Limited compilation success rate for arXiv figures

[lucasgautheron]

First of all: thank you so much for making this code available, this is extremely useful.

I have been running the code on arXiv sources and only about 30-35% of the figures are successfully compiled, using a full texlive installation (Ubuntu 20.04).

I was wondering:

• is this is close to the success rate achieved in the making of the full DaTikZ dataset?
• what do you think are the most recurring causes of failure?

[potamides]

Hi, thanks for your interest! 30-35% sounds about right. However, this is only if you include TikZ pictures from the arXiv split. For other splits (e.g., TeX.SE), the rate should be much higher. This is because for arXiv, we have to extract TikZ pictures with their minimal environment to ensure successful compilation from the whole LaTeX document. The tooling available for programmatically processing LaTeX is extremely limited, so we had to implement our own custom rules to extract these minimal environments. These rules are brittle, as LaTeX documents can be very complex and diverse, which is why there are these high failure rates. I am sure these rules could be improved if you try to identify some recurring failure patterns, though.