Andreas Zeller
LightNews LightNews
banner
andreaszeller.bsky.social
Andreas Zeller
@andreaszeller.bsky.social
1.4K followers 150 following 100 posts
Software researcher at https://cispa.de, working on #Fandango, #S3, #FuzzingBook, #DebuggingBook. Testing, debugging, analyzing, and protecting software for a better world. Find me at https://andreas-zeller.info/
Posts Replies Media Videos
Pinned
andreaszeller.bsky.social
Andreas Zeller @andreaszeller.bsky.social · Nov 28
Happy and honored: I have been selected to become an IEEE Fellow for "contributions to analyzing software and its development process". Thanks a lot to all of you who endorsed and supported me – especially @cispa.de + @saarland-informatics-campus.de, who always encouraged such long-term impact.
cispa.de CISPA Helmholtz Center for Information Security @cispa.de · Nov 28
Congratulations! We are delighted that @andreaszeller.bsky.social will be elevated to IEEE Fellow on January. This is the highest grade of membership in the Institute of Electrical and Electronics Engineers. This recognition highlights once more the excellent research that is carried out at CISPA.
andreaszeller.bsky.social
How do execution features relate to failures? In this new ACM TOSEM paper, Marius Smytzek, Martin Eberlein, Lars Grunske, and I analyze which execution features beyond code coverage correlate best with failures and lead to accurate explanations of failure causes: dl.acm.org/doi/10.1145/...
Fault localization aims to identify code regions responsible for failures. Traditional techniques primarily correlate statement execution with failures; however, program behavior involves diverse execution features, including variable values, branch conditions, and definition-use pairs, which can provide richer diagnostic insights. This paper comprehensively investigates execution features for fault understanding, addressing two complementary goals. First, we conduct an empirical study of 310 bugs across 20 projects, analyzing 17 execution features and assessing their correlation with failure outcomes. Our findings suggest that fault localization benefits from a broader range of execution features: (1) Scalar pairs exhibit the strongest correlation with failures; (2) Beyond line executions, def-use pairs and functions executed are key indicators for fault localization; and (3) Combining multiple features enhances effectiveness compared to relying on individual features. Second, building on these insights, we introduce a debugging approach that learns relevant features from labeled test outcomes. The approach extracts fine-grained execution features and trains a decision tree to differentiate passing and failing runs. The trained model generates fault diagnoses that explain the underlying causes of failures. Our evaluation demonstrates that the generated diagnoses achieve high predictive accuracy. These interpretable diagnoses empower developers to debug software efficiently by providing deeper insights into failures.
February 3, 2026 at 10:10 AM
andreaszeller.bsky.social
* Mail has lost all my emails sent since Monday
* Mail search is broken too
* Search in reminders cannot find anything
* New Keynote is full of ads!?
* Invoke Python-3.13, get 3.14 instead - venvs are messed up
* LaTeX "minted" crashes (likely b/c Python)

