red-green.png · 71 bytes · 2 × 1 pixels
91 programs read this file.
Parsarium is an open testbed of real parsers. It collects 938 verified open-source package paths across 14 formats and eight languages. With standard APIs included, 986 paths have pinned Docker harnesses: you can run one file through them, replay a corpus, or fuzz them with AFL++.
- signature
- IHDR · size and colour type
- IDAT · compressed pixels
- IEND · end of image
- Ccairo 1.18.6width 2 · height 1
- Cgdk-pixbuf 2.44.8width 2 · height 1
- Cimagemagick 7.1.2-35width 2 · height 1
- Cimlib2 1.12.7width 2 · height 1
- Clodepng cci.20260102width 2 · height 1
- Craylib 6.2width 2 · height 1
- Csdl3 3.4.20width 2 · height 1
- Cstb cci.20261021width 2 · height 1
- Cvips 8.18.7width 2 · height 1
- Cwuffs 0.3.6width 2 · height 1
- C++cairo 1.18.6width 2 · height 1
- C++gdk-pixbuf 2.44.8width 2 · height 1
- C++imagemagick 7.1.2-35width 2 · height 1
- C++imlib2 1.12.7width 2 · height 1
- C++opencv 5.1.0width 2 · height 1
- C++raylib 6.2width 2 · height 1
- C++sdl3 3.4.20width 2 · height 1
- C++sfml 3.0.2width 2 · height 1
- C++stb cci.20261021width 2 · height 1
- C++vips 8.18.7width 2 · height 1
- Gogithub.com/andrianbdn/iospng v0.0.0-20180730113000-dccef1992541width 2 · height 1
- Gogithub.com/anthonynsimon/bild/imgio v0.17.1width 2 · height 1
- Gogithub.com/davidbyttow/govips/v2/vips v2.20.0width 2 · height 1
- Gogithub.com/disintegration/imaging v1.6.2width 2 · height 1
- Gogithub.com/dsoprea/go-png-image-structure/v2 v2.0.0-20210512210324-29b889a6093dwidth 2 · height 1
- Gogithub.com/fogleman/gg v1.3.0width 2 · height 1
- Gogithub.com/gameparrot/fastpng v0.0.0-20250305185850-d72e123a2123width 2 · height 1
- Gogithub.com/go-images/png v0.2.0width 2 · height 1
- Gogithub.com/h2non/bimg v1.1.9width 2 · height 1
- Gogopkg.in/gographics/imagick.v3/imagick v3.7.3width 2 · height 1
- Goimage.Decode(io.Reader) with image/png registered 1.27.1width 2 · height 1
- Goimage/png.Decode(io.Reader) 1.27.1width 2 · height 1
- Goimage/png.DecodeConfig(io.Reader) 1.27.1width 2 · height 1
- Javacommons-imaging 1.0.0-alpha6width 2 · height 1
- Javaij 1.54pwidth 2 · height 1
- Javaitextpdf 5.5.13width 2 · height 1
- Javajcodec 0.2.5width 2 · height 1
- Javajcodec-javase 0.2.5width 2 · height 1
- Javajdk-imageicon 21.0.2width 2 · height 1
- Javajdk-imageio 21.0.2width 2 · height 1
- Javajdk-imagereader 21.0.2width 2 · height 1
- Javajdk-toolkit 21.0.2width 2 · height 1
- Javaopenpdf 3.0.5width 2 · height 1
- Javaopenpnp-opencv 4.9.0-0width 2 · height 1
- Javapdfbox 3.0.8width 2 · height 1
- Javapngj 2.1.0width 2 · height 1
- Javathumbnailator 0.4.21width 2 · height 1
- JavaScript@napi-rs/canvas 1.1.0width 2 · height 1
- JavaScript@pdf-lib/upng 1.0.1width 2 · height 1
- JavaScriptcanvas 3.2.3width 2 · height 1
- JavaScriptfast-png 8.0.0width 2 · height 1
- JavaScriptimage-size 2.1.0width 2 · height 1
- JavaScriptparse-png 2.1.0width 2 · height 1
- JavaScriptpng-js 2.0.0width 2 · height 1
- JavaScriptpngjs 7.0.0width 2 · height 1
- JavaScriptprobe-image-size 7.5.0width 2 · height 1
- JavaScriptsharp 0.35.5width 2 · height 1
- Pythonimageio 2.38.1width 2 · height 1
- Pythonimagesize 2.0.1width 2 · height 1
- Pythonmatplotlib 3.11.2width 2 · height 1
- Pythonopencv-python-headless 5.0.0.93width 2 · height 1
- PythonPillow 13.0.0width 2 · height 1
- Pythonpycairo 1.29.2width 2 · height 1
- Pythonpygame 2.6.1width 2 · height 1
- Pythonpypng 0.20220715.0width 2 · height 1
- Pythonrasterio 1.5.2width 2 · height 1
- Pythonscikit-image 0.26.0width 2 · height 1
- Pythontkinter.PhotoImage CPython 3.13.1 with bundled Tk 8.6width 2 · height 1
- Rustimage 0.25.10width 2 · height 1
- Rustimg-parts 0.4.2width 2 · height 1
- Rustload_image 3.5.0width 2 · height 1
- Rustlodepng 3.12.2width 2 · height 1
- Rustminipng 1.0.0width 2 · height 1
- Rustpng 0.18.1width 2 · height 1
- Rustpng_pong 0.12.0width 2 · height 1
- Rustzenpng 0.1.4width 2 · height 1
- Rustzune-png 0.5.2width 2 · height 1
- PHPgumlet/php-image-resize 3.0.1width 2 · height 1
- PHPimagine/imagine 1.5.4width 2 · height 1
- PHPintervention/image 4.4.1width 2 · height 1
- PHPintervention/image-laravel 4.1.1width 2 · height 1
- PHPleague/glide 4.1.0width 2 · height 1
