TeX is a widely-used typesetting system adopted by most publishers and professional societies. While TeX is responsible for generating a significant number of documents, irregularities in the TeX ecosystem may produce inconsistent documents. These inconsistencies may occur across different TeX engines or different versions of TeX distributions, resulting in failures to adhere to formatting specifications, or the same document rendering differently for different authors. In this work, we investigate and quantify the robustness of the TeX ecosystem through a large-scale study of 432 documents. We developed an automated pipeline to evaluate the cross-engine and cross-version compatibility of the TeX ecosystem. We found significant inconsistencies in the outputs of different TeX engines: only 0.2% of documents compiled to identical output with XeTeX and PDFTeX due to a lack of cross-engine support in popular LaTeX packages and classes used in academic conferences. A smaller—yet significant—extent of inconsistencies was found across different TeX Live distributions, with only 42.1% of documents producing the same output from 2020 to 2023. Our automated pipeline additionally reduces the human effort in bug-finding: from a sample of 10 unique root causes of inconsistencies, we identified two new bugs in LaTeX packages and five existing bugs that were fixed independently of this study. We also observed potentially unintended inconsistencies across different TeX Live distributions beyond the updates listed in changelogs. We expect that this study will help authors of TeX documents to avoid unexpected outcomes by understanding how they may be affected by the often undocumented subtleties of the TeX ecosystem, while benefiting developers by demonstrating how different implementations result in unintended inconsistencies.

@inproceedings{paperbot8465,
  title = {Inconsistencies in TeX-Produced Documents},
  author = {Jovyn Tan and Manuel Rigger},
  booktitle = {ISSTA},
  year = {2024},
  doi = {10.1145/3650212.3680370}
}