So glad I'm an expert in debugging /sarcasm
January 30, 2026 at 10:23 AM
andreaszeller.bsky.social
Fuzzing software becomes much more effective if you can generate _valid_ inputs. We have now built the first approach to _statically_ extract complete and precise input grammars from parser code, producing syntactically valid and diverse inputs by construction. Enjoy! dl.acm.org/doi/10.1145/...
Inferring Input Grammars from Code with Symbolic Parsing | ACM Transactions on Software Engineering and Methodology
Generating effective test inputs for a software system requires that these inputs be valid, as they will otherwise be rejected without reaching actual functionality. In the absence of a specification ...
dl.acm.org
January 28, 2026 at 4:05 PM
andreaszeller.bsky.social
After a visit to Max Planck Institute for Security and Privacy (MPI-SP) in Bochum, seeing my awesome colleagues @thorstenholz.bsky.social, @mboehme.bsky.social, Mathias Payer, and many more, now on my way to Paris to celebrate ten years of @softwareheritage.org with the great Roberto Di Cosmo
January 27, 2026 at 5:52 PM
andreaszeller.bsky.social
A researcher used more than 2,000 em-dashes in his papers, revealing AI-based manipulation in 400+ papers since 1985. Professor Zeller claims he "typed" these dashes into the paper by using "two hyphens" and a "typesetting" system.
$ cd ~/Papers/ $ grep -e '[ ~]-- ' */*.tex | wc -l 2258 $
January 8, 2026 at 8:34 AM
andreaszeller.bsky.social
Happy New Year! I am thrilled to report that Jacek Śliwerski, Tom Zimmermann, and I won the ACM SIGSOFT 2026 Impact Award 🏆 for "When do changes induce fixes?" (MSR 2005). The paper introduced the popular SZZ algorithm for linking change histories and bug databases: dl.acm.org/doi/10.1145/...
When do changes induce fixes? | ACM SIGSOFT Software Engineering Notes
As a software system evolves, programmers make changes that sometimes cause problems. We analyze CVS archives for fix-inducing changes---changes that lead to problems, indicated by fixes. We show how ...
dl.acm.org
January 6, 2026 at 3:56 PM
andreaszeller.bsky.social
Problem: Reviewers did not read the paper.
Solution: Write a detailed rebuttal and point to all the places in the paper that answer their questions.
New problem: Reviewers did not read the rebuttal.
December 22, 2025 at 1:15 PM
andreaszeller.bsky.social
In an IPN vision talk last Monday, I sketched how future AI "super-coders" would learn from their own experiments with software to far surpass current LLM-based AI coders.

The talk is now online. Enjoy!

* Recording: www.youtube.com/watch?v=tBO_...
* Slides: andreas-zeller.info/assets/Shoul...
IPN Colloquium 15 12 2025 Andreas Zeller
YouTube video by IPN (ICT Research Platform Nederland)
www.youtube.com
December 19, 2025 at 8:24 AM
andreaszeller.bsky.social
Today at 16:00 CET, I'll give a vision talk "Should AI Coders Experiment More?", paving the way to AI “super coders” that may become way more competent than the most experienced programmers - and also way more competent than any LLM-based coders. Details here: ict-research.nl/2025/11/ipn-...
IPN Colloquium 6: Should AI Coders Experiment More? – ICT Research Platform Netherlands
ict-research.nl
December 15, 2025 at 9:25 AM
andreaszeller.bsky.social
Time to get serious again. New blog post "Reviewer-Author Collusion Rings and How to Fight Them": andreas-zeller.info/2025/12/07/R...
Reviewer-Author Collusion Rings and How to Fight Them
In 2012, I attended a physical meeting of the program committee responsible for selecting the best scientific papers for the ESEC/FSE 2013 conference in Saint Petersburg, Russia. This meeting was part...
andreas-zeller.info
December 8, 2025 at 4:31 PM
andreaszeller.bsky.social
Bonus material for The LaTeX Korrektor! Some of you asked: "What are these photos and posters in the background?" Here they come, enlarged and with some details. Enjoy! #LaTeX #LaTeXKorrektor

In case you missed it, watch all six episodes of the LaTeX Korrektor here: www.youtube.com/watch?v=EhsM...
A description of the items shown in the LaTeX Korrektor background: * Diomidis Spinellis
, Author of
 “Advice for writing LaTeX documents” * LaTeX 2ε 
Cheat Sheet * Helmut Kohl
, German Chancellor
 1982–1988 * A fictitious
 event poster "Lack Leder LaTeX, Hamburg" * A fictitious
 LaTeX 
propaganda 
poster
December 7, 2025 at 12:23 PM
andreaszeller.bsky.social
Series finale! The LaTeX Korrektor 6/6 - Ten Commandments www.youtube.com/shorts/HAodi... #LaTeX #LaTeXKorrektor

Read the LaTeX advice by Diomidis Spinellis (@coolsweng.bsky.social): github.com/dspinellis/l...

