Code caching produces incorrect output with Unicode characters

[mdbrnowski]

It seems that the caching mechanism treats different snippets containing Unicode subscripts (₀, ₁, ₂, etc.) as identical, leading to incorrect reuse of cached results.

Here is a MWE:

\documentclass{article}
\usepackage{fontspec}
\setmonofont{JetBrainsMono} % or Ubuntu, Lato, etc.
\usepackage{minted}

\begin{document}
    \mintinline{julia}{x₀}
    \mintinline{julia}{y₁}
    \mintinline{julia}{z₂}
    \mintinline{julia}{x₃}
    \mintinline{julia}{y₄}
    \mintinline{julia}{z₅}
\end{document}

Expected PDF output:

x₀ y₁ z₂ x₃ y₄ z₅

Actual PDF output:

x₀ y₁ z₂ x₀ y₁ z₂

If I disable caching or use minted2, the issue disappears. Tested with LuaLaTeX and minted 3.7.0.

[gpoore]

Try adding this right after loading minted:

\makeatletter
\let\pdf@mdfivesum\pdf@mdfivesumnative
\makeatother

minted is using \pdf@mdfivesum from pdftexcmds for MD5 hashing across different TeX engines, but apparently under LuaTeX that ignores all code points consisting of more than one byte. \pdf@mdfivesumnative shows up in a different part of the pdftexcmds documentation from \pdf@mdfivesum, and the \pdf@mdfivesum documentation didn't mention any byte limitations.

I'll fix this in the next release, which should be out sometime in the next few days.

[mdbrnowski]

Interesting, thanks!

[davidcarlisle]

If someone could open an issue at https://github.com/ho-tex/pdftexcmds that sounds like a bug