- PHPmarc1706/fast-image-size v1.3.1width 2 · height 1
- PHPnette/utils v4.1.6width 2 · height 1
- PHPphp-getimagesize 8.5.12width 2 · height 1
- PHPphp-getimagesizefromstring 8.5.12width 2 · height 1
- PHPphp-imagecreatefrompng 8.5.12width 2 · height 1
- PHPphp-imagecreatefromstring 8.5.12width 2 · height 1
- PHProsell-dk/webp-convert 2.9.5width 2 · height 1
- PHPspatie/image 3.9.10width 2 · height 1
- PHPyiisoft/yii2-imagine 2.3.1width 2 · height 1
0x0010formats
14 formats, from pixels to protocols.
Every format has small, valid control files that each parser must read correctly before it enters the catalog. These are the first eight bytes of each one. Pick a format to see the parsers that read it.
0x0020use it
Three ways to use it.
Install it from PyPI. It needs Python 3.11 or newer, and Docker to run targets.
$ python3 -m pip install parsarium
Run one file through one parser.
You get a JSON receipt with the input hash, the image ID, the exit status, and the time it took. It is the quickest way to check how a specific library handles a specific input.
$ parsarium show png python pillow $ parsarium run png python pillow my.png
Send the same file to every parser.
Each target writes its own receipt and summary.json collects them. When parsers disagree about the same bytes, that is a good place to start looking.
$ parsarium run-format png --list $ parsarium run-format png \ --file my.png --results runs/png-my-file
Fuzz a parser with AFL++.
All runnable targets have an AFL++ overlay. C, C++, and Rust parsers are instrumented directly. Go, Java, JavaScript, Python, and PHP parsers send their coverage to AFL++ through a small bridge.
$ parsarium afl build png python pillow $ parsarium afl prepare png python pillow runs/pillow $ parsarium afl run runs/pillow --seconds 3600
0x0030engines
Packages and the engines behind them.
Many packages wrap the same decoder. In PHP, most of the PNG packages read images through GD; in JavaScript, most PDF packages build on pdf.js. Parsarium records the parser family of every entry, so you decide what to count.
A study about ecosystems may want every package. A study about parser engines may want one target per family. The bench builds either selection and gives it an ID you can put in a paper.
0x0040manual
A manual for testing parsers.
I am writing these chapters as the notes I wanted when I started: how formats are built, what a harness verifies, how to read coverage feedback, and how to run campaigns across many targets. Each chapter ends with a short quiz and exercises you can run on Parsarium, with hints and solutions.
- 01What a parser doesBytes come in, structure comes out. Why that step matters for testing and security, and how Parsarium organises it.
- 02Formats and grammarsHow formats are built, how a grammar and constraints describe them, and how to read a control file byte by byte.
- 03Controls and receiptsWhat it means for a parser to be verified, what a harness records, and how to read a run receipt.
- 04Choosing targetsCells, adoption ranks, parser families, and standard APIs: how to pick a target set that fits your question.
- 05Coverage feedbackHow AFL++ decides an input is interesting, how Parsarium gets that signal from eight languages, and how to read the numbers.
- 06Differential testingSend the same bytes to many parsers and compare what they say. Where they disagree is where to look.
- 07Fuzzing at scaleFrom one AFL++ campaign to hundreds: seeds, budgets, parallel runs, repetitions, and keeping track of it all.
- 08Triage and disclosureFrom a file in crashes/ to a report a maintainer can act on: reproduce, reduce, deduplicate, assess, and disclose.
- 09Designing an experimentA checklist for studies that use many parsers: question, targets, budgets, repetitions, metrics, and what to publish.
- AGlossaryThe terms used in this manual, in one place.
0x0050about
Why it exists.
I'm José Antonio Zamudio Amaya, a PhD researcher at CISPA working on fuzzing and test generation for programs with complex inputs. Every study I ran started by picking parsers, installing them, and writing adapters. Parsarium is that work done once, carefully, and shared, so experiments can start from the same targets and their results can be compared.