All six episodes of the LaTeX Korrektor: www.youtube.com/watch?v=EhsM...
The LaTeX Korrektor 6/6 - Ten Commandments
YouTube video by Andreas Zeller
www.youtube.com
December 6, 2025 at 8:15 AM
andreaszeller.bsky.social
Why, oh why does your bibliography have all titles in lowercase? WHY? The LaTeX Korrektor 5/6 - Citations: www.youtube.com/shorts/0nk72... #LaTeX #LaTeXKorrektor

Missed previous episodes? This playlist has them all: www.youtube.com/watch?v=EhsM...
The LaTeX Korrektor 5/6 - Citations
YouTube video by Andreas Zeller
www.youtube.com
December 5, 2025 at 8:08 AM
andreaszeller.bsky.social
Can one reject papers because of bad typography? The LaTeX Korrektor 4/6 - Math Mode: www.youtube.com/shorts/mc8ro... #LaTeX #LaTeXKorrektor

Missed previous episodes? This playlist has them all: www.youtube.com/watch?v=EhsM...
The LaTeX Korrektor 4/6 - Math Mode
YouTube video by Andreas Zeller
www.youtube.com
December 4, 2025 at 8:09 AM
andreaszeller.bsky.social
This Friday at 11:00 am, I will be giving a talk on "Language-Based Software Testing" at Imperial College in London, covering our latest and greatest in testing systems with highly complex inputs and interactions. Details here: srg.doc.ic.ac.uk/seminars/25-...
Language-Based Software Testing with José Antonio Zamudio Amaya, Marius Smytzek, Valentin Huber, Addison Crump, Alexi Turcotte, and many others Random test input generators (fuzzers) have become the prime detectors of vulnerabilities in software. While generic fuzzers easily adapt to arbitrary programs under test, they offer very little possibilities to control or shape the generated inputs. In this talk, I present FANDANGO, a novel language-based fuzzer that combines grammars with predicates over input elements to produce inputs that satisfy all the given predicates. Examples of what such predicates can express include input format constraints (“The length field should be equal to the length of the payload”) code features (“Any variable used must be declared beforehand”) statistical distributions (“Across all inputs, the voltage field must follow a Gaussian distribution, but never exceed 20 mV”) data collections (“The credit-card-number field should come from the Python faker library”) and more – actually, any property that can be expressed in a Python expression. In our experiments, FANDANGO efficiently solved complex file formats and satisfied demanding predicates, up to full-fledged programming languages as test inputs for compilers. This opens the door towards personalized fuzzing, where testers can make use of their own knowledge and LLM knowledge to very effectively fuzz systems. Includes live demos! Fandango is available at https://fandango-fuzzer.github.io/ Andreas Zeller is faculty at the CISPA Helmholtz Center for Information Security and professor for Software Engineering at Saarland University. His research on automated debugging, mining software archives, specification mining, and security testing has won several awards for its impact in academia and industry. Zeller is an ACM Fellow, holds an ACM SIGSOFT Outstanding Research Award, and has won two ERC Advanced Grants, Europe’s highest funding for individual researchers.
December 3, 2025 at 10:52 AM
andreaszeller.bsky.social
The LaTeX Korrektor 3/6 - Vertical Space: www.youtube.com/shorts/n9iLf... #LaTeX #LaTeXKorrektor
The LaTeX Korrektor 3/6 - Vertical Space
YouTube video by Andreas Zeller
www.youtube.com
December 3, 2025 at 8:26 AM
andreaszeller.bsky.social
The LaTeX Korrektor 2/6 - Dashes: www.youtube.com/shorts/oDDbg...
The LaTeX Korrektor 2/6 - Dashes
YouTube video by Andreas Zeller
www.youtube.com
December 2, 2025 at 8:04 AM
andreaszeller.bsky.social
For your enjoyment: "The LaTeX Korrektor" www.youtube.com/shorts/EhsMd... #LaTeX #LaTeXKorrektor
The LaTeX Korrektor 1/6 - Quotes
YouTube video by Andreas Zeller
www.youtube.com
December 1, 2025 at 8:08 AM
andreaszeller.bsky.social
Happy and honored: I have been selected to become an IEEE Fellow for "contributions to analyzing software and its development process". Thanks a lot to all of you who endorsed and supported me – especially @cispa.de + @saarland-informatics-campus.de, who always encouraged such long-term impact.
