. `Fiber.new {}.resume` might work too.

000

Benoit Daloze @eregon.me · 18/08/2026

For a while I searched a way to reliably GC an object on CRuby, this is hard given CRuby uses a conservative GC and might keep it alive via the native stack or some register. This seems reliable: m = ObjectSpace::WeakMap.new Thread.new { m[Object.new] = Object.new }.join GC.start p m.keys # => []

Christian Humer @chumer.dev · 13/08/2026

Reposted by Benoit Daloze

This is how we made bytecode interpreters in Java competitive with state of the art assembly tuned C interpreters. They partially evaluate like Truffle interpreters at the same time! Integrated in GraalPy, GraalWasm and Crema! Soon in GraalJS! Great work by Yudi Zheng!

152

Reposted by Benoit Daloze

GraalVM @graalvm.org · 13/08/2026

Escaping the Giant Switch: a deep dive into optimizing bytecode interpreters: medium.com/graalvm/esca...

medium.com

Escaping the Giant Switch

a true story about bytecode interpreters

0143

Reposted by Benoit Daloze

Jean Boussier @byroot.bsky.social · 26/07/2026

While it has been 18 months since I blogged about it, the json gem performance has improved quite significantly since, so now is a good time to add a few more parts to my JSON serie: byroot.github.io/ruby/json/20...

byroot.github.io

Optimizing Ruby’s JSON, Part 8

It has now been about 18 months since I concluded my post series on optimising Ruby’s json.

0233

Reposted by Benoit Daloze

The Verge @theverge.com · 12/06/2026

While his mountain of horrible personal conduct could fill multiple books, one fact in particular stands out: A year ago, Musk’s actions directly led to the deaths of hundreds of thousands of people. He did it knowingly. And, worse — gleefully. Read more: www.theverge.com/tech/949259/...

Verge headline: The world's first trillionaire is a killer
by TC Sottek

Image of Elon Musk doing a "fascist salute" in the form of an oil painting on a wall

28589643758

Reposted by Benoit Daloze

Jean Boussier @byroot.bsky.social · 05/06/2026

I realized it has been ~1.5 years since I concluded my series of post on JSON optimization. So I got curious how much faster it became since then. 2.7.2 was the last version before I took over maintainership. 2.9.0 was the version I released at the end of the blog series.

3498

Benoit Daloze @eregon.me · 01/06/2026

You know it's going well when typing 'github' in my browser has www.githubstatus.com as the first suggestion ...

020

Reposted by Benoit Daloze

Max @bernsteinbear.com · 27/05/2026

New linear scan register allocator in ZJIT by @tenderlove.dev ! railsatscale.com/2026-05-27-a...

072

Benoit Daloze @eregon.me · 14/05/2026

What if Ruby 4.0 features could be implemented in pure Ruby? With TruffleRuby, that’s possible: 73% of the core library is written in Ruby itself. We currently have 10 issues tagged [easy, pure Ruby] waiting for contributors: github.com/truffleruby/...

github.com

Ruby 4.0 support · Issue #4231 · truffleruby/truffleruby

Any help is appreciated whether it is: implementing one of the Ruby 4.0 changes in TruffleRuby, and ensuring it is tested in specs or MRI tests and otherwise add specs. If adding specs please leave...

042

Benoit Daloze @eregon.me · 08/05/2026

I installed the Codex extension in VSCode, the result is it corrupts (!) my git repository when using VSCode as the git editor (EDITOR='code -w'). Filed github.com/openai/codex..., was a pain to figure it out.

010

Benoit Daloze @eregon.me · 05/05/2026

It seems to be macOS-only or maybe clang-only for now.

000

Benoit Daloze @eregon.me · 05/05/2026

More details in github.com/DataDog/dd-t... and how it can cause CI to fail in a very subtle way on 3.3.11 but not on 3.3.10 or any other version!

Benoit Daloze @eregon.me · 05/05/2026

New take about Ruby C extensions: append_cflags "-Wall" or "-Wextra" with -Werror is a bug waiting to explode as it overrides -Wno-$category flags from RbConfig by being added at the end. It's unnecessary too because for Ruby 1.9+ RbConfig::CONFIG["warnflags"] starts with "-Wall -Wextra"

We'd need to statically link libssl and libyaml, set the CA certs path at runtime, support old Linux and macOS, Linux can be done in Docker but macOS requires older machines unless we're OK with macOS 15+ only (the minimum GitHub provides). And look into extra steps in github.com/jdx/ruby + testing

000

Probably yes, see github.com/ruby/setup-r... for the issues with --enable-load-relative that I found back then. Maybe such builds should be just at github.com/ruby/ruby-bu... though, I've been thinking about it, but also it's more complicated than just using --enable-load-relative:

100

Nice post and great to see you blog about Ruby! One thing I'm missing in the blog post is what are the new optimizations in Mustermann 4 which make the performance much better? Maybe something like merging routes Regexp more often or so?

000

Konstantin Haase @rkh.cool · 27/04/2026

Not only @github.com search is broken for PRs for 24+ hours but also just the list of PRs misses some PRs entirely. I have to resort to email to even find my own PRs now.

020

Reposted by Benoit Daloze

Mustermann 4.0 released! sinatrarb.com/2026/04/27/m...

Performance comparison of various Rack-based routers.

Performance comparison of Sinatra with Mustermann 3.1 vs Sinatra with Mustermann 4.0

472

Yeah it's complicated (e.g. libssl and CA certs) though possible because TruffleRuby always ships precompiled and there is no known issue. CRuby has --enable-load-relative but that's not so well tested and for old Ruby versions even `gem i rake` would fail.

100

A good time to switch (pun intended) is when setting up a new machine

100

Benoit Daloze @eregon.me · 27/04/2026

I wish but it sounds unlikely, changing Ruby version Switcher is rather cumbersome (need to reinstall Rubies, learn the new tool, etc) so many people just keep using what they are used to. On a similar note I have a chruby shell function using rbenv/mise because muscle memory and it's shorter.

100

Maybe I should do another blog like eregon.me/blog/2021/06... but the TLDR is: use mise, the other Ruby Switchers don't come close (chruby sets GEM_HOME which causes various issues). In terms of Ruby installers, ruby-build is still the best and is used by both mise and rbenv.