Stefan Marr

I abused my position of power today to teach students that dynamically-typed languages don't generally do type checks at run time.

I'm so excited for VMIL! The program looks great! conf.researchr.org/home/splash-...

Brief weekend reading on another benchmarking pitfall: The Limitations of Running a Workload Generator In the Same JVM as the System-Under-Test norlinder.nu/posts/The-Li... Spoiler: the GC may conspire against you and your data collection!

Brief weekend reading on another benchmarking pitfall: The Limitations of Running a Workload Generator In the Same JVM as the System-Under-Test norlinder.nu/posts/The-Li... Spoiler: the GC may conspire against you and your data collection!

They tell me to stay on top of this technology. Alright, but... "my research subagent went beyond its assigned scope, edited the file and made git commits itself, despite being instructed to only research and report."... it's broken by design. Why'd you give it the capability?

It was amusing to read that the current "AI makes lectures irrelevant" debate has echos from 1781 as related by Boswell.

I thought, I had broken the iteration-time foldout of ReBenchDB. Turns out, it was just terribly slow. Scanning through a 2GB data table on a slow disk takes a while. Fix the primary key order, i.e., the relevant index, and instead of waiting 42sec, it shows in 34ms.

For those wondering, yk is chugging away: support is broadening and in the last couple of days I've merged things which speed up interpreters by at least 5% (some by quite a bit more). There is a lot more juice to be squeezed, of course, but every little helps!

We'll present a paper about ZJIT at VMIL'26! conf.researchr.org/details/spla...

VMIL'26 has a program! Looks very interesting. Static Analysis, optimizing Android's GC, understanding R's benchmarks, optimizing Ruby, and lots more! The SPLASH early registration deadline is September 3rd! Program: conf.researchr.org/home/splash-...

Benchmarking is hard, and there are many pitfalls out there. Let's look at some fictitious examples and think of what could have caused the behavior we are seeing: stefan-marr.de/2026/08/pitf...

Benchmarking is hard, and there are many pitfalls out there. Let's look at some fictitious examples and think of what could have caused the behavior we are seeing: stefan-marr.de/2026/08/pitf...

The video of my QCon talk from this March is now up www.infoq.com/presentation...

Are you working on language implementations and implementation research? Do you want to present your work to an enthusiastic crowed in Oakland? Submit your work to the VMIL Workshop! conf.researchr.org/home/splash-...

Are you working on language implementations and implementation research? Do you want to present your work to an enthusiastic crowed in Oakland? Submit your work to the VMIL Workshop! conf.researchr.org/home/splash-...

How come, we academics are so terrible at "knowledge transfer"? We can't even make sure that this year's organizers of an event know everything they need to know, and the organizers last year already had to learn. I guess, it's not a priority. 😩

Christoph shows how to reduce the size of @graalvm.org native images. We can get pretty far with fairly basic heuristics. Basic, means cheap during compilation. 2026.ecoop.org/details/mplr...

Thomas presenting his work on unifying GCs to be able to have different GCs with different tradeoffs in the same AOT compiled binary. 2026.ecoop.org/details/mplr...

Markus is using pedagogic comments in execution traces to enable students to look at debugging sessions with guiding comments. Learning how to program in the very first year, based on debuggers and visualization. 2026.ecoop.org/details/debt...

Can we use meta-compilation to ahead-of-time compile dynamic language programs from an interpreter? Maybe! Christoph talked about the approach he will try out. 2026.ecoop.org/details/ICOO...

Arrived an hour later than planned, but had to start an hour earlier...

Traveling across central Europe by train tomorrow. Theoretically, a 11h journey with two changes. What are the odds that I will arrive on time? 1. Very good 2. 50:50 3. Deutsche Bahn? 🤣

Lots of us from the SSW will be attending @ecoop-conference.bsky.social, come and talk to use at ICOOOLPS, MPLR, the Demo Track, DEBT, ECOOP Academy, and at the poster session! Christoph A. is presenting early ideas of how to make AOT meta-compilation work for dynamic languages. 1/...

Pleased to announce that the application deadline for the Fourth ACM Europe Summer School on Distributed and Replicated Environments (DARE 2026) has been extended to 19 June 2026! This extension is made possible thanks to additional funding from ACM and SIGSOFT. dare-milan.github.io

Might have won something. Instead of 10h with @deutschebahn.com, they want me to travel 20h with them... Enjoying 10 trains and busses, instead of 3... Got rebooked "automatically" it seems. Thanks AI!

Today, May 22, my book about ZGC is finally ready and available to order! I can recommend reading it if you want to learn more about the magic nature of ZGC. You can order it here (among many places): www.routledge.com/The-Z-Garbag...

REMINDER: Our call for a professorship in the programming language field ends this Sunday! Make sure your application is in the system on time.

Ugh. Past me underestimated how many experiments future me is going to run. ReBenchDB just overflowed 16bit ints on the number of trials (i.e., experiments per benchmark machine). Is this something to celebrate? The number of collected measurements already requires 24bits...

The program of ICOOOLPS is now online: 2026.ecoop.org/home/ICOOOLP... And today is the last day for early registrations. Interested in programming language implementation, compilation, and optimization? Join us in Brussels! Don't know what to expect? 1/2

⏳ 7 days left for early registration at #ECOOP2026! Don't miss your chance to hear Sandrine Blazy take us from mechanized semantics to verified compilation, and find out what 15 years of Viper looks like through Peter Müller's eyes. 📍 Brussels, BE | Jun 29–Jul 3 2026.ecoop.org/attending/re...

Towards Least-Privilege WebAssembly Applications: Transparent Interposition for WebAssembly Components Olav Blaak talked about his approach to fine-grained security policies for Wasm components. youtu.be/a3a_le54tW0 This one was another MoreVMs'26 talk.

Towards Least-Privilege WebAssembly Applications: Transparent Interposition for WebAssembly Components Olav Blaak talked about his approach to fine-grained security policies for Wasm components. youtu.be/a3a_le54tW0 This one was another MoreVMs'26 talk.

At this point, the expectations from the students to get recordings is zero. Maybe it's a good thing. But, I won't be defeated by technology! Wish me luck 🙈

After the uni's browser-based lecture recording system failed in 6 different ways (it worked last term), I am running out of ideas. I guess ffmpeg is my next best option.

Retaining the Renaissance: Lessons Learned For Maintainable Benchmark Suites Lubomír Bulej talk at MoreVMs'26 about keeping a benchmarking suite relevant, and what we may learn from the Renaissance project. youtu.be/SF3T0YHHZMc

Retaining the Renaissance: Lessons Learned For Maintainable Benchmark Suites Lubomír Bulej talk at MoreVMs'26 about keeping a benchmarking suite relevant, and what we may learn from the Renaissance project. youtu.be/SF3T0YHHZMc

Honored to give my talk at #RubyKaigi this year! Slides are available here: blog.peterzhu.ca/assets/rubyk...

Code Layout Optimization Guided by Temporal Profiling Information by Marko Spasic This MoreVMs'26 talk improves the start up of applications by making sure that the code that's needed next is close by. Especially relevant for larger binaries. youtu.be/57KtH-w37B0

Code Layout Optimization Guided by Temporal Profiling Information by Marko Spasic This MoreVMs'26 talk improves the start up of applications by making sure that the code that's needed next is close by. Especially relevant for larger binaries. youtu.be/57KtH-w37B0

The Promise of Static Profiling: Exploring the Limits by Milan Cugurovic Another MoreVMs'26 talk. This one is on using machine learning to develop compiler heuristics. youtu.be/-14jvsxXb3w