cispa.de CISPA Helmholtz Center for Information Security @cispa.de · Nov 28
Congratulations! We are delighted that @andreaszeller.bsky.social will be elevated to IEEE Fellow on January. This is the highest grade of membership in the Institute of Electrical and Electronics Engineers. This recognition highlights once more the excellent research that is carried out at CISPA.
November 28, 2025 at 1:13 PM
andreaszeller.bsky.social
Today 11:20 at #ASE2025/#AgenticSE, we have Kuangxiangzi Liu show how to automatically turn natural language protocol specs into #Fandango formal specs, ready for massive automatic test generation: conf.researchr.org/details/ase-...
Automated testing is essential for the continuous delivery of safety- and security-critical systems, yet the governing requirements are typically written in unstructured natural language. An alternative is to use a large language model (LLM) to interpret the requirements and extract relevant tests automatically. Nevertheless, an LLM, being a powerful tool, does not inherently understand complex logic disguised inside a sentence written by a human using a natural language. It cannot derive a state machine of a system under test (SUT) by simply reading requirements. Furthermore, its output is based on recurring patterns that it has learned from a colossal amount of information, not necessarily always relevant or effective. In this work, we have explored a strategic alternative to the problem, which is to use an LLM that can translate the natural-language requirements into a formal, executable specification and to let automation operate on it. The resulting artifacts will eliminate the random or non-specific behavior of the LLM and add determinism to automation—enhancing traceability, verifiability, and accountability, as they can be checked against real implementations. We demonstrate this approach for protocol testing with Internet RFCs, which provide relatively precise, publicly vetted protocol specifications. We used I/O Grammars, a formalism that captures multi-party interactions along with message syntax, constraints, and states. Our LLM-based pipeline can both synthesize grammars from RFCs and repair them while preserving sentence-level traceability, thereby improving grammar quality and fuzzing efficacy. We evaluated our approach on five widely used network protocols (SMTP, POP3, IMAP, FTP, and TFTP) and observed that the auto-generated grammars achieve an average of 85% state coverage and up to 89% transition coverage across heterogeneous SUTs.
November 20, 2025 at 12:10 AM
andreaszeller.bsky.social
And another student of mine, Bernd Gruner, presenting continuous data flow-based threat modeling with FlowStrider at @aseconf.bsky.social
November 17, 2025 at 6:39 AM
andreaszeller.bsky.social
When you generate test inputs, how can you tell whether they pass or fail? Our BASHIRI tool learns such oracles with 95% accuracy - demo today at #ASE2025 @aseconf.bsky.social: conf.researchr.org/details/ase-...
November 17, 2025 at 5:10 AM
andreaszeller.bsky.social
Today at ASE 2025 @aseconf.bsky.social: “The fault in our stats”, or how the majority of notebooks fails to check even the most basic statistical assumptions
The Fault in our Stats i.e., exploring the degree to which data analysts deviate from statistics best practice, and an automated approach to detect such deviations
November 17, 2025 at 12:36 AM
andreaszeller.bsky.social
On my way to Seoul, South Korea, visiting #ASE2025 and then continuing to Shenzhen and Hong Kong. See you soon!
November 11, 2025 at 1:25 PM
Reposted by Andreas Zeller
gotocon.com
We teach coding.
We barely teach testing.
We ignore debugging.

Prof. @andreaszeller.bsky.social joins Clare Sudbery on the #GOTOpodcast to discuss why debugging is the missing piece in programming education — and why it matters more than ever in the age of AI.

🎧 gotopia.tech/podcast
November 7, 2025 at 1:00 PM