Weird behavior when using \babelfont with two styles in LuaTeX

[NudPiyapon]

Note: This issue stems from stackexchange forum: https://tex.stackexchange.com/questions/637687/weird-behavior-when-using-babelfont-with-two-styles-in-luatex

Consider these MWEs.

\documentclass{article}

\usepackage[english]{babel}
\babelprovide[onchar=ids fonts]{english}
\babelprovide[main,import]{thai}
\usepackage{fontspec}

% \babelfont[thai]{rm}[
%     Script=Thai,
%     UprightFont = *.otf,
%     BoldFont = *-Bold.otf,
%     ItalicFont = *-Italic.otf,
%     BoldItalicFont = *-BoldItalic.otf]
%     {Laksaman}
\babelfont[thai]{rm}{FreeSerif}
\babelfont[thai]{sf}[
    Script=Thai,
    UprightFont = *.otf,
    BoldFont = *-Bold.otf,
    ItalicFont = *-Italic.otf,
    BoldItalicFont = *-BoldItalic.otf]
    {Laksaman}

\begin{document}
    ข้อความทดสอบ Test text.
    \textsf{ข้อความทดสอบ Test text.}
\end{document}

The following code produces expected result. (Correctly apply serif and sans-serif style)

However, when I changed the serif font of Thai to the same as sans-serif one (Laksaman), the product of this code is formatted wrong. (The last test text's style is serif, not sans-serif)

\documentclass{article}

\usepackage[english]{babel}
\babelprovide[onchar=ids fonts]{english}
\babelprovide[main,import]{thai}
\usepackage{fontspec}

\babelfont[thai]{rm}[
    Script=Thai,
    UprightFont = *.otf,
    BoldFont = *-Bold.otf,
    ItalicFont = *-Italic.otf,
    BoldItalicFont = *-BoldItalic.otf]
    {Laksaman}
% \babelfont[thai]{rm}{FreeSerif}
\babelfont[thai]{sf}[
    Script=Thai,
    UprightFont = *.otf,
    BoldFont = *-Bold.otf,
    ItalicFont = *-Italic.otf,
    BoldItalicFont = *-BoldItalic.otf]
    {Laksaman}

\begin{document}
    ข้อความทดสอบ Test text.
    \textsf{ข้อความทดสอบ Test text.}
\end{document}

So, could anyone explains to me what happened here? Thanks in advances!

[jbezos]

I’ve posted a workaround in stackexchange, but I’ll investigate a general solution to this issue once TeXLive 2022 is released (I'm taking a short break).

[jbezos]

While investigating how to deal with these cases and how to improve the current algorithm, I’ve discovered a very dirty trick: add a dummy RawFeature (e.g. RawFeature=+xxxx) to the second font. Seems to work.

[jbezos]

The workaround based on the RawFeature doesn’t work anymore, and what’s worse, the working example (with the expected result) doesn’t work with TeXLive 2023... 🙁 (it worked with 2022).